Improving the Accuracy of Snow and Hydrological Models Using Assimilation by Snow Depth

Abstract

The main aim of this study is to improve the underestimation of spatial snowfall distributions by using assimilation. Although measuring snowfall depth is crucial to evaluate snow water resources and predict snowmelt runoff in spring season, it is difficult to measure snow depth correctly with a gauge because wind speed strongly influences the capture ratio. Snowfall observation errors have a significant influence on the accuracy of hydrological model output. An evaluation of the distributed hydrological model was carried out in the Yoneshiro River Basin in Japan with a modification of the model using snow depth data. To reduce the measurement error using the snowmelt-runoff model, an assimilation policy based on the observed snow depth is included in the snow water equivalent (SWE) model at regular intervals. As a result, the assimilation improves the accuracy of both the snow depth estimation and the snowmelt-runoff simulation. The Nash-Sutcliffe coefficient is improved from 0.63 to 0.86 throughout the year and from 0.21 to 0.82 from March to May. The assimilation of snow depth can contribute to improvement of the hydrological model with higher accuracy compared with direct use of gauge data. Also, how to assimilate snow depth, such as an interval of the assimilation and its applicable timing, is discussed. The model suggested in this study can be helpful for water management–related activities and decision making.