The New Mexico meadow jumping mouse (Zapus hudsonius luteus), which was listed as endangered in 2014, historically ranged throughout the Middle Rio Grande River Valley in New Mexico and along perennial high-elevation streams in New Mexico, southern Colorado, and eastern Arizona. After years of drought, river modifications, and changes to habitat, many previously occupied jumping mouse populations are believed to be extirpated. Currently, Bosque del Apache National Wildlife Refuge in southern New Mexico hosts the only confirmed population of the jumping mouse remaining within the Middle Rio Grande River Valley.
The jumping mouse is dependent on early successional wetland habitat for the food and shelter it needs to survive and reproduce. To create this habitat, the refuge periodically generates disturbances that mimic the historical ones that maintained this early successional habitat, and carefully allocates its limited water supply to maintain soil moisture at necessary levels.
To track the success of habitat management actions, researchers paired field surveys of jumping mouse habitat conditions with remote sensing imagery (Sentinel-2) to create maps of suitable habitat over time. Because conditions in an area can change rapidly within a growing season, three maps were created for each growing season. Researchers used random forest models to predict categories of habitat quality (non-habitat, medium-quality, or high-quality) based on multiple predictor variables. Field survey data (vegetation structure, composition, and soil moisture) from approximately 1,500 surveys per year were matched to the Sentinel imagery layer with the closest date. Remote sensing data included 13 indices of vegetation condition calculated from the Sentinel-2 imagery. Furthermore, 2011 lidar-derived data (vegetation height, aspect, and slope) were also used as predictor variables. Random forest models generally explained habitat conditions well as quantified by the Area Under Curve (AUC) metric, which ranged between 0.94 to 1.00, depending on the model. AUC values range between 0 and 1, with values of 0 not predictive and values of 1 perfectly predictive of class type. These results will be used to evaluate the effectiveness of management actions and track progress toward meeting habitat goals.
Extent of jumping mouse foraging habitat modeled using satellite and lidar data in a Random Forest classifier during 2018, 2019, and 2020 at Bosque del Apache National Wildlife Refuge, New Mexico. Early, Mid, and Late refer to the time periods when habitat was assessed. Early habitat: June to July, Mid: July to Aug, Late: September to October.