The application of state-and-transition models to remotely sensed fractional cover datasets

Assessing vegetation conditions in expansive rangeland ecosystems has long posed a persistent challenge. Recent advancements in remote sensing technologies have provided new avenues for improving this assessment process. In this study, conducted at two ranch sites in Nevada, we integrated existing line-point-intercept monitoring data with additional field observations to evaluate the vegetation condition or “State” at various monitoring points. We then employed machine learning techniques to classify gridded raster datasets, aligning them with existing state-and-transition models (STMs) specific to each study area. Leveraging Landsat-derived fractional cover datasets, along with climate and soil predictors, we aimed to predict vegetation State within specific Vegetation Great Groups (VGGs). The resulting vegetation State maps were then combined to generate a cohesive representation of vegetation conditions across the study sites. Our analysis revealed that relative functional group cover emerged as a superior predictor of vegetation state and ecological processes at the site level. However, we encountered variations in state mapping accuracy ranging from approximately 14% to 44% error. These discrepancies were influenced by factors such as study location, landscape heterogeneity, availability of training data, and species-specific challenges, all of which complicate the accurate classification of remote sensing datasets.