Validating a Time Series of Annual Grass Percent Cover in the Sagebrush Ecosystem

We mapped yearly (2000–2016) estimates of annual grass percent cover for much of the sagebrush ecosystem of the western United States using remotely sensed, climate, and geophysical data in regression-tree models. Annual grasses senesce and cure by early summer and then become beds of fine fuel that easily ignite and spread fire through rangeland systems. Our annual maps estimate the extent of these fuels and can serve as a tool to assist land managers and scientists in understanding the ecosystem’s response to weather variations, disturbances, and management. Validating the time series of annual maps is important for determining the usefulness of the data. To validate these maps, we compare Bureau of Land Management Assessment Inventory and Monitoring (AIM) data to mapped estimates and use a leave-one-out spatial assessment technique that is effective for validating maps that cover broad geographical extents. We hypothesize that the time series of annual maps exhibits high spatiotemporal variability because precipitation is highly variable in arid and semiarid environments where sagebrush is native, and invasive annual grasses respond to precipitation. The remotely sensed data that help drive our regression-tree model effectively measures annual grasses’ response to precipitation. The mean absolute error (MAE) rate varied depending on the validation data and technique used for comparison. The AIM plot data and our maps had substantial spatial incongruence, but despite this, the MAE rate for the assessment equaled 12.62%. The leave-one-out accuracy assessment had an MAE of 8.43%. We quantified bias, and bias was more substantial at higher percent cover. These annual maps can help management identify actions that may alleviate the current cycle of invasive grasses because it enables the assessment of the variability of annual grass−percent cover distribution through space and time, as part of dynamic systems rather than static systems.