Enhancing SWAT’s snow module for multivariate Elevation-dependent snow and streamflow data assimilation

Abstract

The orographic effect is an influential process that controls the spatial distribution of precipitation in mid-high altitude areas. The Satellite-based snow cover fraction (SCF) product from MODIS is valuable data to understand and model such processes. The SCF simulation in the Soil and Water Assessment Tool (SWAT) model has some limitations that restrict the use of satellite-based SCF data. Firstly, the model provides discontinuous simulation (in time) of SCF. In other words, SWAT only simulates the SCF during snowmelt periods and does not provide any estimation of SCF for days in which snow accumulation occurs. Secondly, there is a mismatch between the spatial scale of snow parameters and snow processes in the model. In other words, SWAT considers snow parameters at the coarse (basin or subbasin) scale while simulates snow processes at the fine (Hydrologic Response Unite) scale. Due to these limitations, little effort has been made to use remotely sensed data (specifically SCF) for state and/or parameter estimation of the model. These limitations mostly restrict the model states and parameters estimation at the HRU scale and when the orographic effect is considered by the model. We address these restrictions by modifying the model’s snow processes. We propose a new methodology for multivariate assimilation of MODIS SCF and in-situ streamflow observation into the model when the model considers the orographic effects. Accordingly, we design different univariate and multivariate SCF and streamflow data assimilation (DA) scenarios to estimate the states and parameters of this model. Moreover, we investigate the impact of considering the Elevation Band (EB) capability of SWAT on both types DA scenarios. Results reveal that the EB-based multivariate DA scenario significantly improves the accuracy and robustness of assimilation results. Similarly, the multivariate assimilation improves the streamflow simulation accuracy compared to univariate streamflow assimilation.