Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
CHAPTER 4 Enhancements to Ecological Initiative Geospatial Tool The assessment indicated the need to refine the Ecological Initiative tool to extract a finer- resolution land cover, including impervious surfaces, in the area identified as urban low intensity and urban high intensity in the existing classification. The goal of the tool refinement is to identify the softscape amidst the urban landscape, to map green corridors and nodes that, while they have likely been impacted by urbanization, still contribute to the ecological health of the region. The original tool contained data developed mainly for planning outside of core urban areas, whereas the new data are focused for use inside of core urban areas. As use of the Ecological Initiative tool expands to partner agencies and beyond, a broader range of user needs is emerging. Partners are interested in using the data for green infrastructure planning and site review activities, in addition to regional scale planning. Two pilot project corridors were identified as the location to test the development and applicability of finer- resolution land use and land cover (LULC) information. Two pilot study areas were selected to develop a fine resolution LULC data set. These study areas were I-44 in southwestern Saint Louis County, Missouri, and IL-158 in Central Saint Clair County, IL (Figure 4.1). These study areas circumscribed a diverse mix of cover types and urban densities and thus facilitated the evaluation of the usefulness of the improved spatial resolution LULC in assessing potential environmental impacts. The I-44 study area was approximately 25 square miles in area, and included portions of the municipalities of Crestwood, Fenton, Kirkwood, Peerless, Sunset Hills, and Valley Park. The I-44 study area had a diverse mixture of cover types including forested state park lands, industrial, high density urban, and low density urban. The IL-158 study area was approximately 48 square miles in area and included portions of the municipalities of Millstadt and Columbia within its boundary. Cover types in the IL-158 study area were mostly agricultural, with some low density urban. 20
Figure 4.1. Study areas I-44 in Southern Saint Louis County, Missouri, and IL-158 in Central Saint Clair County, Illinois. Data Enhancements Spring 2012 Lidar To obtain a highly accurate depiction of landscape terrain, as well as vegetation and building structure height to aid in LULC mapping, a digital elevation model (DEM) and digital surface model (DSM) were developed for each study area from raw lidar LAS points at 1-meter spatial resolution. Lidar is an optical sensor that measures distance through multiple laser pulses, thus allowing a user to determine the elevation of the surface and the height of objects on the ground by comparing multiple pulse returns. For the IL-158 study area, lidar data were acquired from March 26 to April 2, 2012, for the Illinois State Geological Survey (ISGS) and have a fundamental vertical accuracy (FVA) of 12.5 centimeters. For the I-44 study area, lidar was collected January 29 to 30 and February 1, 2012, by Surdex Corporation for the United States Army Corps of Engineers (USACE) and has a FVA of 18.5 centimeters. All lidar data were downloaded from the state of Missouri lidar data clearinghouse hosted by Washington University in Saint Louis (ftp://lidar.wustl.edu/). QT modeler software was used to process LAS files for both study areas. The DEM (Figure 4.2) was generated using a grid sample of 1 meter. Adaptive triangulation was used to interpolate elevation in areas where points were less dense, with a maximum distance between points and maximum triangle side of 1000 feet for surface interpolation. The mean z (height) value for each point was calculated for points with class values of 2 (ground) and 8 (model key points). The DSM (Figure 4.3) was generated using a grid sample of 1 meter and adaptive triangulation with a maximum point distance and maximum triangle side of 1000 feet. The maximum z value for first returns was used to obtain the tallest possible point of vegetation or structures. 21
Figure 4.2. Digital elevation model (DEM) from lidar LAS points (MoRAP 2014). Figure 4.3. Digital surface model (DSM) from lidar LAS points (MoRAP 2014). Vegetation and structure height (Figure 4.4) was created by subtracting the DSM, which is the highest point on the surface, from the DEM, which is the lowest point on the surface. The vegetation/structure height data set provides vegetation and structure (building, bridge, utility pole, etc.) height of features on the surface within the study area. Height information of cover types is valuable to the LULC modeling approach. 22
Figure 4.4. Vegetation and structure height surface developed by subtracting lidar DEM from DSM (MoRAP 2014). Additional lidar-derived products such as slope, aspect, solar insolation, and land position were created based on the DEM for each study area. These products provide surface information explaining various landscape variables. LULC classes were defined via a supervised classification using these and other variables (see below). 2012 Aerial Photography Aerial photography from spring and summer 2012 was used to aid in LULC mapping for both study areas. The use of multi-temporal imagery to classify LULC allows for improved classification of forest, grass, and agricultural classes by differentiating spectral changes in vegetation senescence and canopy density between seasons. Leaf-off (spring) imagery reveals structures hidden below the canopy of summer (leaf-on) imagery, which would otherwise be missed in mapping (Figures 4.5 and 4.6). It is uncommon to have high-resolution multi-temporal 23
imagery for a given location within the same calendar year. The use of this imagery allowed for an accurate classification of surface cover conditions as of summer 2012. Spring 6-inch, leaf-off, 4-band, false color-infrared imagery (Figure 4.5) was sampled to 1-meter spatial resolution to match the spatial resolution of the summer United States Department of Agriculture (USDA) National Agriculture Imagery Program (NAIP) imagery. The addition of the infrared band helped especially in separating grassland from cropland. Summer 1- meter, leaf-on, 3-band true color NAIP imagery (Figure 4.6) was downloaded for both study areas from the USDA NRCS Geospatial Data Gateway (datagateway.nrcs.usda.gov/). Figure 4.5. Spring 2012, 1-meter, leaf-off, 4- band infrared imagery aided in mapping of grass and crops and well as structures hidden below the forest canopy (MoRAP 2014). Figure 4.6. Summer 2012, 1-meter, leaf-on, 3-band imagery used to map LULC (MoRAP 2014). Metropolitan Sewer District of Saint Louis Impervious Surface Data Metropolitan Sewer District of Saint Louis (MSD) impervious surface geographic information system (GIS) polygon data, created in 2009, was supplied by East-West Gateway to aid in LULC mapping of impervious surfaces. The MSD impervious surface data include polygons of various impervious surface structures at a fine resolution, including the footprints of houses, sidewalks, sewer drains, and parking lots (Figure 4.7). The use of this data improved mapping of roads, parking lots, sidewalks, and building footprints. MSD only covers a small geographic area of the Saint Louis region, including Saint Louis City and Saint Louis Counties in Missouri. As a result, MSD data were only applicable for the I-44 study area. 24
Figure 4.7. MSD impervious surface polygons aided in mapping roads, building footprints, and other impervious surfaces for the I-44 study area (MoRAP 2014). Results of Enhanced Tool Applied to Pilot Areas LULC Classification A two-date multi-temporal aerial photography data set, together with lidar-based DEM-derived variables, was used to generate a fine resolution (1 meter) LULC data set, current as of June 2012. A supervised classification and regression tree (CART) modeling approach was utilized to classify six LULC classes based on 600 spatially explicit photo-interpreted ground samples for each study area. Samples A stratified random selection of 100 samples per target LULC class (water, urban/impervious, barren, forest, grass, and crop) was generated based on the 15-class, 30-meter resolution 2008â 2009 MoRAP LULC, resulting in a total of 600 training samples per study area (Figure 4.8). The 25
training sample points were used to inform the classifier and produce a six-class LULC map. The original 15 LULC classes were aggregated to fit within the six mapped classes (Table 4.1). The points were manually inspected against the 2012 aerial photography to verify that each point represented the cover type it was assigned to. In the event that a point did not represent the cover type it was assigned to, the point was deleted. Figure 4.8. LULC training samples within the 158 study area (MoRAP 2014). 26
Table 4.1. Crosswalk of 15-Class LULC Used to Generate Training Samples and Six LULC Classes to be Mapped for this Project 6 Class LULC to be mapped for 2012 2008-2009 15 Class LULC Water Open Water Urban/Impervious Impervious High Density Urban Low Density Urban Barren/Sparsely Vegetated Barren or Sparsely Vegetation Forest Deciduous Forest Evergreen Forest Mixed Forest Deciduous Woody/Herbaceous Evergreen Woody/Herbaceous Woody-Dominated Wetland Grass Grass Crop Crop Modeling Six classes were mapped at 1-meter spatial resolution using a supervised CART modeling approach with boosted regression trees with See5 statistical software. The classes that were mapped are: 1. Water 2. Urban/Impervious 3. Barren/Sparsely Vegetated 4. Forest 5. Grass 6. Crop Approximately 600 spatially explicit sample points were used to map LULC based on the statistical relationship of 14 spectral and terrestrial landscape-based variables, including 2012 aerial photography (seven total bands from two dates of imagery), vegetation height, digital elevation model, digital surface model, slope, aspect, solar insolation, and landscape position. The model produced a six-class, 1-meter raster LULC for each study area (Figure 4.9). 27
Figure 4.9. Six-class, 1-meter supervised LULC in the158 study area. (MoRAP 2014) Objects In order to create a vector-based LULC classification, which is valuable for querying specific cover types and improving the resolution and detail of LULC classification, image objects were generated. Image objects are polygons that circumscribe statistically homogeneous features on the landscape based on input data sets using eCognition software. Image objects for this project were generated based on lidar-derived vegetation/structure height and 2012 leaf-on and leaf-off imagery. The modeled six-class, 1-meter raster LULC was used to assign objects LULC values by summarizing the majority of the LULC pixels within each object (e.g., the mode land cover value from pixel centroids was assigned to each image object polygon; see Figure 4.10). 28
Figure 4.10. Objects (black) were populated with LULC by summarizing the majority of LULC pixels within a polygon. (MoRAP 2014) Manual Inspection After image objects were assigned LULC values based on raster classification, they were visually inspected for classification errors. The polygons were compared against vegetation height and the 2012 leaf-off and leaf-on aerial photography data sets. When errors were found, the polygon was assigned to the appropriate class. Errors were inspected and manually edited at a scale of 1:1500 to 1:2000. Common errors included misclassification of shadows around buildings and at the edge of forests as water, grass as crop, and buildings as forest (Figures 4.11 and 4.12). Manually inspecting objects for errors was the most time-consuming aspect of the project. One could reduce the time spent on error correction by editing at a coarser scale (i.e., 1:5000) and/or increasing the minimum size of image objects so that there are fewer objects and details to correct. Overall cross-validated accuracy before editing for the I-44 study area was 79.5%, and was 81.1% for the IL-158 study area. 29
Figure 4.11. Building shadows mapped as water was a common error fixed by manual inspection (MoRAP 2014). Figure 4.12. The 2012 leaf-off aerial photography was used to identify and correct mistakes, such as building shadows being mapped as water (MoRAP 2014). Post-processing/Filtering After objects were inspected for errors, polygons less than or equal to 4 square meters were eliminated and absorbed by the adjacent polygon sharing the longest border to reduce the number of small polygons within the data set. Objects were then dissolved to aggregate adjacent polygons with the same LULC class to clean up and reduce the number of polygons in the data set (Figures 4.13 and 4.14). Figure 4.13. Undissolved image objects in I-44 study area (MoRAP 2014). Figure 4.14. Dissolved objects in I-44 study area aggregates adjacent objects with same LULC type into a single polygon. Reduced number of polygons from 546,931 to 44,276 in I-44 study area (MoRAP 2014). 30
Success and Challenges The 2012 1-meter, six-class LULC improved upon the detail of LULC mapping that previously existed in the Saint Louis region by using higher spatial resolution imagery and lidar data, yet reducing the thematic resolution. The goal of improving the mapping of urban vegetation and riparian corridors within the metro area was achieved (Figures 4.15 and 4.16). The comparison in overall LULC composition between the 2008â2009 30-meter LULC and 2012 1-meter LULC shows that, in a more diverse and urbanized landscape, the mapped LULC composition would be significantly different than that of a more rural, natural, and homogeneous landscape. In the more urbanized I-44 study area, mapped LULC composition changed dramatically, with the 2012 LULC mapping half as much urban/impervious and twice as much grass and crop than the 2008- 2009 LULC (Figure 4.17). In the less urban and more agricultural 158 study area, the 2012 LULC mapped two-thirds-percent less urban, 10% more crop, 8% more forest, and 9% less grass than the 2009-09 LULC (Figure 4.18). The use of higher resolution imagery in mapping the 2012 LULC allowed for a more detailed depiction of all cover types, which can provide a more accurate analysis of the impact of potential development when incorporated into the environmental impact model. Figure 4.15. 2008-2009 MoRAP 30 meter LULC in I-44 study area (MoRAP 2014). 31
Figure 4.16. 2012 MoRAP 1-meter LULC in I-44 study area (same extent as Figure 4.15) shows improved detail in urban vegetation and impervious cover mapping (MoRAP 2014). Acceptability of the Data Refinements Upon completion of the work, the C40B2 team set about to determine the regulatory acceptability of the refined land use/land cover classification. The results of the data refinement were presented to members of the steering committee in February 2014. The background behind the data and the advanced techniques required to produce the data refinements were discussed. Participants contributed insights on how the data can be applied to work within their agencies. The methodology behind the 2012 1-meter LULC classification was widely accepted as a certified approach to refining the 2008â2009 30-meter land cover. It was clear to members that, while the 2008â2009 30-meter land cover proved useful for earlier work, in the 2012 1-meter LULC classification, the fine-resolution six LULC cover class product gives the end user more reliability when using the data for planning activities. The 1-meter LULC provides a much more accurate picture of conditions on the ground. The new LULC provides an improved picture of possible impacts a transportation project can have on the environment, particularly in urban settings, and would be useful to DOTs by providing better site-specific information when corridor studies were conducted and projects occurred in urban locations. The data provide a much better visual depiction of information contained in national data sets, as well as conditions on the ground that are not contained in national-level data. 32
Figure 4.17. Comparison of LULC composition mapped of 2008â09 30-meter LULC and 2012 1-meter LULC in I-44 study area. Improved spatial resolution resulted in significant changes in LULC composition mapped. Figure 4.18. Comparison of LULC composition mapped of 2008â09 30-meter LULC and 2012 1-meter LULC in 158 study areas. Fewer significant changes in LULC composition are mapped due to a more homogenous landscaped of mostly crop. 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.0 Water Urban/Impervious Barren/Sparsely Vegetated Forest Grass Crop Sq . M ile s LULC Cover Classes Comparison of LULC Area Mapped: I-44 Study Area 2008-09 30 m vs. 2012 1 m 08-09 30 m LULC 2012 1 m LULC 0.0 5.0 10.0 15.0 20.0 25.0 30.0 Water Urban/Impervious Barren/Sparsely Vegetated Forest Grass Crop Sq . M ile s LULC Cover Classes Comparison of LULC Area Mapped: 158 Study Area 2008-09 30 m vs. 2012 1 m 08-09 30 m LULC 2012 1 m LULC 33
Often large-scale national data sets do not reflect environmental conditions on the ground in urban areas. That was the case with the 2008â2009 LULC, which was based on input data sets (e.g. 30-meter satellite data) that are used in the development of national land cover data sets. These data essentially consider land that has been disturbed as urbanized. Urban planners see diversity of natural elements in urban land cover, despite development, and seek to take advantage of the natural elements present within the urban landscape. The refined data enhance the Ecological Initiative tool and provide urban planners with a mechanism to proactively plan for ecological improvements in the region. Major efforts are underway in the region to expand greenways and green infrastructure. These efforts are examples of planning efforts that seek to enhance natural elements in the region for broad environmental, social, and economic benefits. The refined data could play a key role in identifying and enhancing key green connections and nodes. The data communicate to the user the location and extent of impervious cover. This information could play a role in stormwater Phase II planning. Currently state DOTs are working with state environmental agencies on permitting requirements in Municipal Separate Storm Sewer Systems (MS4) communities. The refined data could prove useful in the planning for controlling and managing stormwater runoff. Although the 2012 1-meter LULC was only completed for two study areas, it is clear that this refined data could have a lager impact if completed for the entire region. From transportation planning, to watershed conservation efforts, there is a sense that the 2012 1-meter LULC would be a valuable resource not only for state DOTs, but for both local and regional government, environmental organizations, and other non-profits engaged in conservation and preservation efforts in the region. Further discussions are needed to determine how this information can be produced, hosted, and served for use within the region. 34
