GIS RS HABITAT MODELING APPROACHES TO IDENTIFY RIPARIAN COMMUNITIES ON THE PINE RIDGE RESERVATION

We sampled woody riparian populations and habitat at forty-eight locations on the Pine Ridge Reservation in 2007-2008. In 2009, we began analyzing the field data using a variety of statistical and spatial modeling tools. Using Ward’s method on a Euclidian distance matrix with PC-ORD software, we clustered eight species present in the overstory into cottonwood-willow, hardwood, juniper, and Russian olive community types. We tested habitat variables for Plains cottonwood Populus deltoides at the seedling and adult age classes using discriminant analysis in SPSS. The analysis indicated percent bare ground, Pfankuch rating, and percent White River was positively correlated with P. deltoides seedling occurrence, and percent grass was negatively correlated with P. deltoides seedlings. Using the hydrology toolset in the Spatial Analyst extension of ArcGIS, we derived twenty-eight landscape-level variables from SSURGO soils layers and 10-meter DEM derivatives and averaged the variables at a 3^rd order watershed level after generating simulated watersheds from the DEM data. We clustered the watersheds into ten physiographic regions by an isomeans clustering method after transforming the variables into their principal components using Erdas Imagine. We also classified summer and winter 2007 Landsat TM scenes by a supervised classification method using ERDAS Imagine. Our final step is modeling cottonwoods habitat using soil properties, physiographic regions, and classified Landsat layers using an ecological modeling program MaxEnt. The results of the analyses to date indicate: 1) riparian community type on the Pine Ridge reservation is strongly influenced by differences in geomorphic response to precipitation, which can be predicted by bedrock geology, and other landscape level variables, 2) we can use these landscape level variables in a final model to accurately predict riparian community type on the Pine Ridge Reservation, 3) MaxEnt ecological modeling software appears to predict cottonwoods occurrence at a much finer level of spatial detail than was possible using landscape or watershed level analyses.