Home

Geospatial Canopy Cover Assessment Workshop

1. ___ Historical aerial photos classified using OBIA methods Landscape metrics were used to quantify spatial and temporal changes in two classes from a LULC classification 1 forest canopy cover and 2 impervious surface areas are presented in the bar graph Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 12 1 2 2 Accuracy Assessment Performing an accuracy assessment on the classification results is mandatory in order to evaluate the truthfulness of the classified thematic map s representation of real world attributes Congalton and Green 1998 An appropriately designed accuracy assessment compares not only the classification map but also the ground reference test information and to what degree the classifications represented thematically are actually correct to the corresponding ground reference locations Jensen 1996 However when land use land cover classification procedures are performed on historical aerial photography ground reference test information current to the date of image capture cannot be obtained to validate the classification for historic dates If land use land cover classifications had previously been generated for the location those could serve as possible reference data for the accuracy assessment However lacking previous land use land cover classifications for comparison the accuracy assessment can also be performed by using a skille
2. uw F GEOSPATIAL CANOPY COVER ASSESSMENT WORKSHOP pms Software used SPRING 5 1 7 and FRAGSTATS 3 3 Presented by Dr L Monika Moskal amp Dr Diane Styers Tiy 40 Sian Bagh 3 Nd tine ieg Computer Exercise Materials Developed by Justin L Kirsch Remote Sensing and Geospatial Analysis Laboratory Precision Forestry Cooperative at the School of Forest Resources University of Washington Page 2 Table of Contents ILBACKOROUN D crni stictinrat toad otasslabmenat a a a 3 Li Purpose SMG ODJECUVES arichida n a a a a a a NE 5 1 2 Remote Sensing C NCeDIS siyeni naan TT E EA 6 1 2 1 Object Based Image Analysis OBIA amp GEOBIA cccccccccssseeceeeceeseecceeeeeeeeceeeeeeeeeceeeeueneeeeeeas 10 t22 Accuracy ASSE SMENT era r E E a N a aE 12 U2 3 kand cape IVICULICS aioi a E EEEN N a esdideees 14 dP ae DEE Oa T N AE E Pe ere ne 16 ZANAN SU aa A aE A a E a N 18 ZV AVS taline SPRING snae na n Aa an N a a aa 19 22 IMStallie FRAGS TATS srac T R O a NRA 20 2 3 OBIA Project Setup and Segmentation in SPRING ssssesssusssesesensserssrerssrrsseessrersseresersseresseessrersreresees 21 Pacem We Gl sre a ay ears aD Le 81 gt ol eee eee a a a a eee eee 21 Deus Cl CACHE ile ROS CU cccaweaa nae ecctreusce E A a a A a a 22 2 So IMPORTING Man Ey a a anise neshaccaia EAE AE NAA 24 2 DA Sep MeN tat OM E a Oa E T 28 2 OBIA Classification In SPRING enea a N a cas 31 2 5 Post Classificatio
3. Area per Class BE eriou Me Fores ES Background Matris steal dl eo Pa impervious cover IS increasing over time and that increase is a homogenous cover with less and less mixing of other classes such as forest canopy One other finding somewhat surprising is that the number of forest canopy patches has been reduced by over half for Landscape A and over 2 3 for Landscape B moreover the forest cover also became more continuous over that time period Such results are sometimes startling and cannot be taken at face value One of the keys to interpreting results is to understand the data that was used to produce them Ideally we would want to use a remotely sensed image of the same spatial and spectral resolution that was collected over our area of interest at what is often referred to as the anniversary date or a temporal phenological window that overlaps for each year of collection The spectral resolution of the imagery we use for comparision is important because vegetation is easier to distinguish on true color and near infrared imagery than black and white imagery Finally the spatial resolution plays a role because sample pixel area allows us to see a greater amount of detail such as individual tree canopies 1379 1990 2004 204 Landscape A Landscape B Unfortunately it is rare that we can have all of these factors optimized as remote sensing technology changes over time For example NAIP ae
4. if possible image parameters should remain constant i e the same time of year time of day spectral bands sensor sensor look angle spatial resolution and so forth Imagery with obvious features that can be confused with change such as clouds or extreme soil moisture conditions should be avoided Images to be compared must be carefully geo rectified and registered to the same map projection to avoid mistaking misregistration for change An assumption for all change detection techniques is that the areal extent of the changes to be detected is larger than the spatial resolution of the imagery To detect rapid changes to the environment such as residential development a short time span sequence of images is required whereas to detect trends and to forecast a longer time span and larger number of images is required Hame et al 1998 Visual Change Detection For qualitative assessment and display purposes changes can be visualized by loading corresponding bands from multi date images into different computer display channels No change positive change and negative change appear as different colors in the image useful for a general examination of land cover change However the composition and quantity of change cannot be identified or calculated from the visual change detection method Change Direct Multi Date Classification or Post classification Analysis The direct multi date classification approach involves independent classifica
5. Area AREA_ D Radius of Gyration GYRATE_ D Mean MN Mean AM r r MD Area Weighted Median Range St andard Coefficient of RA Deviation SD Variation CV a a a a r r Select Al Select Al NOTE Radius of Gyration Area Weighted Mean GYRATE_AM is equivalent to Correlation Length CL as used in the literature a Ss Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 43 Step N Click the Execute iconon the toolbar to run the process that calculates Statistics mT O OOOO O O O O SLN aali e s Ga d s L Event log Getting user options User options changes cancelled Getting user options User options changes cancelled Getting user options User options read in Checking Checking finished Getting user options User options changes cancelled Step O After the process is complete you should see the message below Click OK and then click the Browse a Results icon on the FRAGSTATS toolbar to bring up Results window Step P The results window will be blank click the Class tab in the bottom left of the window to display the Class metrics statistics W Save ADL file Pate Chass J and Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 44 Step Q Browse the results by clicking on the Class and Land tabs at the
6. Close Help Step D In the middle of the Projections window you will see that the Hemisphere radio buttons are not active and are locked on South this is a problem because we will be using North for our projection To activate the Hemisphere radio buttons click on the Polar Stereographic projection in the Systems box you will see them activate click the North radio button then change your projection back to UTM NAD83 US Step E When you are finished click Apply Your project will now use this projection for all imported images Step F With the Projects window still open in the Bounding Box region of the Projects window click the Planes radio button and make sure both N Hemisphere radio buttons are selected Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 23 Step G Manually enter the following coordinates as is shown circled below X1 537888 X2 539728 Y1 5235270 Y2 5237361 Click Create and then click Load note you will not see the image below right after you click load Step G Visualization of Step G Name Tacoma Projection UTM Reference Projection Projection X1 537888 Yi 5235270 Delete You have just created a project with spatial projection and a bounding box for your project which creates a maximum extent for the project area If an imported image is larger than this box then
7. Page 11 classifications look at signatures to aid in categorizing pixels However the per pixel classifications only have spectral information signatures Object oriented techniques also produce signatures but for a much larger set of information including the above mentioned characteristics such as size orientation texture etc Along with a more robust set of signatures to aid in more accurate pixel categorization object oriented approaches also look at the relationship of signatures among neighboring objects Cothren and Gorham 2005 Object oriented feature extraction using Feature Analyst provides a multi dimensional approach to image classification making more efficient use of the depth of information contained in an image Most importantly the incorporation of spatial relationships and the scaling of objects incorporate the fundamentals of geography and landscape ecology into image classification techniques Applying geographical concepts to image classification techniques most often provides superior results more reminiscent of the natural landscape compared to traditional per pixel methods Blaschke and Strobl 2001 Fundamentally classifications generated under the consideration of spatial relationships are consistent with using landscape metrics which also use the premise of spatial relationships and scale to describe the resulting classifications impervous Forest Background Matrix Aron in Square Mice 4
8. Remote Sensing 20 12 2331 2346 Civco D 1989 Topographic normalization of Landsat Thematic Mapper digital imagery Photogrammetric Engineering and Remote Sensing 55 9 1303 1309 Davis F W and D M Stoms 1996 A spatial analytic hierarchy for Gap Analysis Pp 15 24 in Gap Analysis A Landscape Approach to Biodiversity Planning J M Scott T H Tear and F W Davis Editors American Society for Photogrammetry and Remote Sensing Bethesda Maryland Dymond J 1992 Nonparametric modeling of radiance in hill country for digital classification of aerial photographs Remote Sensing and Environment 39 95 102 Eastman J R 1992 Time series map analysis using standardized principal components In Proceedings ASPRS ACSM RT 92 Convention on Monitoring and Mapping Global Change Volume 1 195 204 American Society for Photogrammetry and Remote Sensing Bethesda MD Elkie P R Rempel and A Carr 1999 Patch Analyst User s Manual Ontario Ministry of Natural Resources Northwest Science and Technology Thunder Bay Ontario TM 002 16 pp Appendices Elvidge C D D Yuan R D Weerackoon and R S Lunetta 1995 Relative radiometric normalization of Landsat Multispectral Scanner MSS data using an automatic controlled regression Photogrammetric Engineering and Remote Sensing 10 1255 1260 Franklin J T L Logan C E Woodcock and A H Strahler 1986 Coniferous forest classification and inventory using Landsat TM a
9. Vegetation Index NDVI is a measure of the amount and vigor of vegetation on the land surface and NDVI spatial composite images are developed to more easily distinguish green vegetation from bare soils In general NDVI values range from 1 0 to 1 0 NDVI NIR RED NIR RED Patch the basic elements of landscapes Also commonly referred to as ecotope biotope landscape component landscape element landscape unit landscape cell geotope habitat or site Scale The pattern detected in any ecological mosaic is a function of scale and the ecological concept of Spatial scale encompasses both extent and grain Segmentation the process of partitioning a digital image into multiple segments by delineating pixels in an image and grouping them together Structure the spatial relationships among the distinctive ecosystems or elements present more specifically the distribution of energy materials and species in relation to 2 the sizes shapes numbers kinds and configurations of the ecosystems Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 56 APPENDIX E Accuracy Assessment Statistics The Kappa or KHAT statistic should also be calculated as it provides a normalized overall accuracy taking into consideration the correct omitted and committed values from the whole table giving a better idea of how the classification algorithm performed as a whole and allowi
10. You do not need to click the Directory button the classification will be stored to the same path as defined in earlier steps Step B Click the Create button to open the Context Creation window This is where you will create a file for the classification Step C In the Context Creation window type in classification1 in the Name field and click the Regions radio button Select all bands by clicking on them and click segmentation_30_100 in the Segmented Images box then click Apply A file for classification1 will be created Step B Step C Classification Context Creation Z Justin Kirsch Workshop seg_class_example sec Name classificationl Contexts Analysis Type Pixel Regions Bands CAT Image Tacorna_1 CAT Image Tacorna_2 CAT Image Tacoma_3 CAT Image Tacorna_4 Segmented Images Step D With the Classification window still open click on classificationl in the Contexts box this will make classifiaction1 active Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 32 Step E With classification1 active click the Extraction of attributes of the regions button This makes the classification boxes in the bottom of the Classification window active Step F After the Extraction of attributes of the regions process is complete click the Training button A window will appear telling you to acquire sample s click OK
11. all areas outside of this extent will be clipped out The outlined region on the image at left visualizes the bounding box step the rectangle represents the bounding box Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 24 2 3 3 Importing Imagery Once a database and a project have been created the next step is to add imagery to a project SPRING works best if imagery is imported using the same spatial projection as the project It is a good idea to know the projection of your imagery before defining the projection of your project For reference the imagery we are using today is projected in UTM NAD 83 Zone 10 North which is why this projection was used when defining the project projection Task 5 Import Imagery Step A With SPRING open click File Import Import Vectorial and Matrical data Step B Click the File box and navigate to the folder where the Tacoma tif file is stored Step C In the File window change the file type from ASCII SPRING spr to TIFF GEOTIFF tif tiff by clicking on the File of type dropdown list The Tacoma tif file will appear click on the Tacoma tif file and then click Open Step B Step C Import Data Conversion Output Look in jo Z Pilot_workshop E yy My Computer Name h Size Type Date Modified B Tacoma tif 158 3 MB tif File 10 13 2010 4 51 26 P kirschj gt 2 3 dJ pilot workshop Fil der 2 13 2011 2
12. areas that have been misclassified as water you can tell if they are water or not by looking at the true color image for reference You can compare your classification with the true color image by toggling classification on and off by clicking the Classified box in the control panel Step Q With the trees class still active click on any region you think are trees but have been classified as something else If necessary repeat these steps for any class that appears to need refining When you are finished click Save and then click Close Step R Now open the Classification window and name the new classification classification2 Choose the same settings as you did in Step J and click the Classify button You will be able to see that your classification will be improved but not perfect The region of canopy shadows that were being grouped into the water class are now being placed into the tree class The classification of water has also been improved in classificationl we can see that some water was being grouped into the roads and buildings classes Also we could have added a class for dry vegetation for this image it would have made sense because it looks as though the image was taken in summer as you can see by the mixture of brown and green grass For this classification we have essentially grouped dry and green grass into one class Needless to say this classification is not perfect There are inevitably going to be areas that are misclassified
13. associated with this method include labeling the change classes and identifying how classes have changed One procedure is simply to combine image bands from multiple dates produce an unsupervised classification and label the clusters as changed or unchanged Alternatively Principal Components Analysis PCA or Tasseled Cap Transformation can be used on the combined data set which produces a new de correlated data set Eastman 1992 One or more of the resulting data channels correspond to the differences change between the original two temporal images Image Algebra e Image subtraction The digital values from one image date are subtracted from the values of another date Areas with change will have large differences in value while those with little change will have small differences e Image ratio A ratio of the band of one image date to the band of another image date is created For areas with no change the value of the ratio tends toward one Areas of change will deviate to higher or lower ratio values How much deviation is required for change is based on selected threshold values Change Vector Analysis The magnitude and direction of spectral change between two image dates can be obtained by plotting for single pixels the values for two spectral variables for the two dates The magnitude of the change is the Euclidean distance between the two points The direction of the vector relates to whether the change is positive or negative
14. bottom left of the window 1 C Users kirs 1 C Users kirs 1 C Users skirs unclassified 1 C Users kirs 24 1 C Users skirs ate 1 C Users skirs trees 1 CNUsers skirs IS 71 4329 1 C Users kirs roads 71 3821 1 C Users skirs Orass Tiari W Save ADJ file Save run 4s Clear all Clear this Close Patch Class Land We can see from the PLAND percentage of land column in the Results window that the percentage of land taken up by canopy is about 26 your results may vary and are determined by your segmentation parameters and your training data In the Type column we can see our list of classes note that water does not appear because we turned it off in our Notepad text file You may also see that there are some classes named with numbers these are unclassified pixels that were not grouped into the unclassified class Do not worry about these too much there aren t usually very many of them to affect your Statistical output significantly Congratulations You have successfully calculated the percentage of canopy coverage for a 3 by 4 mile area in North Tacoma This concludes the workshop on Conducting Canopy Cover Assessments using the freely available programs SPRING and FRAGSTATS Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Lanscape metrics are often reported in tabul
15. or not classified at all but we have reduced this error and it is recommended that one continues to refine a classification until they are reasonably happy with the output For the purposes of this workshop we will stop refining the classification and use the output from classification2 for the next section Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Classification 1 Som ATER L Sie a a ier ra Co ae E Fak L PAZ Aerie RECEP S eet td Ee Zk fe tis ee Kak A enee oe T Sete toes RIM 2 foot sels CI eR Page 35 Classification 2 KEN f a 4 EM rs 3 2 w Ee DANCETI tt cates Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 36 2 5 Post Classification Analysis in SPRING Post classification is a process that refines your classification and also places unclassified pixels into classes It is highly recommended that one runs a post classification especially if there are a large number of unclassified pixels in the classification Like refining your initial classification by acquiring more training data post classification can significantly improve the accuracy of your classification by refining your classes Task 8 Perform a Post Classification Step A From the toolbar click Image Classification and click on Post classification in the Classification w
16. surface varies with elevation To compensate for distortions in the flat surface of an image caused by being captured from an uneven surface of the earth an orthogonal rectification is performed Orthorectification is a mathematical manipulation in the locations of pixels of an image to the actual locations and elevations of certain known features on the ground Mathematical models for orthorectification include nearest neighbor re sampling bilinear interpolation and cubic convolution Information regarding surface elevation is obtained using a DEM Furthermore imagery is often acquired as independent photos covering limited area mosaicking of imagery is often down at the same processing step as orthorectification One key step that needs skilled image analysis is matching the color or shading of imagery during mosaicking The process is often referred to as histogram matching and it aims to eliminate tone differences between images Cut lines are often employed to stitch data sets along natural seams such as roads water bodies or other linear features Land use land cover LULC Land cover describes natural and built objects covering the land surface while land use documents Level Level II Level Ill 1 Urban or Built up Land 11 Residential 111 Single family Units 12 Commercial and Services 112 Multi family Units 13 Industrial 113 Group Quarters 14 Transportation Communications and Utilities 114 Residential Hotels 15 Industrial and Com
17. the information we obtained in the Layer Properties box in ArcMap In the Class properties file field load the classes txt file we created in Notepad In output statistics click the Class and Landscape metrics boxes After you are finished click OK and move onto the next step m Input File Type 8 bit file name cyu sers kirschj Desktop workshop classification2_pos_7 _ tif Landscape Output File C Users kirschj Desktop workshop T acoma_canopy_metrics x Batch Fie Base Name Only Automatically save results m Input Data Type Grid Attributes Analysis Type Ay m Cellsize in meters 1 000 A ASCII Moving window 8 Bit Binary Background Value 399 7 16 Bit Binary Ente Scalia pje 32 Bit Binary Number of Rows y C ERDAS Radius meters Number of Columns x f 840 C IDRISI fo o00 Unique Patch ID s Class properties file C Users kirschi Desktop workshop classes txt x O a Do Not Output ID Image Patch Neighbors Output Statistics Create and Output ID Image 7 Patch Metrics r C InputID Image 4 Col Aule WV Class Metrics ID Fil 1 8 Cell Rule o oK _Dfie V Landscape Metrics Cancel Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Step K Open the class metrics window by selecting the Class Metrics icon on FRAGSTATS toolbar Page 42 the Step L Check the boxes for Total Area CA TA Per
18. the sun atmospheric conditions or different sensors Lillesand and Kiefer 1994 For comparing reflectance values at different times or for quantitative applications of digital imagery these corrections may be necessary Sun elevation correction The season of the year determines the position of the sun relative to the earth The sun elevation correction normalizes the image data taken at different dates to values assuming the sun was at the zenith position Earth sun distance correction The season also governs the distance between the sun and the earth which influences the amount of solar irradiance reaching the earth s surface The earth sun distance correction calculates the amount of solar irradiance for a given scene at the mean earth sun distance Atmospheric correction The amount and kinds of particles in the atmosphere affect the amount of solar irradiance reaching objects on the ground and the amount of light scattered haze that is detected by the sensor One method to compensate for haze in the atmosphere is to observe the radiance value of an area that is theoretically zero deep clear lake for example Any radiance detected by a sensor in these areas is assumed to be due to the scatter from haze for each image channel the dark object value is calculated and subtracted Conversion to absolute radiance Absolute radiance is essential for modeling the reflectance properties of physical objects or biophysical processe
19. 16 42 PN Esta ferramenta possibilita a importa o dos seguintes tipos de arquivos ASCII SPR ARCINFO SHAPE DXF DWG JPEG IDRISI SURFER TIFF GEOTIFF JPEG2000 KML Obs Para um mesmo tipo de arquivo e entidade poss vel selecionar e importar varios arquivos simultaneamente Step D After loading the Tacoma tif file click the Bounding Box button to open the Bounding Box window Step E Click the Project and Planes radio buttons If both Hemisphere S radio buttons are selected change them both to N Step F If there are decimal places after the X1 X2 Y1 and Y2 coordinates then click back and forth between the Active PI and Project and Geographic and Planes radio buttons until they disappear If this step was necessary then be sure to return your radio buttons back to Project and Planes Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 25 Step G When you are finished click Apply Step D Steps E F and G Bounding Box gt S Ser 2 Pilot_workshop Tacoma tif Select below the file Projection Datum Spring will ii Active PI automatically convert the data and adjustit to the project KA g If the project does not exist it will be created with the Cursor No C Yes same information p a x Project UTM Datum gt NADB3 US Pixel Size ji Coordinates Geogr
20. 30 m pixel Landsat data have been available since 1972 The level of spatial detail is too coarse for extracting information at the forest stand or sample plot 10 x 10 m level However the Landsat TM data does give information on the larger landscape context which can have impacts on local forest health Furthermore because of a slightly higher temporal resolution Landsat data can show multiple inter annual observations for a location however this ability is reduced in the Pacific Northwest due to cloud cover For preprocessing imagery and extracting ground information useful in classification another required source of data is the digital elevation model DEM Digital elevation data is similar to imagery except that each pixel contains a height value instead of a gray level intensity Continuous coverage DEM data are available through base mapping programs such as the Shuttle Radar Topography Mission SRTM which offers freely available data nationwide at 30 m per pixel resolution Data pre Processing Aerial and satellite remotely sensed data undergoes an extensive pre processing prior to utilization and although a majority of this work is done by the data providers it s critical to understand some of these concepts as these can have an impact on the accuracy of output products Below we focus mainly on pre processing of aerial data However as more and more high spatial resolution data sets are available from space borne sensors it i
21. After you have outlined your box click on the Draw tool to zoom into the area Here you can see how well your segmentation has delineated features The segmentation seems to make sense features like roads buildings and groups of similarly colored patches of trees are grouped together Step H After you are finished inspecting your segmentation click the Reset orZoomIL amp amp tool to return the image to its full extent and move on to the next section Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Figure 1 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 31 2 4 OBIA Classification in SPRING After completing the segmentation you are now ready to begin the classification steps of the OBIA method A classification uses the delineated pixels from the segmentation to automatically categorize pixels in an image into similar classes You will first have to provide training data for the classification to work more simply you have to tell SPRING what to call what For example if you wanted to create a classification with 2 classes you would tell SPRING to call some pixels trees trees and all other pixels non trees We will cover these steps in detail in the following section Task 7 Context Creation and Classification Step A Go to Image Classification to open the Classification window
22. C Here we will create a text file for our classes Open Notepad it should be in the Accessories folder Step D From the Layer Properties window in ArcMap we know the order of classes Type in Ic what you see at the right with no spaces between File Edit characters your class orders should be the same 55 l 0 unclassified true false This will tell FRAGSTATS how to order classes 1 water false true Notice that the water class is the only one that is 2 trees true false false true This means that we will not be 2 oads true false CA l l 4 buildings true false factoring in water in our calculation of the total 5 grass true false land area This is important if you want to know how much of the land is taken up by canopy and do not want to factor in water as a percentage of the total area Format View Help Step E When you finish entering in the list of classes click File Save and name the file Notepad file classes save the file to the same folder your other data is stored in Step F With ArcMap still open double click the classification2_pos_7_7 layer to bring up the Layer Properties window Click the Source tab Under Raster Information look at the Columns and Rows values Columns 1840 and Rows 2091 Keep note of these values we will need them in the later steps mn A Raster Information Columns and Rows 1840 2091 Number of Bands Cellsize X Y 1 1 Uncompressed Size 3 67
23. For example a negative change might be a loss of vegetation whereas a positive change might represent vegetative re growth Combination of Analyses Traditional classification of a reference image to develop base classes can be combined with information derived from a change image To extract information on the classes that have changed in the second image only the area under the binary mask need then be classified By limiting the second classification to areas of change the classification requirements are not as complicated Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 18 2 ANALYSIS Geospatial Software A wide range of geospatial software is available commercially as freeware and as open source Segmentation is an analysis imbedded in the following image analysis software ENVI ERDAS Imagine and IDRISI OBIA based classifications are possible as extension software to ERDAS and ARCGIS called Feature Analyst or as standalone packages Berkeley Image Segmentation and Definiens eCognition The cost of these packages can be prohibitive with eCognition toppling the cost scale We chose freely available software for this workshop although there are pros and cons to using freeware we believe cost is often the most limiting factor thus we hope that the detailed steps presented in this workbook help to overcome some of the downfalls of the freeware The following two free sof
24. MB Format TIFF Source Type discrete Pixel Type unsigned integer Pixel Depth 8 Bit Data Source Data Type File System Raster Folder G tacoma_wkshp Raster B2 dassification tif Set Data Source OK Cancel Apph Step G Open FRAGSTATS and Click File New on the toolbar to create a new project Step H On the FRAGSTATS toolbar click File Save to save your project Navigate to the folder where your other data is stored name the project Tacoma_canopy_metrics and click Save Step I From the toolbar click Fragstats Set Run Parameters to open the Run parameters window You can also get to the Run Parameters window by clicking the Set 4 Run Parameters icon on the toolbar Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 41 Getting user options _ User options changes cancelled Step J In the Run Parameters window fill out the Run Parameters as shown below In the Input Data Type filed click the 8 Bit Binary radio button Click the 8 bit file name button and load the classification2_pos_ 7_ 7 tif file Click the Output File button and in the File Name field name your output Tacoma_canopy_metrics and click Open In the Grid Attributes box put in 1 for cell size 999 for background value and 2091 for number of Rows and 1840 for number of Columns circled below this is the size of our imagery and is
25. Size Standard Deviation Mean Patch Size fee ee eee Total Edge m Perime Edge Metrics Edge Densit Ari Mean Patch Edge Wer re a Sum of each patches permeter divided by the square root of patch Mean Shape Index area for each class or all patches and adjusted for square standard vide r the iber of patche The mean shape index multiplied by the patch area divided by the total class area Wean Perimeter Area Ratio apna each patches perimeter area ratio divided by the number of Jolie The sum of 2 times the logarithm of patch perimeter m divided by the logarithm of patch area m2 for each patch of the corresponding patch type divided by the number of patches of the same type the raster formula is adjusted to correct for the bias in perimeter Shape Metrics Area Weighted Mean Shape Index Mean Patch Fractal Dimension The mean patch fractal dimension multiplied by the patch area divided by the total class area the raster formula Is adjusted to correct for bias in the perimeter Area Weighted Mean Patch Fractal Dimension Nearest The sum of the patch area divided by the squared nearest edge to Neighbor edge distance between the patch and the focal patch of all patches Metrics Mean Proximity Index of the corresponding patch type whose edges are within a specified distance of the focal patch summed across all patches in the y the total number of patches The sum of the distance to the nearest patch of the sam
26. To perform a classification we first have to tell SPRING what groups of pixels belong to what features For example by looking at the segmentation we can see that like pixels are grouped together for example tree canopies that are next to roads are separated by lines During the training portion of the classification step we will provide names classes for groups of like pixels This is called a supervised classification because the user is determining the number of classes and the pixels that belong in those classes In other words the user tells SPRING what to call what Step G In the Training window type the name water in the Name field and choose a blue color for water by clicking on the Color box then click Create You have just created a class for water water Total Num of Pixels 0 The name water will appear in the Themes box with zero acquired pixels You have just created a class named water which you will place regions of water pixels into by selecting them from the image This is another crucial step in the OBIA method because it determines what pixels are going to be grouped into what classes Now that we have a class for water we will have to acquire training samples for the water class Step H In the Training window make sure the Region radio button is selected Zoom to some water in the image you may have to click A the Zoom IL icon on the toolbar first to display your image in full Step l Click the Cross Curso
27. With your training steps complete click the Classification button in the Classification window In the Type of Classification drop down box Tipo do Classificador choose the Bhattacharya method Step M In the Acceptance Threshold drop down pick an acceptance threshold of 99 9 then click the Sample Analysis button You can choose to save the output if you wish but you do not have to Click Close to close the Sample Analysis window Image Classification oe g E Atributos do Classificador Sample Analysis Acquisition Classif Accuracy 100 00 Tipo do Classificador Bhattacharya r Acceptance Threshold 99 9 Texture File Omission Errors 0 00 Comission Errors 0 00 3 Themes Theme Confusion Matrix water Total Num i water 100 00 trees trees 0 00 Themes Area of Acquisition Q Sample Analysis roads roads 0 00 Imagem de Saida water 100 00 trees 0 005 roads 0 00 CAT Image Name dassification1 Classify Close Help Step N In the Name field name your classification classification1 and click the Classify button to run the classification After the classification process is complete take some time to inspect the output By comparing the true color image with your classification you will be able to see that there are misclassified regions and regions that have not been classified at all We will need to refine our classification by selecting n
28. a nternational Journal of Remote Sensing 18 3211 3243 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 51 Hame T Heiler and J S Miguel Ayanz 1998 An unsupervised change detection and recognition system for forestry International Journal for Remote Sensing 19 6 1079 1099 Jaeger J A G 2000 Landscape division splitting index and effective mesh size new measures of landscape fragmentation Landscape Ecology 15 115 130 Jensen J R 1996 Introductory digital image processing a remote sensing perspective 2 Edition Prentice Hall Inc Upper Saddle River NJ 07458 Leckie D G and M D Gillis 1995 Forest inventory in Canada with emphasis on map production The Forestry Chronicle 71 1 74 88 Leopold A 1933 Game Management Charles Scribners New York Li X and A H Strahler 1985 Geometric optical modeling of a conifer forest canopy IEEE Transactions of Geoscience and Remote Sensing 30 276 292 Lillesand T M and R W Kiefer 1994 Remote sensing and image interpretation 3 Edition John Wiley and Sons Inc McGarigal K and B J Marks 1994 Fragstats Spatial Pattern Analysis Program for Quantifying Landscape Structure Version 2 0 Forest Science Department Oregon State University Corvallis Oregon Mas J F 1999 Monitoring land cover changes a comparison of change detection techniques International Journal of Remote Se
29. aphic Planes Xi 537888 X2 539728 Hemisphere N G 5 N MS Step H With the Import window still open click the Output tab Click the Category button and change the category to CAT_Image and click Apply Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 26 Step I In the IL info layer field change the IL name to Tacoma and click Apply If the message below appears after you click Apply click OK Click back to the Data tab and click Apply Once this process finishes click Close Ds Tacoma tif SSelect below the file Projection Datum Spring will automatically convert the data and adjust it to the project If the project does not exist it will be created with the same information Data exist are of the area of the project bba Bounding Box Pixel Size Entity Image v Dummy Value After you are finished with the above steps your imagery will be loaded but it will not be displayed We will cover how to load imagery in the following steps but first we will briefly cover the contents of the SPRING control panel window The control panel is the screen on the left with the Tacoma_1 through Tacoma_4 bands This is where the information of your imagery is displayed Our imagery has four bands one for red green blue and infrared These bands each contain information
30. ar and graph formats These reports are also supported by visualizations For exmaple below are GEOBIA classificaiton results for two growing communities between the years of 1997 to 2004 The red color in the calssification represents impervious areas the green forest canopy and the other color summarizes the background matrices all other classes in the LULC classification We can observe that in both landscapes the total area of the impervious class increases over time however it s more difficult to distinguish changes in the forest class area More improtantly we can observe for Landscape A the impervious areas in the 1990 image help to establish transportation corridors that serve as vectors for impervious area expansion near forested areas However the pattern is slightly different for Landscape B the impervious areas increase but the increase is centralized in the lower left hand corner Again it s difficult to assess the change in area of the forest class for Landscape B Although the visual products of GEOBIA LULC are powerful detailed and informative these do not provide us with numbers such as percent change in forest class or area increase in impervious class The visual products of GEOBIA LULC classification often need to be supported with tabular and numerical reauslts For Page 45 3 REPORTING Landscape A Landscape B example FRAGSTATS allowed us to calculate many landscape metrics for these t
31. ata International Journal of Remote Sensing 2 15 41 Turner S J R V O Neill W Conley M R Conley and H C Humphries 1991 Pattern and scale Statistics for landscape ecology Pp 17 49 in M G Turner and R H Gardner Editors Quantitative methods on landscape ecology Springer Verlag New York NY Woodcock C E J B Collins S Gopal V D Jakabhazy X Li S Macomber S Ryherd V J Harward J Levitan Y Wu and R Warbington 1994 Mapping forest vegetation using Landsat TM imagery and a canopy reflectance model Remote Sensing and Environment 50 240 254 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 52 APPENDIX A Data Sources You will find current and extensive list of data mostly freely available on the RSGAL Geoportal site created specifically for Urban Forest Assessment http depts washington edu iufa Integrated Urban Forest Assessment Geoportal This site is hosted by Dr L M Moskal s Remote Sensing amp Geospatial Analysis Laboratory at the University of Washington UAPP ER SITY OF ff WASHINGTON This project is driven by a partnership between Dr Moskal s RSGAL and the Green Cities Research Alliance UW IUFA Discussion Board NEWS UW IUFA datasets are available on the IUFA ArcGIs com site Partners can obtain login information from iufa uw edu esri Geoportal Server Brta release of 2002 and 2009 Sea
32. atable on future dates other locations or other datasets e g canopy cover impervious surfaces agriculture land use land cover classifications 3 Skills in quantifying and assessing the accuracy of the technique and how results can be used to develop a field sampling regime for more in depth canopy assessment 4 Skills in producing and understanding landscape metrics for multi temporal change analysis 5 Understanding of how the products can be used as the 1st step in land use land cover mapping 6 Asummary report of their findings This workshop module description of method training exercise and report template will then be freely Verification and available online to the general public This workshop Re evaluation fills a need for city and county personnel in the state of Washington who may not otherwise be able to afford a training in the use of these new technologies in order N i J to protect natural resources in their communities After the workshop Dr Moskal and Dr Styers will work on a revision to the protocol based on the feedback provided by the workshop participants The report will oo Classification Change be on the project website hosted by RSGAL along with and Mapping Detection revised workshop manual and module at http depts washington edu rsgalwrk canopy ea To the right is a conceptual flow diagram of the process Honos etrics described in this workshop Geospatial Canopy Co
33. but only one project can be viewed at a time The following steps will cover how to create a project and will also cover how to define a spatial projection for your project area While it is possible to create a project without spatial projection this is not recommended One would still have to define a projection when exporting any file created within that project To avoid doing this step later we will define a spatial projection when we create the project Task 4 Create a project and define its projection Step A With your database loaded go to File Project Project to open the Project window In the Name field name your project Tacoma Step B Click on the Projection box in the Project window to open the Projections window Step C Click on UTM in the Systems box and Datum gt NAD83 US in the Earth Models box Type 10 into the Zone field and click away in the Long box you will see coordinates for Longitude appear Step B step C P Projects sz lal Projections Projects TE NO PROJECTION a Datum gt ITRF WGS84 UTM Datum gt SIRGAS2000 MERCATOR Datum gt SAD69 GAUSS TM Datum gt CorregoAlegre LAMBERT MILLION Datum gt AstroChua o m LAMBERT _ Datum gt SICAD r so Datum gt NAD83 US Bounding Box Coordinates Geographic Planes Long o 1230 0 Long1 Long2 Lati Lat2 z Standard Parallel First Lat C Create Load unload Change _ Seg Lat
34. centage of Landscape PLAND and Number of Patches NP then click OK This tells FRAGSTATS to output only these class metric statistics V Total Area CA TA V Percentage of Landscape PLAND IV Number of Patches NP Patch Density PD Largest Patch Index LPI Landscape Shape Index LS Normalized LSI NLS Total Edge TE Edge Density ED Background Boundary f Do not count any as edge Cou roll Cr spec nt all as edge m ff city the proportion to treat as edge Patch Area AREA D Radius of Gyration GYRATE_ I Mean Area Weighted MN Mean AM E E MD E Median Range Standard Coefficient of RA Deviation SD Variation CV L m r NOTE Radius of Gyration Area Weighted Mean GYRATE_AM is equivalent to Correlation Length CL as used in the literature Step M Open the landscape metrics dialog box by clicking the Landscape Metrics on the FRAGSTATS toolbar Check the box for Total Area CA TA to calculate the area of the image then click OK V Total Area CA TA Percentage of Landscape PLAND Number of Patches NP Patch Density PD Largest Patch Index LPI p L f fh icon total Total Edge TE Edge Density ED Background Boundary f Do not count any as Count all as edge D edge Specify th y the proportion to tre j fo at as edge Patch
35. cing rapid population growth due to economic expansion in leading industries such as aerospace biotechnology information technology and international trade City of Seattle 2010 To meet the infrastructure needs of this growing population forested and agricultural lands are being converted to human modified urban uses at staggering rates For example increases in military and civilian personnel and their families due to growth at Joint Base Lewis McChord has resulted in a 201 population increase in the nearby City of DuPont since 2000 U S Census Bureau 2010 A visual scan of historical and current aerial photographs below quickly reveals a substantial amount of land conversion from natural cover to urban development west of DuPont to the Nisqually Delta Although historical aerial photography has been available for over 60 years due to improvements in image processing tools it is just now rapidly evolving as a management tool Morgan et al 2010 Land cover and land use classes such as forest canopy and impervious areas can be automatically classified and extracted from high resolution aerial and satellite imagery using new Geographic Object Based Image Analysis GEOBIA OBIA techniques Hey and Castilia 2008 Quantifying changes in cover using landscape pattern metrics such as forest patch size and connectivity helps us identify and visualize areas that are changing quickly Dunbar and Moskal 2004 This information can assist decision ma
36. ction of new SPRING offers a 64 bit version the SPRING sigorthms and methodologies II 2 b Scientific Citation of SPRING software perfo rms op tima y In a 3 It hed citation oh sar hee in pears reports piense use a reference to the paper S e a jellin zamara S a Freitas UM Garrido J Cor mere amp Graphics z atmosphere Mata See Also News in the Spring Version 5 1 Mirror Sites for SPRING download Step C Choose install complete to install the SPRING IMPIMA SCARTA package in the English column and then click run SPRING Step C will now be added to your list of programs Home Download Support Manuals Data Publications News Links Contact Step B Download Downloads Support Manuals Data Publications News Links ye RIN Spring English gt Download Download WELCOME Here you can download SPRING 5 1 7 32Bits for WINDOWS SPRING WIN environment is composed by the following programs Spring contains Geoprocessing and Image Processing functions Scarta is used for interactive map production application a a Impima is used for converting satellite images or part of them to the SPG image format Vee ood te be o aapi weer te Geetteed PRE The wees oe cteetfed Hy Her eee e8ientes F you eet te outnet WWE aot yew oe p ogee woe pet et yer et obia eet oh ee totter Vere 8 pee oe WOT o ragetered weer chet the ton Sebec
37. d air photo interpreter to verify the classification results with the source image Skirvin et a 2004 Workstation Assessment The skilled air photo interpreter assessments can be performed at the workstation A workstation assessment involves randomly selecting multiple locations on the land use land cover classification noting its class assignment and comparing it to its actual interpreted land use land cover as determined from the image Random points can be generated using an extension to ArcGIS called Hawth s Spatial Ecology Tools gt Generate Random Points tool http www spatialecology com Hawth s Tools is an open source extension for ESRI ArcGIS Hawth s tools is designed to perform ecology related spatial analysis It is a useful tool to aid in sampling design allowing for a variety of sampling techniques stratified random sampling grid sampling polygon sampling Field Validation Protocols for collecting field data will vary according to the classification and change detection methods employed In general there would be a requirement for field data to train the classifier and verify the accuracy of classification then a separate field data set would be required to focus on the changes in classification that are revealed by the multitemporal change detection analysis In the case of unsupervised classification minimal field data are required initially although the biological context would be difficult to genera
38. d to build pyramids Keep ArcMap open we will use data from the classification2_ pos 7 7 tif Properties window in the next section Add Data xs infolaye dbf linktable dbf metadata dbf pictures dbf Name dassification2_pos7 7 tif Add Show of type pstasets and Layers v Cancel O Untied ArcMap cinto Fie Ed View Bocima imet Selection Geoproceming Cuitomuze Windows Help jas Ly 3 lt SSSEO amp ve ARMOR Layer MO clamification2 por 724 v A i y iz By Tatie Ot Contents aranci ak TEREE amp layers E classification pos 77 18 is itt chee Fah ee gt aa AS ce T RUIG ie med 2 X amp F Spee g Sue ci dagr E Tapas Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 39 2 7 Calculating Landscape Metrics with FRAGSTATS In this section we will run our statistics for canopy cover on our Tacoma imagery We will use the classification2_pos_7_7 output to estimate percent canopy cover and the total area of canopy for our imagery But first for FRAGSTATS to make sense of our data we need to assign our pixels names which can then be attributed to our classes For example for our blue colored pixels to be called water we need to provide FRAGSTATS with information that connects the name water to blue pixels We will use ArcMap and the Notepad program to d
39. e distribution of patches with respect to each other regularly dispersed vs clumped Contrast The relative difference among patch types shape complexity The relative amount of edge per unit area the fractal dimension Adjacency contagion The tendency of a patch type to occur next to another patch type Connectedness The functional joinings between patches Boundary The transition zone between patches Boundary Richness The number of different boundary types Boundary Convergence The proportion of of points where more than three patches converge Boundary Diversity Index The number and the area evenness of boundary types Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 15 According to Jaeger 2000 fragmentation measures should 1 increase monotonously when new sites are converted into intensively used areas e g into settlement areas and roads 2 have an intuitive explanation 3 not be too sensitive to the omission or addition of very small residuals 4 not require much data input 5 be as simple as possible from a mathematical point of view New metrics and empirical data supporting the interpretation of metrics relative to ecological criteria continue to be introduced in the literature at a high rate According to Simberloff 1999 the most critical issue facing landscape ecologists is the verification of the ecological relevance and meaning of landscape metr
40. e most optimal dates and frequencies for our analysis However factors such as lag effect of LULC change due to population increases need to be considered Furthermore pre 1930 s aerial photography is not available and the more historic data sets might be in hard copy format requiring pre processing Population 1860 1860 1900 1920 1940 1960 1980 2000 Year Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 48 4 CONCLUSION Here we summarize the most important take home messages from the workshop Free and open sources software and data greatly reduces the cost of remote sensing applications such as urban canopy mapping Such software requires patience and extensive troubleshooting some computer skills are required to make this process successful A geospatial scientist will have an easier time using this type of software then an average user as concepts such as projections pre processing orthorectification are already known to the user citizen scientists and volunteers will need training NAIP imagery is freely available continuous often collected using the near infrared range and provides historical datasets nationwide however eastern and central U S are acquired at higher temporal frequencies then PNW Temporal analysis of classified images can show areas of change as well s capture the rate of change The data is spatial thus can be c
41. e type based on nearest edge to edge distance for each patch in the Sa a ca ca os with a neighbor divided by the number of patches with a Contagion and Minus the sum of the length of each unique edge type involving the Interspersion corresponding patch type divided by the total length of edge involving Metrics Interspersion Juxtoposition Index the same type multiplied by the logarithm of the same quantity summed over each unique edge type divided by the logarithm of the number of patch types minus 1 multiplied by Diversity and a OO Shannon s Diversity Index ieee the sum across all eles types of the pears Evenness atch i lat p on Metrics Minus the sum across all patch types of the proportional Shannon s Evenness Index ase aa m m a a by that proportion divided Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 58 APPENDIX G Future Readings To assure that we can keep you most up to date we have created an online reference resource using the Mendeley com site You can view the content of this reference through the link below UW RSGAL Canopy Assessment Readings http www mendeley com groups 917001 uw rsgal cano assessment readings If you wish to comment on the literature and uppload additional fields please send an email to rsgal uw edu explaining your request and use the group name in the email subject Geospatial Canopy Cover Assessment Worksh
42. ed and the unsupervised methods in a single procedure One of the critical issues that must be addressed after selecting the broad approach is the choice of algorithm statistical non parametric fuzzy logic evidential reasoning neural nets others Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 9 Supervised Classification A supervised classification utilizes the analyst s knowledge of features on the ground Mathematical algorithms use the spectral properties of known ground features training sets to determine the class identity of unknown image pixels Unsupervised Classification An unsupervised classification uses clustering algorithms to distinguish pixels sharing similar spectral characteristics The relationship of the resulting spectral clusters to ground features is subsequently identified Per Pixel Classification Traditional classification methods have used Landsat satellite imagery inset a in figure below to produce LULC maps e g 2001 National Land Cover Database by assigning individual pixels to a specific class based on spectral signature inset c in figure below The spatial resolution of this imagery is 30m meaning that anything on the ground less than 30m in size or 900m2 9 688ft2 will be generalized to the main class represented in that pixel 9 688ft is larger than most parcels in Seattle so detail beyond Anderson Level or Il cannot be
43. er exploring initial accuracy assessments The end map user should be mostly concerned with the users accuracy as it lets them know the type of accuracies to expect on the ground but a check of the producer s accuracy can help to determine in the most appropriate methods were used while discussing this with the analyst The structure of the error matrix allows for the production of overall accuracy of the entire classification and a producer s accuracy and user s accuracy for each class All three of these values should be reported since the future use of the classification is unknown Felix and Binney 1989 Below is an example of an Error or confusion matrix which can be based on field validation auxiliary data validation or even visual interpretation of remotely send data Visually Interpreted nenial Forest Background Errors of Class P Matrix Total Commission o Impervious a jaw A 2 Forest ee ee K R K Y Matrix Total n E of NPE A D R T Y Z Omission Overall Classification Accuracy A R Y n The overall accuracy is calculated by dividing the number of correctly classified locations by the total number of locations assessed for accuracy n 100 The correctly classified locations are those on the diagonal within the matrix A R Y Story and Congalton 1986 The errors of commission those that yield the user s accuracy are found in a column on the right side These values represent that the classificatio
44. ew training data for our classes This is okay refining a classification can take a long time however it is important to be comfortable with your classification before applying it to a larger data set We will now try to refine this classification by selecting new training data for our classes for reference misclassification is often caused by shadows In the image below you can see that there are lot of water pixels too far inland These are mostly shadows from tree canopies and buildings that are being placed into the water class because shadows and water are both dark We will have to supply more training data to make these classes more accurate We will fix these regions in the following steps Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 34 ap i a P sl aw gt a pe Like wa FB DE L LLA EE Iz O viet tain e ERE R age Rs Fas BEE e a5 Lh Sie AT Ee Oe PER A ay F mST 4 ee 7 Inet Cpe hears he i a X at M sE AN t To Sa Lie 7 Beas ACENTE dj gh LEL N T a a NE Step O Turn on segmentation _30_100 by highlighting it and then check the Labeled box in the bottom of the control panel Turn classification1 on by highlighting it and then check Classified box in the control panel Step P Reopen the Training dialog box and make the trees Theme active by clicking on it Click on some of the shadowed
45. field studies Validation of landscape metrics with biological phenomena is particularly complicated when targeting a wide phylogenetic range of taxa such as aimed for in a biodiversity monitoring program because patterns of the landscape may affect various organisms differently Readily available landscape analytical programs such as FRAGSTATS McGarigal and Marks 1994 and Patch Analyst Elkie et al 1999 can be used to calculate a large number of metrics for the landscape patches derived from a classification of remotely sensed imagery Since an overwhelming number of metrics are described in the literature the choice of appropriate sets of landscape metrics must be based on criteria meaningful to the program at hand Monitoring forest biodiversity requires that landscapes be compared temporal changes be evaluated and possibly that landscape effects be predicted The criteria for selecting metrics should include e Simplicity of interpretation Simplicity of mathematics Low correlation statistical independence among the set of metrics Relevance as disturbance fragmentation indicator General Description Composition The variety and relative abundance of patch types Richness The number of different patch types Diversity The relative abundance of different patch types Structure The configuration of the landscape size distribution The relative abundance or frequency of patches in different size classes Dispersion ee i pers Th
46. for specific wavelengths It is beyond the scope of this workshop to cover wavelengths in detail however it is important to know that Tacoma_1 corresponds to red Tacoma 2 corresponds to green Tacoma_3 corresponds to blue and Tacoma_4 corresponds to infrared To display an image in true color what you would see in the real world you would match the Tacoma_1 band with red and so on However instead of doing this we will load our image in false color by loading the infrared band instead of the red band this will make it easier to identify vegetation in our imagery Step J If the control panel window is not open click the control panel icon on the toolbar Step K To display our image in false color highlight the _ Conte Panel Tacoma_4 band and check the box for R in the bottom portion Main 4 of the window Do this for Tacoma_2 Available Infolayers Selecteds Infolayers for blue and Tacoma_3 for green If your image a does not appear click the Zoom IL icon to zoom into your image Tacoma 1 With the image loaded and in view we are now ready to conduct o pomi a segmentation A Ahamd Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 27 True Color False Color Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Task 6 Perform a segmentation Step A With your image loaded in fal
47. h the ls Since then hundreds of professionals have enyoved the use ease ag 5 Dus to mts aor same version number 3 1 but will be assigned 2 unique build number beginning with Fi populanty the program was completely revamped in 2002 version 3 3 The Pe arate Fire build 1 If you are not suce what build cumber you have it is listed is the About we corrently a ie smother mayor revamping which wall result tn the release of version 4 0 RAGSTATS desenption available from the Help Menu after openning the FRAGSTATS Si sometime m 2 Shortcourses execetable The latest build will be available frost the download inik below and the changes urses associated with each build will be desenbed in the release gotes Substantial changes to the mp purpose of this web site s to facthcate Gasemmation of the software currently versne software primarily associated with the addition of new features e g mew metrics wiil be 3 and to facilitate commumcation among FRAGSTATS users feleased as a new version mumber Again all changes associated with each release will be documented m the release notes Thus it is important to check the release notes before About The Developers downloading a new vernon The original version of FRAGSTATS published in 1995 was developed by Dr MoGarizal aed Barbara Marks o f Oregon State University Ms Marks was the programmer and Note in vernon 3 3 a number of the mew metrics described in the FRAGSTATS Metne docmses
48. ial to maintaining close ties with individuals and communities sharing knowledge to inform decision making and in turn listening to questions concerns and feedback regarding their needs The Remote Sensing and Geospatial Analysis Laboratory RSGAL is an applied science research laboratory meaning that analysis techniques developed by the lab are applied to real world issues through pilot studies hands on workshops and presentations to community focus groups Through this free workshop we aim to educate personnel of small local government and public organizations who are an under represented audience in the field of remote sensing technologies in hopes of providing innovative time and cost efficient approaches to sustainable urban management The target audience is urban foresters city planners parks personnel etc who are interested in learning more about their city s forest resources but who don t have the budget to do so Each workshop attendee will receive training in one of the most technologically advanced image analysis techniques which is valuable for both critically evaluating subcontracted work or conducting your own assessments now and in the future Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 5 1 1 Purpose and Objectives Successful management strategies require a base of knowledge and information about forest resources in order to develop action plan
49. ibe F yey See Cyes ver peeved OF fhe oomai belt bee put pere Check the improvements of this new version at News me Voeget Pewewerd anre Ton et eens of met EP fe peered het etre amp ow Gtebeee Choose the Site for downloading SPRIING 5 1 7 WINDOWS 32Bits Steet eee eee eee ee Source Portuguese Spanish English French POPS te riean NTa o DPI INPE Brazil Lene LEAME README aLL A Instala Complete m Banco Demo Inatala Simbolos em Inatalacion Simbolos em Symbol instalation Lnstallation de symboles BMP BMP BMP BMP Spain LEIAME LEAME README IME Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 20 2 2 Installing FRAGSTATS In this section we will install the freely available program FRAGSTATS version 3 3 FRAGSTATS is a landscape ecology tool used to quantify landscape structure from data derived from imagery It is provides the user with metrics to statistically describe a landscape Task 2 Install FRAGSTATS 3 3 Spatial Pattern Analysis Program for Categorical Maps Step A Navigate to the FRAGSTATS website Ctrl left click on the following link http www umass edu landeco research fragstats fragstats html Step B Click FRAGTSTATS Download in the left hand column Step C Double click the FragSetup33zip file and save it to the desktop Step D Unzip the folder on the desktop right click and click E
50. ics Based on this literature review we recommend metrics in Appendix F Landscape Scale It serves well to identify the scale level at which the relationship between structure function and change can most accurately explain the landscape dynamic at hand Landscape metrics deal in three levels of scale landscape class and patch Forman and Godron 1986 The smallest unit in a landscape is the patch A patch is a relatively homogenous nonlinear area that differs from its surroundings Forman 1995 Metrics for this scale contain information for every individual patch regarding its spatial character as well as measures of deviation from other patches within the same class and patches from other classes within the landscape An assemblage of patches based on a common attribute such as land use land cover type constitutes a class Class metrics summarize the overall spatial character of all patches within the class as well as the composition and spatial configuration of the class An assemblage of classes and their respective patches constitutes a landscape mosaic Fragstats 2002 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 16 1 2 4 Change Detection A number of approaches for change detection using remotely sensed multi temporal data sets are described in the literature see Lillesand and Kiefer 1994 Jensen 1996 In general to reduce the need for image calibration
51. indow Step B In the Post Classification window click on classification2 and choose a Weight of 7 and a Threshold of 7 then click Apply Step C When the post classification is complete rename it from classification_pos to classification_pos_7_7 by right clicking it clicking on Rename then click OK to close the rename window Displayed in Figure 2 are two different post classification with two sets of parameters for weight and threshold one with a weight of 5 and a threshold of 2 and another with a weight of 7 and a threshold of 7 Notice that the original classification did not classify all pixels you can see this in the upper left corner and middle of the image labeled none a post classification was not run on this image Also notice that the dark line near the top portion track shows up in the 7_7 post classification but not the 5 2 Because this line might not actually be grass it appears that the 7_7 has misclassified these regions It appears that the 5 2 post classification has outperformed the 7_7 because of this we will choose the 5 2 parameters for use in later steps Move onto section 2 6 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 37 Figure 2 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 38 2 6 Exporting Imagery from SPRING Once we are comfortable with the output of our c
52. ink Processing System Downloads Support Manuals Data Publications News Links oh j A C Install the SPRING 5 1 7 software package A SPRING is a state of the art GIS and remote sensing image processing system with an object oriented data model which by C li l C ki ng O n Down oad i l n t h e u p pe r l eft provides for the integration of raster and vector data representations in a single environment SPRING is a product of Brazil s National Institute for Space Research INPE DPI Image Processing Division with assistance from e EMBRAPA CNPTIA Brazil s Agricultural Research Agency e IBM Brasil e TECGRAF Computer Graphics Technology Group Step B To register your copy of SPRING gt PETROBRAS CENPES e K2Sistemas e nte r yo u r e m a i a d d ress a pa SSWO rd a n d The SPRING project has received substantial support from CNPq National Research and Development Agency through its nd PROTEM CC GEOTEC project programs RHAE a choose the SPRING for Windows 32 version serine main teatures e An integrated GIS for environmental socioeconomic and urban planning applications fro mMm t h e Ve rs i O n d ro p d OW n ist wh ile A multi platform system including support for Windows95 98 NT XP and Linux e A widely accessible freeware for the GIS community with a quick learning curve e To be a mechanism of diffusion of the knowledge developed for the INPE and its partners with the introdu
53. into account More advanced methods of correction have been developed that model the relative position of trees and the amount of shadowing based on slope and aspect Gu and Gillespie 1998 or that model the mixture of sun lit crown shadowed crown and background within a pixel Li and Strahler 1985 Woodcock et al 1994 Aerial topography can be corrected using DEM s produced from stereo photo pairs using photogrammetric techniques while some also rely on the above mentioned SRTM data However one needs to be careful when using a coarser resolution dataset for topographic corrections especially in heterogeneous areas such as urban landscapes In these cases LIDAR is becoming the more appropriate source of DEMs Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 8 Geometric corrections Orthorectification and Mosaicking During the process of acquiring digital imagery geometric distortions occur due to variations in the sensor s platform the earth s curvature relief and so forth The company that supplies imagery frequently corrects systematic distortions in house or provides the values of variables for use with correction algorithms Ground control points GCP s or accurately located ground features that are identifiable on the image are useful for correcting random geometric distortions Distortions in scale occur because the distance between the sensor and the earth
54. kers with allocating funds and resources to areas that have been negatively impacted from development and need the most attention to prevent further canopy cover loss Remote sensing technologies can provide a means to explore a variety of continuous environmental variables over large areas including canopy cover and other land use land cover types Remote assessments are reasonably simple and can be conducted quickly inexpensively and without access or disturbance issues encountered in ground based data collections These assessments provide a means to measure and monitor complex urban environments and their dynamic ecologies For instance Canopy cover surveys and forest pattern metrics are useful to help a city quantify current tree cover status Hunsinger amp Moskal 2005 determine the locations and drivers of canopy loss or gain Turner amp Gardner 1991 and monitor these trends over time Moskal et al 2004 These data can then be used to select inventory sampling sites establish tree protection requirements for new developments assist with urban forest health management and determine target areas for planting projects Remote sensing techniques can be applied to the analysis of other human and environmental dynamics within urban systems to aid in sustainable planning and management of these areas Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 4 Outreach is essent
55. lassification and post classification we are now ready to export the data for viewing in GIS or for use in the statistical analysis program FRAGSTATS Task 9 Exporting your classification Step A In the control panel window highlight the classification_pos_7_7 then from the toolbar click File Export Export Vectorial and Matrical Data to bring up the Export window Step B In the Export window make sure that the format type is TIFF GeoTIFF in the Format drop down box and click Save Step C In the Save File window choose a folder to save your file into we recommend using the same folder your database and imagery are stored in and name the file classification2_pos 7 7 and click Save When you are finished close the Export window and close SPRING Step B External Data t TIFF GeoTIFF Mono RGE Infolayer classification2_pos_ _ Step C la Save File wy m b workshop gt Organize New folder Fr Favorites _ MB Desktop ib Downloads 1 Recent Places 1 est2 classification po Tacoma LJ Libraries TT Documents a Music k Pictures Videos jm Computer amp MAIN C cay BACKUP E File name classification2_pos_7_7 Save as type Arquivo tif Hide Folders Step D To view your classification and its properties open ArcMap and start a new blank map Step E Use the Add Data button to add the classification2 pos 7 7 tif file to your map click OK when you are aske
56. magery that is more suitable for vegetation studies as vegetation vigor is prominent in the near infrared region of the spectrum Example of black and white monochromatic true color and near infrared imagery note the differences between vegetation and water in these types of image More information about NAIP imagery can be found in Appendix B For about a decade now LIDAR has been flow commercially and consortia have established regionally Puget Sound LIDAR Consortium and national USGS Click Unfortunately LIDAR is not currently collected by a nationwide program at a yearly temporal interval and some regions still do not have LIDAR coverage while others have audited coverage due to the quickly changing technologies One should remember that even older LIDAR data can provide very useful data for example the 3D point cloud can be processed to extract a surface model which can then be used to produce a DEM Canopy models and building models have also been attempted and are very useful inputs for remote sensing analysis More information on LiDAR can be obtained on the RSGAL LiDAR workshop website at http depts washington edu rsgalwrk lidar Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 7 Another source of free data for monitoring large areas such as counties and states is the digital satellite imagery of Landsat TM Although comparatively low spatial resolution
57. mercial Complexes 115 Mobile Home Parks 16 Mixed Urban or Built up Land 116 Transient Lodging 17 Other Urban or Built up Land 117 Other 2 Agricultural Land 21 Cropland and Pasture 22 Orchards Groves Vineyards Nurseries and Ornamental Horticultural 23 Confined Feeding Operations 24 Other Agricultural Land human uses of the landscape Remote sensing imagery captures characteristics of the Earth s surface but it takes an interpreter s knowledge about shape texture patterns and site context to derive information about land use activities from information about land cover LULC classifications typically utilize some modification of the Anderson hierarchical system with generalized LULC classes described at Levels amp Il and more detailed classifications for Levels III amp beyond Classification and Mapping For monitoring the forest landscape over time conceptually simple and practically similar approaches to classification of digital data is required across the sample design The databases used for detecting changes must be as consistently and accurately mapped as possible in order to avoid errors in accuracy Often categorized as supervised and unsupervised several approaches for multispectral image classification Lillesand and Kiefer 1994 are available in standard image analytic software packages Intermediate between these two are a series of modified approaches which attempt to capitalize on the strengths of the supervis
58. n Analysis in SPRING eesessssssesesesersressererereresrererersrossererereeesserorerreosreressreersreressrrerss 36 Z2 6 EXpOrting Imagery Onm SPRING airen N a a a 38 2 7 Calculating Landscape Metrics with FRAGSTATS cc ccssccccsssececcesececceeseceseeseceeeeneceeseesecessunecesseens 39 3 REPORTING cocada a a a sae eae ea eee eae 45 ASCON CLHUISTON r A a A natal tareee 48 SZOFNERRESOURCES nen eer ear ry eee PCr pnt Ur yee rg er ee yO Pee eae Pere tre erie Per Sere eee 49 REFERENCES cae ecstasy tenance aa es sae ee aes aeons aman eae outa 50 APPENDIX Az Data SOURCES cic vacuriciecsaccatedcacin vehinncesvad ue aiienpasnaiieadua awe oica E eee Gabe braid es bend aaa 52 APPENDIX B Aerial Photography and NAIP aineenaan ess ui daeceatunttessiane oad N ads aun n et 53 APPENDIX C Satellite Data Pre PrOCeSsin ess iassscxtic certs s access aa estaes ded oes eee ee 54 APPENDIX DO Concepts ana DeTinitlOns srarsiaee catcnsGiseen toad N a N a 55 APPENDIX E Accuracy Assessment Statistics ccccscccseccseccseccseccsecceecceeeceusceuscesscesccessceseeeseceseeeseceseenseetseenes 56 APPENDIX F tandscape Metrics Descriptio NS aeriranieinen ian a A A ANAN R EI 57 APPENDIX G FUtUrE REJANG S mrina n a a a A AAA 58 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 3 1 BACKGROUND With over 3 4 million residents U S Census Bureau 2010 the Seattle Tacoma Bellevue area is experien
59. n algorithm generated J number locations to a specific land use land cover class Only A number of those locations were an instance of correct class assignment Q W committed omitted from their correct class A J generates the probability that a pixel classified on the map actually represents that category on the ground Story and Congalton 1986 The errors of omission those that yield the producer s accuracy are found in a row at the bottom These values represent that within that year s study area there were D number of instances of validation locations within that land use land cover class column A number of those locations were actually produced as such class with B C omitted committed to other classes A D generates the probability of a reference pixel being correctly classified Story and Congalton 1986 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 14 1 2 3 Landscape Metrics Landscape metrics quantify the composition and pattern of patches in the landscape and can be categorized according to their general function in the table below In addition to simple description of the landscape certain metrics attempt to measure aspects of landscape pattern thought to reflect or influence underlying ecological processes In most cases however the relationships between landscape metrics and biological phenomena have not been validated with
60. n collected under the same conditions as the reference image Elvidge et al 1995 describe another relative radiometric calibration called the Automated Scattergram Controlled Regression method Using the same two sensor bands from different images a scattergram and regression line is plotted Pixels close to the regression line are considered to be unchanged and are used to perform the radiometric correction Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 55 APPENDIX D Concepts and Definitions The best place for looking up remote sensing terminology is the Canadian Centre for Remote Sensing glossary at http www ccrs nrcan gc ca glossary index e ph We provide the most often used terminology here Accuracy Assessment a procedure that_compares not only the classification map but also the ground reference test information and to what degree the classifications represented thematically are actually correct to the corresponding ground reference locations Change the alteration in the structure and function of the ecological mosaic over time Change Detection a comparison of images from multiple dates with the intent of detecting changes in areal extent over time and space Classification a technique that automatically classifies all pixels in an image into land cover classes or themes Classification can be unsupervised controlled by the computer which decides to gro
61. n of those objects into themes or classes An assumption of GEOBIA is that the landscape is made up of homogenous patches which can be separated by their spectral signatures The simplest and longest standing approach to image classification in remote sensing is the per pixel based supervised and unsupervised classifications The per pixel classification categorizes remotely sensed image pixels by a theme such as land cover The classification uses the numerical data of spectral information contained within each pixel as the basis for this categorization Lillesand and Kiefer 2000 The user then has the ability to select the classification algorithm or decision rule based on the nature of their input data and their desired results The algorithm or decision rule determines the manner in which pixels are assigned to categories by evaluating each pixel s spectral information and the spectral information of its neighbors Types of decision rules include parallelpiped minimum distance and maximum likelihood classifiers Jensen 1996 Feature extraction classification is an object oriented approach to image classification Feature extraction uses relationships such as size shape texture directionality repetition context as well as spectral information to categorize pixels Lillesand and Kiefer 2000 Both object oriented and per pixel Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA
62. nd digital terrain data IEEE Transactions on Geoscience and Remote Sensing GE 24 139 146 Franklin J and C E Woodcock 1997 Multiscale vegetation data for the mountains of Southern California Spatial and categorical resolution In Scale in Remote Sensing and GIS D A Quattrochi and M F Goodchild Editors CRC Press Inc New York Franklin S E 1991 Topographic data and satellite spectral response in subarctic high relief terrain analysis Arctic 44 15 20 Franklin S E and P T Giles 1995 Radiometric processing of aerial and satellite remote sensing imagery Computers and Geosciences 21 413 435 Franklin S E and E E Dickson 1999 Approaches for monitoring landscape composition and pattern using remote sensing In D Farr S E Franklin E E Dickson G Scrimgeour S Kendall P Lee S Hanus N N Winchester and C C Shank Monitoring Forest Biodiversity in Alberta Program Framework Alberta Forest Biodiversity Monitoring Program Technical Report 3 Draft Report Goward S N C J Tucker and D G Dye 1985 North American vegetation patterns observed with the NOAA 7 Advanced Very High Resolution Radiometer Vegetatio 64 3 14 Gu D and A Gillespie 1998 Topographic normalization of Landsat TM images of forest based on subpixel sun canopy sensor geometry Remote Sensing and Environment 64 166 175 Hame T A Salli A Andersson and A Lohi 1997 Biomass estimation of Boreal forest using NOAA AVHRR dat
63. nd for ease of identification of photosynthetically active vegetation Imagery acquired from NAIP can be used for a wide variety of purposes because the NAIP coverage is not restricted to agricultural lands but rather imagery is often includes an entire state This means that NAIP imagery can be used for assessment of canopy change over time in an urban areas For links to NAIP imagery and coverage maps refer to the links below The image to the right provides the NAIP coverage from 2003 2010 it s a modified version of information at http www fsa usda gov Internet FSA File naip03_ O9covermaps pdf Green areas were collected at 1m resolution and brown areas at 2 m resolution White areas show states with no NAIP coverage for the year Washington State Orthoimage Portal http www geography wa gov imageextractorjs Page 53 APPENDIX B Aerial Photography and NAIP 2010 NAIP COVERAGE 2008 NAIP COVERAGE 2007 and 2009 NAIP imagery of Washington Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 54 APPENDIX C Satellite Data Pre Processing Extensive corrections are performed on satellite data We discuss main concepts below most of these are often performed by the image provider Radiometric correction Radiometric corrections compensate for differences in radiance values among images due to the seasonal position of
64. ng for comparison to any other normalized matrix of research Congalton 1991 The following formula taken from Jensen 1996 was used for the generation of the KHAT statistic r r N x g pee KHAT 2 N De p where r number of rows in the error matrix Xii number of observations in row and column i on the major diagonal Xia marginal total of row X41 Marginal total of column i N total number of observations included in matrix The returned statistic is a value ranging between O and 1 explaining the agreement between a classification map and the validation data KHAT values 0 81 1 00 demonstrate an almost perfect agreement between the classification and the validation locations KHAT values 0 61 0 80 demonstrate substantial agreement and KHAT values 0 41 0 60 demonstrate moderate agreement KHAT values below 0 40 show fair to slight agreement Landis and Koch 1977 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 57 APPENDIX F Landscape Metrics Descriptions Metric Catego Metric Descri ption Sum of areas of all patches belonging to a given class Number of Patches Total number of patches in the bias e or given class Mean Patch Size ha Average patch size Area Metrics Median Patch Size ha The middle patch size Patch Size Standard Deviation standard Deviation of Patch Areas Patch Size Coefficient of Variance Patch
65. nsing 20 1 139 152 Moskal L M and D M Styers 2010 Land use land cover LULC from high resolution near infrared aerial imagery costs and applications Factsheet 12 Remote Sensing and Geospatial Application Laboratory University of Washington Seattle WA Digital version of the fact sheet can be downloaded at http dept washington edu rsgal O Neill R V 1996 Recent developments in ecological theory Hierarchy and scale Pp 7 14 in Gap Analysis A Landscape Approach to Biodiversity Planning J M Scott T H Tear and F W Davis Editors American Society for Photogrammetry and Remote Sensing Bethesda Maryland Riitters K H R V O Neill C T Hunsaker J D Wickham D H Yankee S P Timmins K B Jones and B L Jackson 1995 A factor analysis of landscape pattern and structure metrics Landscape Ecology 10 1 23 39 Simberloff D 1999 The role of science in the preservation of forest biodiversity Forest Ecology and Management 115 2 101 111 Stoms D M 2000 Actual vegetation layer In A Handbook for Conducting Gap Analysis Version 2 0 0 16 February 2000 http www gap uidaho edu handbook LandCoverMapping Stoms D M and W W Hargrove 2000 Potential NDVI as a baseline for monitoring ecosystem functioning International Journal of Remote Sensing 21 2 401 407 Strahler A H 1981 Stratification of natural vegetation for forest and rangeland inventory using Landsat imagery and collateral d
66. o this We will use the information in the Layer Properties window to create a txt file which links pixels to names of classes Task 10 Calculate tree canopy metrics using FRAGSTATS Step A With ArcMap still open double click the classification2_pos_7_7 layer to bring up the Layer Properties window Step B In the Layer Properties window click the Symbology tab and click Unique Values in the Show field ArcMap will ask you to build an attribute table click OK From here we can see what order our classes are in This is important in creating the text file for our classes Notice that there are 6 classes The white class labeled O corresponds to unclassified pixels For display below we have changed the unclassified pixels to pink to easily identify them in our classification In the count column we can see how many unclassified pixels there are Some of these were Classified in the post classification but not all of them This is okay there typically aren t enough unclassified pixels to alter your statistics significantly If there are then you would want to return acquire more training data and conduct another classification ie m we ee ee ee ce ee ee A A 4 Mii 4 t h l ieii leh ch IET t UF Y B 3 jor aoe x a m agii a ar i Aai Aedes e e oar cee sae Mina ical _ gt Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 40 Step
67. obtained It is now a well accepted principle that this moderate resolution is not appropriate for LULC mapping in urban areas Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 10 1 2 1 Object Based Image Analysis OBIA amp GEOBIA A new classification method called Object Based Image Analysis OBIA appears to work best on high resolution satellite and aerial imagery as well as LIDAR data e g 1m NAIP Figs 1b amp 2 This form of feature extraction allows for use of additional variables such as shape texture and contextual relationships to classify features Fig 1d This method can be used on free publicly available high resolution nationwide NAIP imagery which can be classified at Anderson Level Ill and higher to achieve very detailed LULC maps for urban based planning management and scientific research To the left is a comparison between a Landsat 30m pixel resolution image a classified using per pixel method c and a National Agricultural Imagery Program NAIP 1m pixel resolution image b classified using Object Based Image Analysis method d Note the higher image detail in the OBIA classification and the greater number of classes possible Geographic object based image analysis GEOBIA is a method of classification involving the delineation segmentation of similar pixels into discrete objects and is followed by the classificatio
68. ompared and utilized with other spatial datasets Accuracy assessment is critical and necessary workstation assessment can sometimes substitute field based assessments Sampling for field based assessment is a complex process Keep spatial metrics simple for ease of interpretation Results should be interpreted carefully taking care to understanding the data differences Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 49 5 OTHER RESOURCES These will be posted on the final workshop website http depts washington edu rsgalwrk canopy Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 50 6 REFERENCES Box E O B N Holben and V Kalb 1989 Accuracy of the AVHRR vegetation index as a predictor of biomass primary productivity and net CO flux Vegetatio 80 71 89 Bruzzone L and D Fernandez Prieto 2000 An adaptive parcel based technique for unsupervised change detection nternational Journal of Remote Sensing 21 4 817 822 Cao C and N S N Lam 1997 Understanding the scale and resolution effects in remote sensing and GIS In Scale in Remote Sensing and GIS D A Quattrochi and M F Goodchild Editors CRC Press Inc New York Chuvieco E 1999 Measuring changes in landscape pattern from satellite images short term effects of fire on spatial diversity International Journal of
69. op by UW RSGAL March 3 2011 University of Washington Seattle WA
70. orm numerous segmentations with different parameters to get the ideal segmentation No one way is correct but you should be aware of which options work best for your imagery Settings that are too fine or too coarse could create issues in later steps of the OBIA process Step B Segmentation Region Growing CAT Image Tacorna_1 CAT Image Tacoma_2 CAT Image Tacorna_3 CAT _Image Tacorna_4 Similarity 30 Area pixels 100 Initial MD Band of Exclusion Output Category CAT _ Image IL Name segmentation_30_100 Arc Smoothing Yes No Bounding Box Apply Close Help Step E Click Apply to run your segmentation This will take about 5 minutes and would take much longer if we were working with a larger image i e an image with more pixels or had chosen smaller values for similarity and pixels While your segmentation is running take time to read the passage below for additional information on the Similarity and Area parameters and also to look at Figure 1 which highlights differences in segmentations with different values for these two parameters Step F After your segmentation is complete a window titled Assistant will appear and will draw your segmentation close this window and move on to Step F Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 29 The Similarity and Area parameters dictate how pixels are gro
71. r on the toolbar if it is not already selected then click some water in the upper right of the image A group of pixels will be outlined in blue with these pixels selected click Get in the Training window Do this for 4 more regions of water pixels and then move onto Step G x f a Training R Name water Themes Create water Total Num of Pixels 481680 ete m Mode Normal Group Ungroup C Show All Type Acquisition Test Contour Polygon Rectangle Region gt De hod Aquisition 1 Num of Pixels 236589 B N Change Tar j p Aquisition 2 Num of Pixels 245091 a ee y te z l Delete Dn reales iy 5 v Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 33 Step J Repeat Steps D F for the classes trees roads buildings i e impervious surfaces and grass create your classes in this exact order Notice that in this false color image vegetation is red this is because living vegetation reflects in the infrared wavelength which gives it a red appearance Also notice that the grass in the track in the lower left of the image is not red This is because this is artificial turf and not actually grass which absorbs infrared giving it a gray appearance You would not want to acquire training samples for grass from this field Step K After acquiring all of your samples click Save and then click Close Step L
72. rial photography APPENDIX B was originally collected in black and white from the 1930 s to about the 1970 s but is currently mostly collected in near infrared Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 47 Below we can observe the types of imagery that was used for the analysis we see that the earlier imagery was in black and white so was the 1990 imagery The 2003 imagery was a false color near infrared and the most recent imagery was true color Not only were the spectral characteristics of the imagery different but so were the spatial characteristics as shown in the table below Thus the dramatic changes observed between 2003 and 2004 is actually due to the landcape metrics being impacted by both the spatial and spectal resolution of the data used for the analysis We are often constrained in our analysis by the lowest resolution of our data be it spatial or spectral Some things that can be done to avoid the these discrepancies in data include removing the most different imagery from the analysis or resampling data to the same spatial resolution 1979 Aerial Photograph Black amp White One other key decision one needs to make when undertaking image analysis is the date and the temporal resolution of the imagery The graph to the left shows the changes in population for Landscape A and B When many dates of imagery are available we can choose th
73. rks and wall be included the mex version as soon as we have thoroughly tested them Also all of the FRAGSTATS documentation is viewable as HTML fies from the documentatson web page and via the on line Help function in the FRAGSTATS software Disclaimer This software is in the pubic domain and he recent may aot assert any proprietary nights thereto nor sepresent it to Tar as other than an Oregon State University coe program version 2 x or University of Massachusetts produced program version 3 RAGSTATS is provided as is without warranty of asy kind inctading but not imited to the amped warranties of merchantability and fitmess for a particular purpose The user assumes all Dp for the accuracy and suitateary of this program for a specific application Iz no event will the authors or the University be liable for any damages tt mchadeng lo Ost profits lost savings or other meadencal or consequential damages amang from the use of or the inability to use thes program Download FRAGSTATS Last ur d 101700 FRAGST TATS software verior 33 build as g 2 several files associated with the latest software release including Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 21 2 3 OBIA Project Setup and Segmentation in SPRING 2 3 1 Creating a Database In order to work with data in SPRING you must first create a database Databases are where all of your project folder
74. s and establish long term goals The mission of the University of Washington UW Precision Forestry Cooperative PFC is to develop advanced technologies to improve the quality and reliability of information needed for planning implementation and monitoring of natural resource management to ensure sustainable forest management and to increase the competitiveness of Washington s forest sector As part of PFC the Remote Sensing and Geospatial Analysis Laboratory RSGAL aims to provide a research rich environment and exceptional resources that drive scientific investigations of multi scale dynamics of landscape change through innovative applications of remote sensing and geospatial tools and promotes a transdisciplinary approach for sustainable management solutions to pressing environmental issues Consistent with these missions the main purpose of this project is to provide guided analytical training to urban foresters land managers and city planners in an innovative technique to quantify tree canopy cover using high resolution aerial imagery calculate forest change metrics and select sampling sites for ground based tree inventories The benefit of undergoing such training is the ability to deliver the results from the report produced in the workshop with confidence and authority for advocacy purposes Participants will be provided with 1 Preprocessed sample imagery 2009 NAIP 2 An accurate technique for analyzing these data that is repe
75. s are stored Here we will create our database and define its spatial projection Task 3 Create a database Step A Open SPRING a window titled SPRING 5 1 New Features will appear close this window Step B Click on File Database in the upper left corner of the SPRING window Step C Click on the Directory button and navigate to the Desktop and click OK Step D In the Databases Name field type in your last name Step E Choose a database type from the Manager drop down list we recommend using Access or dbase as your database type Click Create you will be prompted to assign a password click no unless you want a password for your database it is not required Step F Now click Apply to load your database Steps C D and E Step C Databases cs Browse For Folder C Wsers Desktop Databases se A Kirsch Justin L jJ Computer th Network d FragSetup33 di New folder J RSGAL_THUMB di tacoma_wkshp 1 Martger Access ma Change Passy rd M Aiii mia di workshop a Create Apply Delete Ez Name your last name here Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 22 2 3 2 Creating a Project After creating a database you are now ready to create a project within your database This is where all of your imagery and outputs specific to a project are stored More than one project can be created within a database
76. s important to be aware of the processes involved in generating these data Satellite pre processing is even more extensive and is discussed in more detail in Appendix C Topographic Corrections If the slope of the study area exceeds a certain value usually about 25 degrees then a topographic correction is required in order to accurately classify imagery Topographic slope and aspect can contribute to distortions in remotely sensed data This is particularly true in mountainous areas completely shadowed by terrain where the region of interest may be blocked from receiving solar irradiance Requiring a DEM as input topographic slope aspect corrections aim to compensate for differences in pixel values that are due to the object s orientation to the sun s position rather than to the reflectance properties of the objects on the ground The amount of solar light that each pixel theoretically receives is calculated from the DEM which is modeled to determine the brightness value to add or subtract from the original brightness values of the image For surface normalization simple cosine corrections have shown improvements to classification accuracy Franklin 1991 Franklin and Giles 1995 and Dymond 1992 However both over corrections and under corrections occur using these simple cosine models Civco 1989 Although the sun to sensor angle is corrected to that of a flat terrain differences in tree canopy shadow due to slope and aspect are not taken
77. s on the ground and atmosphere e g water pollution CO concentration and net primary productivity For the detection of disturbance over short periods of time when vegetation succession is not evident calculating reflectance values from raw radiance values may not be necessary Hame et al 1998 On the other hand for detecting trends in disturbance patterns over longer time periods absolute calibration may be required depending on the method of change detection In addition for detecting subtle and progressive change in reflectance values due to vegetation succession over a 50 year trajectory calibration to absolute radiance is necessary As an alternative to radiometric calibration which converts the entire dataset from digital number values into ground reflectance values relative radiometric normalization of multi date imagery can be used for change detection Jensen 1996 Relative radiometric normalization does not require that reflectance data be taken on the ground during the time of image acquisition an impossible task when using older image data For example one method of relative radiometric normalization has the following steps Jensen 1996 A reference scene is chosen against which to calibrate The next step is correlating the brightness values among certain invariant targets in the reference scene to other scenes Regression equations are applied to the other image data to calculate what the brightness values would be had they bee
78. se color click Image Segmentation on the toolbar to open the segmentation window Step B Select all four bands by clicking on them in the Similarity field type in 30 and an in the Area pixels field type in 100 Step C In the IL Name filed type in segmentation 30 100 to give a name to your segmentation Step D In the Segmentation window click the Bounding Box button to bring up the Bounding box window Here repeat the same steps as you did on section 2 3c Step F to get rid of the decimal places trailing your coordinates Again be sure to return your radio buttons to Planes and Project and make sure both Hemisphere N radio buttons are selected Page 28 2 3 4 Segmentation In this section we will perform a segmentation sometimes referred to as a delineation Segmentation is the first step in the object based image analysis OBIA process which delineates features in an image by grouping similar pixels together and dividing those pixels into regions The segmentation is one of the most critical steps in the OBIA process so it is important to be comfortable with the segmentation output You can change the parameters of the segmentation which changes how closely similar pixels are grouped together and also how large those groups of pixels are allowed to be We will display 4 different segmentations which use various parameters Figure 1 The same parameters will likely not work for 2 separate images so one likely has to perf
79. tabon and heted in the dialog boxes are sot yet operasional Thes a matrics have bees primary techmcal support person for the original release She now works for Hewlett Packard and no longer peovedes technical support The current version 3 was developed by Dr Keven MecGargal with programming by Eduard Ene and additional programmung assistance by Chris Holmes Chris Holmes was respoeasble for the wtitsal reprogramenns Hes now works for Compag Computers Eduard Ene an mdependemt consultant and associate of the UMass Landscape Ecology om is now the principal programmer Dr McGarigal continues to be the pricipal developer o FRAQOSTATS and is the primary contact person for questions and comments Dr Sam Cushman and Dr Maile Neel provided v aluable input dursig the development and testing of version 3 3 are mcboded as authors of the current version What s In This Website This web site rs logically organized into the following pages e FRAGSTATS Documentation This page contains inks to al FRAGSTATS documentation including shde presentations from prior workshops e Dowslloads This page is where you go to dowelent the software and documentation The orgpmal FRAGSTATS version 2 and the new release version 3 eee seleases can be PEEP OF Ryo Thre reese crteene enseres te Graiiant s sabad coumetrrene mbebe crrmetwens deactivated in the dialog boxes so it is impossible to select them All of these options are in the wo
80. te without access to field observations in essence a field program to label the clusters would be required in subsequent years In the supervised and modified supervised approaches a significant field program is required to develop and implement the classification in the first year and in subsequent years the field program would be aimed specifically at areas of change For example the figure shows a random two phase stratified sample for Seattle first stratified by sampling region then by zoning Sampling Regions n a Region Commercial Mixed Use Developed Park or Boulevard Downtown Major Institutions Manufacturing Industrial Multi Family Parks Natural Area E Single Family Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 13 1 2 2 2 Error or Confusion Matrix Beyond providing an average classification accuracy remote sensing data should always include an error or confusion matrix this matrix can help one assess per class accuracies This is particularly useful when we are only interested in one class form the LULC classification such as forest canopy For example the overall classification might be 85 but often canopy classes have lower accuracies which are lost when only the overall accuracy is reported The matrix also helps the analyst to determine if the algorithm and data sets performed well producer accuracy often the algorithm is revised aft
81. tions of imagery from different dates being compared to detect changes in the landscape One advantage of this method is that absolute calibration of the imagery is not required if the changes are prominent or if good ground data for training sets are available Disadvantages are that errors in classification may be compounded in the change detection analysis resulting in a misinterpretation of change Supervised classification First independent supervised classifications using imagery from different dates are produced of the study area A matrix can then be made which compares the classifications on a pixel by pixel basis allowing to and from analysis of change Unsupervised classification A disadvantage to many methods that rely on supervised classification is the need for high quality training data often not available for change detection studies Unsupervised classification which does not require training data is sometimes combined with other methods for change detection see Combination of Analyses section Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 17 Classification of Multi temporal Data Sets Data sets composed of imagery from more than one date are combined and classified using either a supervised or unsupervised approach This method is only successful if the spectral value of the change classes differs significantly from the unchanged classes Problems
82. ttle Object Based Image Analysis OBIA Landuse Landcover LULC REMOTELY SENSED DATA We share large datasets with our partners through the UW RSGAL Pogoplug link to tutorial here Or go to fA Project Related Publications cities throus gh scienc City County State Regional Federal Other This data are available to through WAGDA to UW RSGAL IUFA research partners only contact us to request the data iufa uw edu City of Seattle GIS Data Center GIS Data Download SUMMARY City of Tacoma GIS Data Center GIS Data Download SUMMARY City of Bellevue GIS Data Center City of Marysville GIS Data Center GIS Data Download City of Bainbridge Island GIS Data Center City of Mercer Island GIS data center NOTE Not all Cities maintain a GIS website but may have GIS data if contacted directly Access is restricted but IUFA partners may contact UW to discuss help options with this at iufa uw edu S1 Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA The National Agriculture Imagery Program NAIP is a product of the United States Department of Agriculture USDA and is an aerial photography program that acquires imagery during the growing season The NAIP program began in 2003 and imagery is acquired at 1 meter resolution in natural color red green and blue but in 2007 also began acquiring data in an infrared ba
83. tware packages will be demonstrated in the applied section of this workshop 1 SPRING e http www dpi inpe br spring english e SPRING is a freely available product of Brazil s National Institute for Space Research INPE and is a state of the art GIS and remote sensing image processing system with an object oriented data model which provides for the integration of raster and vector data representations in a single environment 2 FRAGSTATS e http www umass edu landeco research fragstats fragstats html e FRAGSTATS is a computer software program designed to compute a wide variety of landscape metrics for categorical map patterns Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 19 2 1 Installing SPRING SPRING is a freely available object oriented remote sensing software package used for image processing of both raster and vector data It allows the user to automatically classify imagery which groups together similar pixels in an image into land cover classes or themes This workshop is aimed at providing an accurate and repeatable technique for extracting canopy cover from imagery and analyzing its change over time After installation of both SPRING and FRAGSTATS section 2 1b we will focus on calculating canopy cover for an area of Tacoma using imagery from 2009 Task 1 Install SPRING 5 1 7 Step A Step A Navigate to the SPRING website a a left click on the following l
84. up pixels together based on a probability of similarity in radiation or the supervised controlled by the user who decides which pixels belong together DEM digital raster file consisting of a sampled array of elevations for a number of ground positions at regularly spaced intervals Forest fragmentation occurs when forests are cut down in a manner that leaves relatively small isolated patches of forest known as forest fragments or forest remnants Function the interactions among the spatial elements that is the flows of energy materials and species among the component ecosystems GEOBIA geographic object based image analysis sometimes also referred to as OBIA or HOBIA hierarchical object based image analysis A method of classification of involving the delineation segmentation of pixels into discrete objects and is followed by the classification of those objects An assumption is that the landscape is made up of homogenous patches Landscape a term with varying definitions in its coarsest definitions it includes an area of land containing a mosaic of patches or landscape elements Landscape ecology involves the study of landscape patterns the interactions among patches within a landscape mosaic and how these patterns and interactions change over time LULC land use land cover Land cover describes natural and built objects covering the land surface while land use documents human uses of the landscape NDVI Normalized Difference
85. uped together Similarity of 10 being a higher similarity than a similarity of 30 and the size of those groups Area of 40 being a smaller group of pixels than an area of 100 Different combinations of these two parameters will yield different segmentation results Though we will only run one segmentation during this workshop it is highly recommended to run many segmentations before going through with the next step of the OBIA process By carefully examining Figure 1 you will notice that pixels are not always put in groups that make sense Sometimes tree pixels are grouped with road pixels This is a source of error that one will never fully eliminate rather through manipulation of the segmentation parameters one can attempt to reduce this error before moving on to the later steps of the OBIA process For comparison let s see how different parameters affect our segmentation Shown in Figure 1 are 4 images with combinations of values for Similarity and Area pixels of 10 amp 40 10 amp 100 30 amp 40 and 30 amp 100 You can see from the figure below how a Similarity of 10 groups more similar pixels together than a similarity setting of 30 You can also see that that an Area of 40 groups a smaller number of similar pixels together than an Area of 100 Step G Inspect your segmentation by clicking on the Zoom cursor wy and drag a box around either the field in the bottom left of the image or the patch of forest in the upper left of the image
86. ver Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 6 1 2 Remote Sensing Concepts Remote sensing is the science of obtaining information about an area through the analysis of data acquired by a device that is not in contact with the area Lillesand 2008 This typically involves imagery acquired through aerial photography or by sensors located on satellites orbiting the Earth The process from data acquisition to presentation of results to stakeholders involves many steps where a vast majority of time is dedicated to interpretation and analysis Data Digital imagery must form the basis for the development of remote sensing products to guarantee data consistency and the understandable application of scientific methods Ancillary data from field observations vegetation and forestry maps photos and other records are also important sources of information for image classification and verification Remotely sensed data is not always free however some free sources include the National Agricultural Imagery Program NAIP which has been collecting nationwide aerial photography for decades Some older aerial photographs can also be acquired from USGS at no cost and range as far back at 1930 s NAIP imagery was acquired in black and white up until the 1980 s true color was then acquired up until about 2000 and presently near infrared false color imagery is typically flown It is the near infrared color i
87. wo landscapes for four different temporal observations These calculations are reported in the graphs to the left We only use few of the simplest landscape metrics Area per Class Number of Patches and Average Patch Size to report changes on the landscape Thus our visual maps can be supported by numerical summaries Here we see that Landscape A and B lost only little forest canopy over time and in fact for Landscape B there is a slight increase in canopy from 1990 to 2003 We also observe the dramatic increase of impervious areas that is consistent with our visual interpretation of the output maps What the graphs cannot tell us is the spatial associations between impervious areas developing along linear corridors in Landscape A Geospatial Canopy Cover Assessment Workshop by UW RSGAL March 3 2011 University of Washington Seattle WA Page 46 and clustering of these areas in Landscape B On the other hand what was a difficult to visually observe on the maps in terms of the changes in forest canopy cover is now clearly demonstrated by the Average Patch Area metirc Over time the forests are becoming fragmented this Number of Patches is true and consistent for ss both landscapes We also observe that the Average number of impervious patches is decreasing over time and the Average s Patch Area metric informs us that the impervious s pe patches are getting larger in Average PatchA size This tells us that the
88. xtract All then click Extract Step E After the file has download there will be an unzipped folder named FragSetup33 on the desktop Open the folder and double click on the Setup Application FRAGSTATS will now be added to your list of programs Step B Step C FRAGSTATS Spatial Pattern Analysis Quicklinks mosis FRAGSTATS on Potion Aali Pr for C ical M rere Program for Categorical Maps NALCC PEAGSTATS ogram for Categorical Maps NALCC aes Decementation FRAGSTATS Nome Page FRAGSTATS rescsrars Downloads 7 Den abeads 2 he oofteare bout t ja at is FRAGSTATS CAPS This page rs where you go to download the software and leam about the latest release CAPS f FRAGSTATS Release Notes HAB 5 FRAGSTATS is a computer software program designed to compute a wide variety of HABIT r g IT landecape metrics for categoncal map patterns The orspeeal software vermon 2 was released FRAGSTATS 3 3 was developed as research tool Akhough we have done comundecable in the public domain durmg 1995 amsocintion wi zh the ation of a U SDA F orest RMLands PRAGSEATS testing of the software it is haghby Mceby that users will discover Sugr that arise under RMLands Service General Techarcal Report icGargal ans Miarka 1 spec ific conditions that we did not test please report all bugs immediately to Dr Kevin Vernal poo Vernal pools FRAGSTATS McGangal megarigalk Feco umasa edu Patches to comect for fags wi be released wit

Geospatial Canopy Cover Assessment Workshop

Contents

Download Pdf Manuals

Related Search

Related Contents