- This study was conducted to determine if bacterial DNA present streams could be used to predict upstream watershed characteristics. Previous studies have found that bacterial composition in soil is influenced by land use. It was hypothesized that if the bacteria present in a stream is known that it can be used to predict upstream watershed characteristics. Collecting bacterial data involved sampling at 62 different sites in Oregon. The bacterial DNA from these samples were then extracted resulting in a spreadsheet of operational taxonomic units (OTUs). Land cover characteristics for each site were obtained by delineating each site’s watershed in StreamStats. The OTU and StreamStats data were used as inputs to create a model using support vector regression (SVR) in python to predict land cover characteristics. The SVR inputs kernel and C value were manipulated to improve the model along with the prevalence of OTUs. The largest Nash-Sutcliffe efficiency (NSE) value obtained when manipulating the model for forest and shrub cover was 0.26 using an ‘rbf’ kernel, C value of 20433 and a prevalence greater than 91%. This indicates that the model produces a better prediction of land coverage than using the average of all the sites’ land cover.
- Key Words: land cover, bacterial DNA, support vector regression, Nash Sutcliffe efficiency