Iowa GAP is scheduled for completion in December, 2001. Check the state and national GAP webpages...

1
Iowa GAP is scheduled for completion in December, 2001. Check the state and national GAP webpages for updates and links to acquire the reports and GIS data. General Information www.iowagap.iastate.edu Data Download cctr.ait.iastate.edu IMS Site siberia.gis.iastate.edu/iagapims/viewer.htm IOWA’S LAND COVER MAP CREATION PROCESS Printed 11/21/01 Final Products and Data Access: The Iowa Gap Analysis Program will publish a final land cover report explaining details of the classification process. In conjunction with this report, an Atlas of Iowa Land Cover will also be published. The documents will be available on CD and on the Iowa GAP website in Adobe Acrobat .pdf format. The GIS datasets used to create the land cover and the land cover data itself will be available for downloading via ftp service. This same GIS data will be available for interactive viewing and querying over the Internet using ESRI’s Internet Map Server software. Introduction and Background The effort to comprehensively map the land cover of Iowa from satellite imagery for the Gap Analysis Program (GAP) began in 1996. The Geological Survey Bureau of the Iowa Department of Natural Resources (DNR) processed Landsat 5 imagery provided by the Multi-Resolution Land Characteristics Consortium (MRLC) into six land cover classes (Trees, Grass, Cropland, Artificial, Barren and Water) plus a Cloud class. This product is referred to as Phase 1 land cover; the processing occurred in Iowa City and was completed in late 1998. The satellite imagery was processed scene by scene using an unsupervised classification in EASI/PACE imaging software. The MRLC provided pre- processed satellite images that contained 240 clusters along with the seven original TM bands of data for two dates per scene. In some scenes the date chosen was not optimal for showing vegetation differences or there was excessive cloud cover. Therefore, additional dates for 11 out of 12 scenes were purchased with funds from several sources. This additional imagery, plus the original MRLC purchase, was used in the Phase 2 mapping process. This process further classified the six classes into map units compatible with the National Vegetation Classification System (NVCS). Creation of the Vegetation Alliances Index and Map Labels List for Iowa National GAP has realized that no one mapping strategy will work for all states. They provided guidelines for land cover products but the methods to achieve those products were up to each state to develop. Phase 2 of Iowa GAP used the image classification protocol developed by the Upper Midwest Gap Program as a starting point. Their methods were tested with our data and modified to fit our needs and capabilities. This poster is a summary of our methods along with a discussion of certain issues unique to the Iowa landscape. The Iowa Gap Analysis Program adopted “An Alliance Level Classification of the Vegetation of the Midwestern United States” as our basic classification framework. It was published in 1997 by The Nature Conservancy in a cooperative effort with National GAP to provide a vegetation classification system that complies with the Federal Geographic Data Committee’s NVCS. Naturally occurring vegetation is described hierarchically with physiognomic criteria at the higher levels and floristic criteria at the lower levels. A team of six Iowa botanists reviewed the Alliance Level Classification and modified it to more accurately reflect the existing vegetation alliances that they have encountered in the field. The result was a Preliminary Index of Natural Vegetation Alliances for Iowa” , printed in 1998. Twenty alliances listed in the Alliance Level Classification that did not include Iowa in the range were added and eight alliances were created. Conversely, alliances listed in the classification as occurring in Iowa but which team members believed not to occur here, were removed from the Vegetation Alliances Index. Special Processing Techniques General Classification Process Post Processing Techniques To improve the utility and accuracy of the final land cover product, three processing techniques were used to enhance the general unsupervised classification. Iowa Gap Land Cover Cooperators: Before Afte r Example area showing NWI Aggregation Model results Robin McNeely created a model in ERDAS Imagine to split the Phase 1 Tree class into evergreen or deciduous subclasses based on a cutoff value in Band 5 of the earliest Spring date. The model takes the tree masked Spring date, (March or April work best), and assigns a value of one to all pixels with a Band 5 value of 90 or greater and the rest are assigned zero. The cutoff value varies by five in either direction based on the imagery date and quality. The Spectral Profiler was used to get sample band values in known sites of red cedar, pine and deciduous forest and a cutoff value was selected that is closer to the signature of deciduous forest. Use Phase 1 as Base Data Compile Ground Referenced Data from Variety of Sources Enhance Three Phase 1 Classes Using the General Classification Procedure Create Iowa Vegetation Alliances Index and Map Labels List Write Metadata for Statewide Land Cover Distribute Land Cover Data via CD, ftp and Internet Map Service Post- process Barren Class This index in turn provided a framework for a Working List of Land Cover Map Labels for Iowa Gap Analysis”, finalized in 1999. Essentially, this is the list of descriptions for land cover that correspond to the map labels. The Vegetation Alliances Index was condensed and modified to reflect the limitations of Landsat TM imagery to discriminate certain spectrally similar alliances. Non-vegetation classes (Barren, Cropland, Artificial and Water) were added to accommodate actual land cover distinguishable by the satellite imagery. Unlike the Alliance Level Classification, which was created by botanists before the land cover mapping process began, the Map Labels List evolved over a year of classifying land cover. When alliances could not be reliably separated by the available dates of satellite imagery, they were aggregated to a map label. For example, the six upland oak alliances could not be spectrally separated from each other or from the maple-basswood alliance, so they all were grouped into the upland deciduous forest map label. The Phase 1 land cover was used as a base for creating the Phase 2 data. The Tree, Grass and Artificial classes were individually processed into more detailed map labels. The first step was to isolate Cloud pixels, if they existed, from Phase 1. These were used to mask the Spring date of imagery and an unsupervised classification with 50 classes was run. All available ground referenced data were used to assign map labels. The next step was to overlay the NWI data as described in the previous panel; these pixels were not modified after this step. The Tree class was processed by applying the Tree Separation Model as described in the previous panel. The result of the model is a mask that is used with the Spring imagery in an unsupervised classification and produces almost entirely deciduous forest with some mixed evergreen/deciduous forest. The remaining original tree pixels (evergreens) are run through a separate unsupervised classification and generate groups of evergreen forest or woodland and some mixed evergreen/deciduous forest or woodland. Both classifications used all six bands as input in generating 100 groups. The tree map labels were then processed with digital soils data as described in the previous panel. Landsat Scene Coverage for Iowa Tree Separation Model Function Spectral Profiler results showing May signatures Red Ceda r Pine s Decid Fores t The Grass class used a late Spring date when available. All six bands were used in the unsupervised classification, resulting in 100 groups. The majority of map labels were cool grass with about 25% being warm grass. Upland shrubland and grass with sparse trees were also found in this portion. Using a Fall date, usually October, worked to differentiate the two Artificial map labels. It was beneficial to have some contrast between the growing vegetation and hard surface material (pavement, gravel). The Spectral Profiler tool was used to get signatures for comparison between residential and commercial areas. Based on that signature data, the three best bands for separation were subset and used in an unsupervised classification. The resulting 100 groups were assigned map labels using digital ortho photo quads (DOQs) as a background. Ground Referenced Data Sources Vegetation Survey Various ISU Research Projects DNR State Park Ecoplans District Forester Stands TNC Prairie Survey DNR Prairie Survey DNR Funded County and State Park Surveys The key to generating the most accurate land cover from satellite imagery is having enough ground referenced data for each map label. When Iowa GAP began in 1997, a vegetation survey was sent to all 99 County Conservation Board offices in the state and to DNR parks and wildlife management areas. The survey was a form and asked for textual descriptions of uniform areas of vegetation based on the Vegetation Alliances Index. The uniform areas were to be outlined on Farm Service Agency (FSA) aerial section photos. When received at ISU, the photos were scanned and registered to digital topographic quads or DOQs. The outlined vegetation polygons were digitized onscreen in ArcView and a shapefile was created with attributes from the survey form. Unless otherwise noted, map labels were assigned using the survey shapefile, other ground referenced data sources and false CIR display of imagery. Screen shots of Tree classification for overlay comparison on DOQ and Spring satellite image. Tree pixels are being swiped across the underlying image. Crop Cool Grass Crop Red Cedar Forest After the Phase 1 land cover data were created, it was realized that the Barren class included map labels besides actual barren sites. Post-processing of the land cover focused on the Phase 1 Barren class and was done in two passes. Pass 1 - It was decided to process only 1/2 section and larger clumps; this translated to clumps >= 360 pixels. There were 141 clumps created in Imagine and individually examined using false CIR imagery display, DOQs, a mines location shapefile and clump shape. Roughly 60% of the clumps were recoded to cropland, 7% stayed Barren, 17% went to a newly created map label called barren/mixed vegetation and the rest of the clumps were distributed over wetlands, water and cool grass. The new map label functions as a flag to focus future land cover classification. The clumps contain a possible mix of barren, grass, trees and cropland. This map label will not be further analyzed in the current Iowa Gap Analysis Program effort. Pass 2 - Clumps in the size range of 1/4 - 1/2 section (180 - 359 pixels) were automatically recoded to cropland. This decision was based on the analyst’s examination of smaller clumps surrounding the target clumps in Pass 1. It was felt that the accuracy of the Barren class would improve after this recoding. Issues, Concerns and Recommendations There were several issues that the Iowa GAP team felt were affecting not only the final land cover accuracy but the types of vegetation found in the map labels themselves. The extensive human disturbance of the Iowa landscape has resulted in a highly fragmented land cover pattern. The agricultural based economy has produced cropfields bounded by hedgerows, farmsteads protected by windbreaks, pastures with forest islands and many small towns. The use of 30 meter resolution pixels to classify such a landscape means many pixels represent the reflectance of more than one distinct land cover type. This will ultimately affect the land cover accuracy. Most of the satellite imagery used in the classification was selected and purchased by people with little knowledge of the specific timing of phenological characteristics of Iowa vegetation. In order to reliably discriminate between some vegetation types, two dates of imagery must be used that show different signatures for the types. Generally, the supplied dates worked reasonably well for separating evergreen and deciduous trees and warm and cool season grasses. It is possible a mid to late October date would discriminate between some upland deciduous forest alliances that were aggregated in this classification. In order to take advantage of a nationwide purchase of imagery at no cost to our state program, Iowa GAP had to use satellite imagery mainly flown in 1992. Our classification effort and ground referenced data collection began in late 1997 and continued for three years. The five year minimum difference has likely caused some misclassification and definitely created some confusing situations during the entire process. This time lag will extend to nine years if field based accuracy assessment is done in 2001. It is known that grasslands, croplands and urban areas have changed significantly in the last nine years. Because of the processing time involved in creating statewide land cover, current satellite imagery must be used to generate a timely, useful data set. Map Labels List Phase 1 Credits Jim Giglierano, IA DNR John McKinness, IA DNR Phase 2 Credits Robin McNeely, ISU Bret Giesler, ISU Mark Alexander, IA DNR Use Special Processing Techniques to Improve Classificati on The Des Moines Lobe region of Iowa has many small wetlands which are important habitat to a variety of vertebrate species being modeled by GAP. The 30 meter resolution of the TM imagery was too coarse to find many of the small wetlands so National Wetlands Inventory (NWI) data were used as an overlay. Digital NWI data exists for Iowa and was used to directly write wetland areas onto the Phase 1 land cover. Todd Bishop, IA DNR, wrote an aggregation model in ArcInfo that processes NWI arc and polygon data. Frequently, wetlands exist in complexes of several types and each type may be too small to retain its existence when converted to raster format. The aggregation model used certain relationship rules to dissolve polygon boundaries and assign the deepest water regime label to the resulting polygon. The model was originally written to estimate duck breeding numbers in wetlands and aggregated some wetland types that GAP needed to delineate. This necessitated a two-step process to apply NWI data to the land cover; model results were first overlaid on the land cover and then forested, shrubland and saturated wetlands from the original NWI data were overlaid. NWI Wetland Aggregation An early Spring date in March or April works best for delineating evergreen map labels. However, the signature of red cedar at this time is close to the signature of a mixed content pixel that includes a water/bank vegetation interface. The digital soils data were used as a mask to select evergreen or red cedar pixels occurring in the floodplain and recoding them to deciduous forest or woodland. There is a chance that pine plantations or red cedar may occur in the floodplain and these will be erroneously changed, but it was felt that those instances will be rare. Floodplain Soils Mask Tree Separation Model Once all twelve scenes were processed, they were merged together into one statewide image. The overlay order was determined by the classification analyst based upon overall scene content quality. Metadata was written. Poster created by Robin McNeely, [email protected] and edited by LIO. FSA aerial photo with outlined polygons Phase 2 land cover after post- processing Phase 2 land cover area percentages 59.88% Cropla nd 21.21% Cool Grass 4.85% Warm Grass 5.78% Deciduous Forest 59.82% Cropla nd 30.56% Grass 6.94% Trees 0.76% Water 1.05% Artifi cial 0.86% Barre n Merge 12 Scenes into Statewide Image Deciduous Evergreen Mixed Cutoff Value at 90 Phase 1 land cover area percentag es Temporarily Flooded Forest Upland Deciduous Forest Red Cedar Forest White Pine Forest Robin McNeely, Department of Animal Ecology, Iowa State Robin McNeely, Department of Animal Ecology, Iowa State University, Ames, Iowa University, Ames, Iowa

Transcript of Iowa GAP is scheduled for completion in December, 2001. Check the state and national GAP webpages...

Page 1: Iowa GAP is scheduled for completion in December, 2001. Check the state and national GAP webpages for updates and links to acquire the reports and GIS.

Iowa GAP is scheduled for completion in December, 2001. Check the state and national GAP webpages for updates and links to acquire the reports and GIS data.

General Information www.iowagap.iastate.edu

Data Download cctr.ait.iastate.edu

IMS Site siberia.gis.iastate.edu/iagapims/viewer.htm

IOWA’S LAND COVER MAP CREATION PROCESS

IOWA’S LAND COVER MAP CREATION PROCESS

Printed 11/21/01

Final Products and Data Access:

The Iowa Gap Analysis Program will publish a final land cover report explaining details of the classification process. In conjunction with this report, an Atlas of Iowa Land Cover will also be published.

The documents will be available on CD and on the Iowa GAP website in Adobe Acrobat .pdf format. The GIS datasets used to create the land cover and the land cover data itself will be available for downloading via ftp service. This same GIS data will be available for interactive viewing and querying over the Internet using ESRI’s Internet Map Server software.

Introduction and Background

The effort to comprehensively map the land cover of Iowa from satellite imagery for the Gap Analysis Program (GAP) began in 1996. The Geological Survey Bureau of the Iowa Department of Natural Resources (DNR) processed Landsat 5 imagery provided by the Multi-Resolution Land Characteristics Consortium (MRLC) into six land cover classes (Trees, Grass, Cropland, Artificial, Barren and Water) plus a Cloud class. This product is referred to as Phase 1 land cover; the processing occurred in Iowa City and was completed in late 1998. The satellite imagery was processed scene by scene using an unsupervised classification in EASI/PACE imaging software. The MRLC provided pre-processed satellite images that contained 240 clusters along with the seven original TM bands of data for two dates per scene.

In some scenes the date chosen was not optimal for showing vegetation differences or there was excessive cloud cover. Therefore, additional dates for 11 out of 12 scenes were purchased with funds from several sources. This additional imagery, plus the original MRLC purchase, was used in the Phase 2 mapping process. This process further classified the six classes into map units compatible with the National Vegetation Classification System (NVCS).

Creation of the Vegetation Alliances Index and Map Labels List for Iowa

National GAP has realized that no one mapping strategy will work for all states. They provided guidelines for land cover products but the methods to achieve those products were up to each state to develop. Phase 2 of Iowa GAP used the image classification protocol developed by the Upper Midwest Gap Program as a starting point. Their methods were tested with our data and modified to fit our needs and capabilities. This poster is a summary of our methods along with a discussion of certain issues unique to the Iowa landscape.

The Iowa Gap Analysis Program adopted “An Alliance Level Classification of the Vegetation of the Midwestern United States” as our basic classification framework. It was published in 1997 by The Nature Conservancy in a cooperative effort with National GAP to provide a vegetation classification system that complies with the Federal Geographic Data Committee’s NVCS. Naturally occurring vegetation is described hierarchically with physiognomic criteria at the higher levels and floristic criteria at the lower levels.

A team of six Iowa botanists reviewed the Alliance Level Classification and modified it to more accurately reflect the existing vegetation alliances that they have encountered in the field. The result was a “Preliminary Index of Natural Vegetation Alliances for Iowa” , printed in 1998. Twenty alliances listed in the Alliance Level Classification that did not include Iowa in the range were added and eight alliances were created. Conversely, alliances listed in the classification as occurring in Iowa but which team members believed not to occur here, were removed from the Vegetation Alliances Index.

Special Processing Techniques General Classification Process Post Processing Techniques

To improve the utility and accuracy of the final land cover product, three processing techniques were used to enhance the general unsupervised classification.

Iowa Gap Land Cover Cooperators: Before After

Example area showing NWI Aggregation Model results

Robin McNeely created a model in ERDAS Imagine to split the Phase 1 Tree class into evergreen or deciduous subclasses based on a cutoff value in Band 5 of the earliest Spring date. The model takes the tree masked Spring date, (March or April work best), and assigns a value of one to all pixels with a Band 5 value of 90 or greater and the rest are assigned zero. The cutoff value varies by five in either direction based on the imagery date and quality. The Spectral Profiler was used to get sample band values in known sites of red cedar, pine and deciduous forest and a cutoff value was selected that is closer to the signature of deciduous forest.

Use Phase 1 as Base Data

Compile Ground Referenced Data from Variety of

Sources

Enhance Three Phase 1 Classes Using the General Classification Procedure

Create Iowa Vegetation Alliances Index and Map

Labels List

Write Metadata for Statewide Land

Cover

Distribute Land Cover Data via CD, ftp and Internet Map

Service

Post-process Barren Class

This index in turn provided a framework for a “Working List of Land Cover Map Labels for Iowa Gap Analysis”, finalized in 1999. Essentially, this is the list of descriptions for land cover that correspond to the map labels. The Vegetation Alliances Index was condensed and modified to reflect the limitations of Landsat TM imagery to discriminate certain spectrally similar alliances. Non-vegetation classes (Barren, Cropland, Artificial and Water) were added to accommodate actual land cover distinguishable by the satellite imagery. Unlike the Alliance Level Classification, which was created by botanists before the land cover mapping process began, the Map Labels List evolved over a year of classifying land cover. When alliances could not be reliably separated by the available dates of satellite imagery, they were aggregated to a map label. For example, the six upland oak alliances could not be spectrally separated from each other or from the maple-basswood alliance, so they all were grouped into the upland deciduous forest map label.

The Phase 1 land cover was used as a base for creating the Phase 2 data. The Tree, Grass and Artificial classes were individually processed into more detailed map labels. The first step was to isolate Cloud pixels, if they existed, from Phase 1. These were used to mask the Spring date of imagery and an unsupervised classification with 50 classes was run. All available ground referenced data were used to assign map labels. The next step was to overlay the NWI data as described in the previous panel; these pixels were not modified after this step.

The Tree class was processed by applying the Tree Separation Model as described in the previous panel. The result of the model is a mask that is used with the Spring imagery in an unsupervised classification and produces almost entirely deciduous forest with some mixed evergreen/deciduous forest. The remaining original tree pixels (evergreens) are run through a separate unsupervised classification and generate groups of evergreen forest or woodland and some mixed evergreen/deciduous forest or woodland. Both classifications used all six bands as input in generating 100 groups. The tree map labels were then processed with digital soils data as described in the previous panel.

Landsat Scene Coverage for Iowa

Tree Separation Model Function

Spectral Profiler results showing May signatures

Red Cedar

PinesDecidForest

The Grass class used a late Spring date when available. All six bands were used in the unsupervised classification, resulting in 100 groups. The majority of map labels were cool grass with about 25% being warm grass. Upland shrubland and grass with sparse trees were also found in this portion. Using a Fall date, usually October, worked to differentiate the two Artificial map labels. It was beneficial to have some contrast between the growing vegetation and hard surface material (pavement, gravel). The Spectral Profiler tool was used to get signatures for comparison between residential and commercial areas. Based on that signature data, the three best bands for separation were subset and used in an unsupervised classification. The resulting 100 groups were assigned map labels using digital ortho photo quads (DOQs) as a background.

Ground Referenced Data Sources•Vegetation Survey

•Various ISU Research Projects

•DNR State Park Ecoplans

•District Forester Stands

•TNC Prairie Survey

•DNR Prairie Survey

•DNR Funded County and State Park Surveys

The key to generating the most accurate land cover from satellite imagery is having enough ground referenced data for each map label. When Iowa GAP began in 1997, a vegetation survey was sent to all 99 County Conservation Board offices in the state and to DNR parks and wildlife management areas. The survey was a form and asked for textual descriptions of uniform areas of vegetation based on the Vegetation Alliances Index. The uniform areas were to be outlined on Farm Service Agency (FSA) aerial section photos. When received at ISU, the photos were scanned and registered to digital topographic quads or DOQs. The outlined vegetation polygons were digitized onscreen in ArcView and a shapefile was created with attributes from the survey form. Unless otherwise noted, map labels were assigned using the survey shapefile, other ground referenced data sources and false CIR display of imagery.

Screen shots of Tree classification for overlay comparison on DOQ and Spring satellite image.

Tree pixels are being swiped across the underlying image.

CropCrop

Cool GrassCool Grass

CropCrop

Red Cedar Forest

Red Cedar Forest

After the Phase 1 land cover data were created, it was realized that the Barren class included map labels besides actual barren sites. Post-processing of the land cover focused on the Phase 1 Barren class and was done in two passes.

Pass 1 - It was decided to process only 1/2 section and larger clumps; this translated to clumps >= 360 pixels. There were 141 clumps created in Imagine and individually examined using false CIR imagery display, DOQs, a mines location shapefile and clump shape. Roughly 60% of the clumps were recoded to cropland, 7% stayed Barren, 17% went to a newly created map label called barren/mixed vegetation and the rest of the clumps were distributed over wetlands, water and cool grass. The new map label functions as a flag to focus future land cover classification. The clumps contain a possible mix of barren, grass, trees and cropland. This map label will not be further analyzed in the current Iowa Gap Analysis Program effort.

Pass 2 - Clumps in the size range of 1/4 - 1/2 section (180 - 359 pixels) were automatically recoded to cropland. This decision was based on the analyst’s examination of smaller clumps surrounding the target clumps in Pass 1. It was felt that the accuracy of the Barren class would improve after this recoding.

Issues, Concerns and Recommendations

There were several issues that the Iowa GAP team felt were affecting not only the final land cover accuracy but the types of vegetation found in the map labels themselves.

The extensive human disturbance of the Iowa landscape has resulted in a highly fragmented land cover pattern. The agricultural based economy has produced cropfields bounded by hedgerows, farmsteads protected by windbreaks, pastures with forest islands and many small towns. The use of 30 meter resolution pixels to classify such a landscape means many pixels represent the reflectance of more than one distinct land cover type. This will ultimately affect the land cover accuracy.

Most of the satellite imagery used in the classification was selected and purchased by people with little knowledge of the specific timing of phenological characteristics of Iowa vegetation. In order to reliably discriminate between some vegetation types, two dates of imagery must be used that show different signatures for the types. Generally, the supplied dates worked reasonably well for separating evergreen and deciduous trees and warm and cool season grasses. It is possible a mid to late October date would discriminate between some upland deciduous forest alliances that were aggregated in this classification.

In order to take advantage of a nationwide purchase of imagery at no cost to our state program, Iowa GAP had to use satellite imagery mainly flown in 1992. Our classification effort and ground referenced data collection began in late 1997 and continued for three years. The five year minimum difference has likely caused some misclassification and definitely created some confusing situations during the entire process. This time lag will extend to nine years if field based accuracy assessment is done in 2001. It is known that grasslands, croplands and urban areas have changed significantly in the last nine years. Because of the processing time involved in creating statewide land cover, current satellite imagery must be used to generate a timely, useful data set.

Map Labels List

Phase 1 Credits

Jim Giglierano, IA DNRJohn McKinness, IA DNR

Phase 2 Credits

Robin McNeely, ISUBret Giesler, ISUMark Alexander, IA DNR

Use Special Processing

Techniques to Improve

Classification

The Des Moines Lobe region of Iowa has many small wetlands which are important habitat to a variety of vertebrate species being modeled by GAP. The 30 meter resolution of the TM imagery was too coarse to find many of the small wetlands so National Wetlands Inventory (NWI) data were used as an overlay. Digital NWI data exists for Iowa and was used to directly write wetland areas onto the Phase 1 land cover. Todd Bishop, IA DNR, wrote an aggregation model in ArcInfo that processes NWI arc and polygon data. Frequently, wetlands exist in complexes of several types and each type may be too small to retain its existence when converted to raster format. The aggregation model used certain relationship rules to dissolve polygon boundaries and assign the deepest water regime label to the resulting polygon. The model was originally written to estimate duck breeding numbers in wetlands and aggregated some wetland types that GAP needed to delineate. This necessitated a two-step process to apply NWI data to the land cover; model results were first overlaid on the land cover and then forested, shrubland and saturated wetlands from the original NWI data were overlaid.

NWI Wetland Aggregation

An early Spring date in March or April works best for delineating evergreen map labels. However, the signature of red cedar at this time is close to the signature of a mixed content pixel that includes a water/bank vegetation interface. The digital soils data were used as a mask to select evergreen or red cedar pixels occurring in the floodplain and recoding them to deciduous forest or woodland. There is a chance that pine plantations or red cedar may occur in the floodplain and these will be erroneously changed, but it was felt that those instances will be rare.

Floodplain Soils Mask

Tree Separation Model

Once all twelve scenes were processed, they were merged together into one statewide image. The overlay order was determined by the classification analyst based upon overall scene content quality. Metadata was written.

Poster created by Robin McNeely, [email protected] and edited by LIO.

FSA aerial photo with outlined polygons

Phase 2 land cover after post-processing

Phase 2 land cover area percentages

59.88%Cropland

21.21%Cool Grass

4.85%Warm Grass

5.78%Deciduous Forest

59.82%Cropland

30.56%Grass

6.94%Trees

0.76%Water 1.05%

Artificial

0.86%Barren

Merge 12 Scenes into Statewide Image

DeciduousDeciduous EvergreenEvergreen

MixedMixed

Cutoff Value at 90

Phase 1 land cover

area percentages

Temporarily Flooded Forest Upland Deciduous Forest Red Cedar Forest White Pine Forest

Robin McNeely, Department of Animal Ecology, Iowa State University, Ames, Robin McNeely, Department of Animal Ecology, Iowa State University, Ames, IowaIowa