Based on the concept of accumulating overland flow energy, an erosion potential model (nLS+) was calibrated and validated for five military installations (Fort Benning, Fort Hood, Fort Riley, Schofield Barracks, and Pohakuloa Training Area [PTA]). By integrating data from digital elevation models (DEMs) and landuse/landcover (LULC) assessments in a geographic information system (GIS) environment, the nLS+ model determines where surface water runoff transitions from overland sheet flow to concentrated flow and, as a result, where the potential for soil erosion and gully formation is highest.

Three quantitative performance objectives were identified for this project: (1) identify the critical threshold for accumulated nLS+ values for each study installation; (2) determine whether a single critical threshold for accumulated nLS+ values is adequate for all installation study sites; and (3) use the nLS+ model in a predictive mode to forecast areas where gullies are likely to form in response to future military training events.

The nLS+ model is designed as a series of four sub-models to predict the current location of gullies or forecast the formation of new gullies in response to a training event. Data sets required for gully prediction include LULC and DEM inputs. From these data layers, the Manning’s n, slope, profile curvature, and other required intermediate data products are computed. In forecast mode, input data include a filled (i.e., depressionless or pit-removed) DEM and Manning’s n grid (output from the model in prediction mode) and a form of vehicle tracking data from Global Positioning System (GPS) devices that includes, at minimum, vehicle coordinates and velocity. Model LULC and DEM inputs can be obtained from nationally available geographic datasets such as the National Landcover Database (NLCD) and National Elevation Dataset (NED) to minimize data acquisition and preparation times.

At the NED DEM resolution, 131 of 222 total validation gullies (59%) were predicted to within 10 meters (m) of their actual location. This percentage increases to 79% and 89% for distances within 20 and 30 m, respectively. Mean distance to the correct location ranged from a low of 4.93 m (Keamuku Parcel at PTA) to a high of 14.73 m (Fort Hood). Percentage correct predictions (<= 10 m) for individual installations ranged from a high of 85% (Keamuku Parcel) to a low of 50% (Fort Benning), with the total range in distances varying from 20 m (Keamuku Parcel) to nearly 96 m (Fort Hood). Only the predictions for Keamuku Parcel met the success criteria of having more than 80% of gully locations correctly identified using installation-specific accumulated nLS+ thresholds within the required distance precision. Overall model accuracy was 86%, though commission error for the predicted gully class was nearly 100%. Although the model suffered from extreme overprediction of gully locations, the total land area identified as likely erosion sites ranged from a low of 496 hectares (ha) (Kahuku Range) to a high of 16,862 ha (Fort Hood), with the total installation area susceptible to predicted gully erosion falling between 12% (Fort Riley) and 21% (Keamuku Parcel).

The nLS+ model was developed with the ArcGIS 10.0 (Esri, Redlands, CA) suite of GIS software applications routinely utilized across Department of Defense (DoD) agencies and installations. The model is packaged as a “toolbox” with multiple sequenced tools than can be downloaded and used on a local computer workstation running ArcGIS version 10.0. The toolbox and associated modeling tools also includes an integrated help system to explain the underlying processes, data requirements, and recommendations for default settings or values as they apply to execution of the model. Feedback from installation Integrated Training Area Management (ITAM) GIS technicians helped shape the final form of the modeling tool as a sequence of smaller and quicker performing submodels rather than one large model. Comments and questions generated during site visits were useful during creation of the integrated help system to guide users through the geoprocessing procedures.

The final model—Rapid Soil Erosion Assessment Toolbox—is comprised of six total submodels labeled within the toolbox by number and title. Users run the model and arrive at final results by executing each sub-model in numeric sequence from #1 (Prepare Filled DEM) to #4 (Predict Gully Locations). Sub-models #5 and #6 can be used to forecast future gully locations by modifying the installation DEM and Manning’s n grid with GPS-derived vehicle tracks. Average run time for the model, including all sub-models and assuming a non-light detection and ranging (LIDAR) DEM and satellite resolution LULC grid for the installation is already on hand, ranged between 1.1 and 8.3 minutes on a typical desktop computer workstation. When working with spatial data at the spatial resolution of installation LIDAR datasets (3 m), processing time was three to four times longer in duration.

Project results were largely positive, despite not meeting the success criteria established for all performance objectives. One issue that remains with interpreting model results is the large rate of false positive gully predictions (i.e., errors of commission). Despite this issue, when used as a guide to identify areas with the most potential for gully erosion, the model results support time and cost savings by significantly reducing the total area of land being considered for more frequent and/or intensive field monitoring.

If the nLS+ model finds widespread use across DoD installations, it is important that an organization be identified to maintain the model and make any updates and/or revisions necessitated by future GIS software versions. As with any GIS-based model for which installation use is expected, the nLS+ model is an ideal candidate to be delivered to end-users as a web-based geoprocessing service made available and maintained by a central environmental organization. Uniform data and information products could then be delivered on a scheduled basis to installations for action. All data products required by the nLS+ model, and other environmental/sustainability indicators, could be stored in a geospatially-enabled relational database to facilitate access to current data. Storing data in the geospatially enabled relational database supports distributed viewing and editing by land managers and their staff, as well as distribution via mapping and geoprocessing web services similar to the “ITAM Map Viewer” web application used by Fort Riley (http://services.geog.ksu.edu/frk_rtla).