Georgia Harrison | Search USGS

USGS2025

An accuracy assessment of satellite-derived rangeland fractional cover

Satellite-derived maps of vegetation cover provide detailed information about vegetation spatiotemporal patterns and are increasingly used to better understand and manage rangelands. Despite their utility, questions remain regarding the regional and site level accuracy for these maps, especially compared to field-collected data. We conducted an accuracy assessment of the Rangeland Analysis Platform (RAP), using over 17,000 field plots sampled through nationwide rangeland vegetation monitoring programs in the continental U.S. We observed higher overall nationwide map error compared to previous validations of RAP, and absolute error (Mean Absolute Error [MAE] and Root Mean Square Error [RMSE]) was highest for perennial herbaceous and bare ground and lowest for trees (MAE range = 2.98 –10.22 %). There were also differences in map agreement with field data across ecoregions. Generally, map agreement was highest in the Great Basin and lowest in the Great Plains and Desert Southwest. Additionally, we assessed the suitability of using RAP in riparian and wetland areas, which are absent in the current version’s training. Errors for bare ground in riparian areas were lower than errors of upland accuracy assessments (upland MAE = 10.22 %, riparian MAE = 7.22 %), but for all other functional groups, riparian error was higher (ΔMAE range: 0.21 – 20.49 %). We examine how our results could inform regional applications of fractional cover data while considering error and uncertainty and identify areas for potential model improvement. Our findings inform the use of RAP regionally and provide a technique for evaluating other vegetation mapping products for use in rangeland management.

Ecological Indicators↗

USGS2025

Spatial scale dependence of error in fractional component cover maps

Geospatial products such as fractional vegetation cover maps often report overall, pixel-wise accuracy, but decision-making with these products often occurs at coarser scales. As such, data users often desire guidance on the appropriate spatial scale to apply these data. We worked toward establishing this guidance by assessing RCMAP (Rangeland Condition Monitoring Assessment and Projection) accuracy relative to a series of high-resolution predictions of component cover. We scale the 2-m and RCMAP predictions to various focal window sizes scales ranging from 30 to 1 500 m using focal averaging. We also evaluated variation in scaling effects on error at ecoregion and pasture (mean area of 1 050 ha) scales. Our results demonstrate increased accuracy at broader windows, across all components, and most increases in accuracy level off at ∼200–600 m scales. At the scale with highest accuracy, cross-component average correlation ( r ) increased by 6.5%, and root mean square error (RMSE) was reduced 46.4% relative to 30-m scale data. Scaling-related improvements to accuracy were greatest in components such as shrub and tree with more spatially heterogeneous cover and in ecoregions with more spatially heterogenous cover. When components were aggregated at the pasture scale, r increased 10% and RMSE decreased 34.3% on average relative to the 30-m scale. Our results provide empirical data on the scale dependence of error, which fractional cover data users may consider alongside their needs when using these data. Although the general principle remains that remotely sensed products are intended to address landscape-scale questions, our analysis indicates that applying data at finer than landscape spatial scales and grouping even a handful of pixels resulted in lowered error compared to pixel-level comparisons. Our results quantify the trade-offs between data granularity and error related to scale for fractional vegetation cover.

western United States↗