Predicting exotic annual grass abundance in rangelands of the western United States using various precipitation scenarios

Expansion of exotic annual grass (EAG), such as cheatgrass (Bromus tectorum L.) and medusahead (Taeniatherum caput-medusae [L.] Nevski), could cause irreversible changes to arid and semiarid rangeland ecosystems in the western United States. The distribution and abundance of EAG species are highly affected by weather variables such as temperature and precipitation. The study's goal is to understand how different precipitation scenarios affect EAG abundance estimates and dynamics, and we develop a machine learning modeling approach to predict how changes in annual and immediate past precipitation patterns could affect the abundance of EAG. The machine learning predictive model used seed source from previous years, weather variables, and soil profiles to drive its predictions. We achieved excellent training accuracy (r = 0.95 and median absolute error [MdAE] = 2.36% cover) and strong test accuracy (r = 0.79 and MdAE = 4.54% cover). We developed five versions of EAG abundance maps for 2022 with different precipitation scenarios: 9 yr of average precipitation, half of the average, three-fourths of the average, one and one-half times the average, and two times the average. The approach presented can be replicated to new study domains and easily modified for use with other precipitation scenarios. Developing multiple versions of a year's EAG spatially explicit abundance dataset predictions from multiple weather-based scenarios can provide important information to land managers as they prepare for variable EAG dynamics each year. Informed annual predictions based on weather scenario−driven models have the potential to improve fire preparation decisions.