Random Forest models were trained using 2010-2020 skipper observations and a stratified random sample of available areas. These data were related to a suite of covariates that were summarised at small (50 m radius) to large (800 m radius) landscape scales, as well as their native scale. The suite of covariates contained environmental data, such as potentially undisturbed grass or perennial grass and forb net primary productivity, spectral data such as Sentinel-2 derived MSAVI, and abiotic data related to climate, topography, and soils. Models were applied to 30 m resolution imagery. Outputs include classification (binary values indicating relative absence/occurrence) and relative probability of occurrence. Use this app to explore the models.
This app is intended to support the identification of landscape-scale landcover thresholds to prioritize landscapes for grassland restoration for wildlife benefits.
Weighted mean occurrence for an ensemble of species distribution models (Baird's Sparrow, Chestnut-collared Longspur, Sprague's Pipit, Thick-billed Longspur).
The Prairie Pothole Joint Venture developed a mask for potentially undisturbed lands (PUDL), utilizing multiple years of CLU data .We conducted supervised classification to classify the landcover types potentially undisturbed grass, potentially disturbed grass, shrub, and other cover classes. This layer represents the classified landcover classes potentially undisturbed grass, potentially disturbed grass (having qualities similar to training data that represents restored grasslands), and shrub within the PUDL layer. The extent includes multiply US Migratory Bird Joint Venture boundaries in the Great Plains.
