Sierra Nevada snowpack and runoff prediction integrating basin-wide wireless-sensor network data

We focus on characterizing snowpack and estimating runoff from snowmelt in high elevation area (>2100 m) in Sierra Nevada for daily (for use in, e.g. flood and hydropower forecasting), seasonal (supply prediction), and decadal (long-term planning) time scale. Here, basin-wide wireless-sensor network data (ARHO, http://glaser.berkeley.edu/wsn/) is integrated into the USGS Precipitation-Runoff Modeling System (PRMS), and a case study of the American River basin is presented. In the American River basin, over 140 wireless sensors have been planted in 14 sites considering elevation gradient, slope, aspect, and vegetation density, which provides spatially distributed snow depth, temperature, solar radiation, and soil moisture from 2013. 800 m daily gridded dataset (PRISM) is used as the climate input for the PRMS. Model parameters are obtained from various sources (e.g., NLCD 2011, SSURGO, and NED) with a regionalization method and GIS analysis. We use a stepwise framework for a model calibration to improve model performance and localities of estimates. For this, entire basin is divided into 12 subbasins that include full natural flow measurements. The study period is between 1982 and 2014, which contains three major storm events and recent severe drought. Simulated snow depth and snow water equivalent (SWE) are initially compared with the water year 2014 ARHO observations. The overall results show reasonable agreements having the Nash-Sutcliffe efficiency coefficient (NS) of 0.7, ranged from 0.3 to 0.86. However, the results indicate a tendency to underestimate the SWE in a high elevation area compared with ARHO observations, which is caused by the underestimated PRISM precipitation data. Precipitation at gauge-sparse regions (e.g., high elevation area), in general, cannot be well represented in gridded datasets. Streamflow estimates of the basin outlet have NS of 0.93, percent bias of 7.8%, and normalized root mean square error of 3.6% for the monthly time scale.