Comparing stability in random forest models to map Northern Great Plains plant communities in pastures occupied by prairie dogs using Pleiades imagery

Abstract. Black-tailed prairie dogs (Cynomys ludovicianus) have been described as a keystone species and are important for grassland conservation, yet many concerns exist over the impact of prairie dogs on plant biomass production and consequently livestock production. The ability to map plant communities in pastures colonized by prairie dogs can provide land managers with an opportunity to optimize rangeland production while balancing conservation goals. The aim of this study was to test the ability of random forest (RF) models to classify five plant communities located on and off prairie dog towns in mixed-grass prairie landscapes of north central South Dakota, assess the stability of RF models among different years, and determine the utility of utilizing remote sensing techniques to identify prairie dog colony extent. During 2015 and 2016, Pleiades satellites were tasked to image the study site for a total of five monthly collections each summer (June–October). Training polygons were mapped in 2016 for the five plant communities and used to train RF models. Both the 2015 and 2016 RF models had low (1 %) out-of-bag error rates. However, comparisons between the predicted plant community maps using the 2015 imagery and one created with the 2016 imagery indicate over 32.9 % of pixels changed plant community class between 2015 and 2016. The results show that while RF models may predict with a high degree of accuracy, overlap of plant communities and interannual differences in rainfall may cause instability in fitted models. A final RF model combining both 2015 and 2016 data yielded the lowest error rates and was also highly accurate in determining prairie dog colony boundaries.