Winter habitat indices from Landsat 8 and Sentinel-2 imagery for the contiguous US

In seasonally cold ecosystems, ecological processes and biotic communities are strongly influenced by winter conditions. Climate change is affecting these conditions, particularly in the Northern Hemisphere, as temperatures during the cold season continue to warm and alter patterns of frozen ground and snow cover. Yet, the current understanding of the ecological impacts of these changes is limited. A fundamental first step in addressing this knowledge gap is to quantify winter conditions with ecologically meaningful indices across large areas and at spatiotemporal resolutions relevant to on-the-ground management. Here, our goal was to combine Landsat 8 and Sentinel-2 (L8S2) data to derive three 30-m indices of winter conditions (winter habitat indices or WHIs): snow season length, percentage of days of frozen ground without snow, and snow cover variability, for the contiguous United States. We assessed the accuracy of the L8S2 WHIs using a nationwide network of meteorological stations and examined their error rates by land cover type, elevation, and the number of cloud-free observations available for each index calculation. Last, we compared the spatial patterns and errors in the L8S2 WHIs with those in WHIs derived from coarse-resolution MODIS (Moderate Resolution Imaging Spectroradiometer) imagery (500 m). We found that all three L8S2 WHIs accurately characterized winter conditions on the ground. They also had very similar accuracy and spatial patterns as the MODIS WHIs, despite having a lower imaging frequency and the lack of extensively validated snow cover products akin to those from MODIS. The accuracies of the WHIs were generally highest in mountainous areas of the western US and in vegetated areas, and they were lowest in parts of the midwestern and eastern US where cloud-free observations were less frequent, and in developed and barren areas. Having more cloud-free observations improved the accuracy of the WHIs, especially snow season length. Owing to their higher resolution, the L8S2 WHIs detected far more spatial detail than those from MODIS, particularly in topographically complex regions where winter conditions are highly heterogeneous over short distances. Our results demonstrate that 30-m WHIs derived from L8S2 data accurately capture winter conditions across the contiguous US, with accuracies that rival those from MODIS. Thus, the L8S2 WHIs offer exciting opportunities for ecological applications at a 30-m resolution across large spatial scales.