Measurement error in remotely sensed fractional snow cover datasets: implications for ecological research

Abstract Snow cover and snow melt patterns are important features of the Arctic environment, with wide-ranging repercussions for ecology. Datasets based on satellite imaging—often freely available—provide a powerful means for estimating snow cover. However, researchers should be aware of the possible error and bias in such datasets. Here, we quantify measurement error in commonly used data on snow cover, and demonstrate how biases have the potential to alter conclusions of ecological studies. We established 38 quadrats (80 m × 50 m) across a study site of Arctic tundra near Utqiaġvik, Alaska. At each quadrat, we estimated fractional snow cover (FSC) and the timing of snow melt using data from moderate resolution imaging spectroradiometer (MODIS), visible infrared imaging radiometer suite (VIIRS), and Sentinel-2 satellites. We compared satellite-based estimates with data from drone imagery to quantify measurement error and bias. We then evaluated whether the measurement error and bias alter conclusions about the relationship between the timing of snow melt and the breeding phenology of a population of pectoral sandpipers Calidris melanotos . We found that satellite datasets tended to overestimate FSC, leading to late estimates for snow melt dates. The Sentinel-2 dataset gave the most accurate results, followed by VIIRS, with MODIS giving the least accurate results. The degree of error varied substantially with the level of FSC, with biases reaching up to 60% for MODIS and VIIRS datasets at intermediate FSC values. Consequently, these datasets resulted in substantially different conclusions about how snow melt patterns were related to settlement and nesting dates of pectoral sandpipers. Our study indicates that measurement error in FSC can be large with substantial variation in the degree of error among satellite products. We show that these biases can impact conclusions of ecological studies. Therefore, ecologists should be conscious of the limitations of satellite-derived estimates of snow melt, and where possible should consult studies validating snow measurements in environments comparable to that of their study system.