Seasonal snow cover and its melt dominate regional climate and hydrology in many mountainous regions in the world. The Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover products have been widely used for regional hydrological modeling. However, data gaps in snow products due to frequent clouds remain a serious problem, particularly for daily products. This paper presents a spatio-temporal modeling technique for filling up data gaps in daily snow cover estimates, based on time series of Terra/Aqua MODIS images. The spatio-temporal modeling technique integrates MODIS spectral information, spatial and temporal contextual information, and environmental association within a Hidden Markov Random Field (HMRF) framework. The performance of our new technique is quantitatively evaluated by comparing our snow cover estimates with in situ observations at 33 SNOTEL stations as well as to original MODIS snow cover products over the Upper Rio Grande Basin during 2006–2008 snow seasons. Mainly due to cloud obscuration, there are as high as 32% data gaps in original Terra/Aqua combined MODIS snow products. Our HMRF technique reduced cloud-cover related data gaps to < 1% and achieved a snow-mapping accuracy of 88.0% for the gap-filled areas. For the areas not covered by clouds, our HMRF-based technique also improved the snow cover estimate accuracy of original MODIS snow products by 3.5%, from 85.1% to 88.6%. When spatio-temporal contextual information and environmental association information are progressively incorporated within the HMRF framework, the overall snow mapping accuracies are improved and omission errors are reduced. Particularly, our HMRF-based technique increased the snow product accuracy by 4.2% during whole transition periods, and by 6.2% in March during snow melt. The snow mapping accuracies were significantly improved over evergreen forests and mixed forests.
All Science Journal Classification (ASJC) codes
- Soil Science
- Computers in Earth Sciences