Resetting the baseline: Machine learning predicted meadows for 60 watersheds in the Sierra Nevada

This data publication contains the geospatial data layers generated from machine learning models. Random forest models were developed to identify potential historical meadow habitats in 60 watersheds of the Sierra Nevada, California in 2023. The models were trained using over 11,000 mapped extant meadow polygons from the Sierra Nevada MultiSource Meadow Polygons data. Geospatial predictor variables representing topographic position, relative elevation, flow accumulation, snowpack, and distance to stream channels were used to train the models to predict locations with similar hydrogeomorphic characteristics to modern meadows. This data publication includes prediction rasters representing continuous meadow probability values from 0-1 for each watershed generated by both local watershed-scale models and a Sierra Nevada-wide model. Polygon layers representing aggregated high probability meadow areas for each watershed from the local models and Sierra Nevada model are also provided. These polygons were generated by selecting contiguous pixels with values greater than 0.5 in the prediction rasters and converting to vector polygons. The provided data layers can be used to identify potential areas for meadow restoration that could increase groundwater storage, floodplain connectivity, biodiversity, and resilience to wildfire and climate change across the Sierra Nevada mountain range. The mapped historical meadow habitats greatly expand the known extent of meadows in the region.
We aimed to understand where and how frequently meadows historically occurred to reset the baseline condition and provide insight into their restoration potential. We trained machine learning algorithms to identify potential meadow areas with similar hydrogeomorphic conditions to extant meadows while ignoring their unique vegetative characteristics since we hypothesized that vegetation would change but geomorphology would remain.
For more information about this project and these data, see Cummings et al. (2023) as well as Cummings and Pope (2023).