Spatio-temporal Modeling Of Snow Cover And Snowmelt Runoff Modeling

Snow is an important component of the Earth's surface. Seasonal snow cover and its melt not only dominate the regional climate and hydrology, but also constantly affect the global energy balance and climate change. Therefore, accurate estimation of snow cover extent, analysis of spatio-temporal variation of snow cover, and simulation of snowmelt runoff are important to water resources management and climate change studies.The Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover products have been widely used for regional snow cover extraction and hydrological modelling. However, as an optical sensor, MODIS observations are severely affected by cloud cover, resulting in large data gaps, particularly for daily snow products. In addition, the snow product accuracy is significantly lower for forested areas and topographically complex regions, and also for the time periods when snow cover is thin and ephemeral. Spatio-temporal dynamics of seasonal snow cover is able to reveal the regional climate change and water resource balance. To monitor and analyze the spatio-temporal variation of seasonal snow cover, previous studies generally investigated the changes in the spatial extent and seasonal durations of the snow-covered period by using pixel-based methods from time series of snow cover images. However, these pixel-based methods cannot reveal the spatio-temporal evolutionary relationships among snowpacks. In addition, the ephemeral snow cover during transition period cannot be captured by these pixel-based methods, thus would induce bias in the analysis results of seasonal duration. Snowmelt Runoff Model (SRM) is one of the most widely used models for modeling and forecasting snowmelt runoff, in which snowmelt runoff is calculated from daily snow covered area derived from satellite images. In SRM, daily snow cover is reconstructed by employing the snow depletion curve since the standard MODIS daily snow cover products often suffer from cloud contamination. The detailed analysis of SRM simulated runoff difference relative to different snow cover has not been conducted in previous studies.To fill the previous research gaps on snow cover, this paper improved MODIS snow cover products with a Hidden Markov Random Field (HMRF)-based spatio-temporal modeling technique, proposed an object-based spatio-temporal analysis of snow accumulation and depletion processes, and analyzed the SRM-simulated runoff difference resulting from snow cover difference in snowmelt season. The research objectives and important conclusions are:(1) Improving MODIS snow cover products with a HMRF-based spatio-temporal modeling technique.This research presents a spatio-temporal modeling technique for producing daily snow cover estimates without data gaps, based on time series 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 the new technique is quantitatively evaluated by comparing the snow cover estimates with in situ observations at 33 SNOwpack TELemetry (SNOTEL) stations as well as original MODIS snow cover products over the Rio Grande Basin during 2006-2007 snow season. There are as high as 34.1% data gaps in the original daily Terra/Aqua combined MODIS snow products over 2006-2007 snow season, mainly due to the influence of clouds. The presented HMRF-based spatio-temporal modeling technique achieved the snow cover estimate for the area of original data gaps with an accuracy of 88.1%. For the areas where are not covered by clouds, the HMRF-based technique also improved the snow cover estimate accuracy of original MODIS snow products by 3.3%, from 85.3% to 88.6%. Our analysis shows that the effects of data gaps filling and overall accuracy of snow cover area estimates are significantly improved, when spatio-temporal contextual information and environmental association information are progressively incorporated within the HMRF framework. Original MODIS snow products have a relatively low accuracy during the snow transition periods and in forest-covered areas. Our HMRF-based technique increased the snow product accuracy by 4.2% for the whole transition periods, and by 5.8% in March during the snow depletion period. Significant accuracy improvements on snow cover estimate have been also achieved over evergreen forests and mixed forests.(2) Object-based spatio-temporal analysis of snow accumulation and depletion processes.Snowpack accumulation and depletion is a complex and dynamic geophysical phenomenon. This study proposed an object-based spatio-temporal framework for analyzing snow accumulation and depletion processes for the first time, in which snowpack is defined as a field-object. This framework organizes data into snow zones, snow sequences, and snow processes based on the levels of spatio-temporal aggregation of snowpack. A snow zone is a spatially continuous aggregate of snow pixels identified from a snow thematic image. A snow sequence represents a continuum of snow zones over space and time. A snow process is defined as a spatially and temporally continuous aggregate of sequences. The thematic attributes, spatial attributes, and temporal attributes, as well as the thematic relationships, spatial relationships, and temporal relationships of snow zones, snow sequences and snow processes are organized and stored by employing an objected-based method. A case study was conducted in Upper Rio Grande basin to analyze the snow accumulation and depletion process during 2006-2007 snow season. The results indicated that the proposed object-based analysis is able to describe the snow accumulation and depletion process of each snowpack during its life-cycle, and to reveal the spatio-temporal evolution of intra- and inter-snowpack.(3) Analysis of the SRM simulated runoff difference with respect to the influence of snow cover difference during the snowmelt season.This study investigated the SRM simulated runoff difference with respect to the influence of snow cover difference between MODIS standard snow cover products and HMRF-improved snow cover products. Daily snow cover area from MODIS snow products is obtained from the construction of the snow depletion curve. This snowmelt runoff simulation was conducted in Rio Grande Headwater basin during 2007 snowmelt season. Snow depletion curves developed from two snow products were generally comparable in whole snowmelt season. However, the one improved by HMRF method is able to capture more detailed spatio-temporal variation of snow cover. The simulated streamflow based on MODIS snow products and HMRF-improved snow products are both significantly correlated with in situ streamflow. Particularly, the simulated runoff obtained from HMRF-improved snow products shows better accuracy. The results show that the simulated runoff differences resulting from snow cover differences depend on the snowmelt period. In the early snowmelt period, runoff contributed by snowmelt is quite low due to the constant low air temperature, thus snow cover differences cannot generate obvious differences on simulated runoff. In the middle and late snowmelt period, runoff is more sensitive to snowmelt due to the warming air temperature, thus snow cover differences directly lead to the simulated runoff differences.