Improving short term streamflow forecast using SNOTEL data assimilation

Streamflow forecasts rely on the quality of hydrologic initial conditions and weather forecasts. Therefore, they might benefit from the assimilation of snow water equivalent (SWE) and soil moisture observations. Although networks of soil moisture observations are nearly nonexistent, under some conditions streamflow can provide an index to antecedent soil moisture, and SWE is measured at SNOTEL/CDEC sites over much of the mountainous western U.S. The time of concentration of a basin is a key determinant of the factors controlling potential streamflow prediction skill; for lead times less than the time of concentration forecast skill is dominated by hydrologic initial conditions, whereas for longer lead times, weather forecast accuracy becomes increasingly important. Furthermore, for lead times shorter than the time of concentration, the flow forecast is quasi deterministic (because much of the water that will arrive downstream during the forecast lead time is already in the river). Spatially distributed weather forecasts (and hence calibration approaches) are desirable if consistent forecasts at multiple forecast points within a basin are desired. We evaluate here an approach for improving short to medium range deterministic streamflow forecasts in snowmelt driven basins through assimilation of SNOTEL/CDEC SWE data several days prior to the forecast. In this approach, the reference spinup of the hydrologic model uses gridded station data set to force a hydrology model. The 10-day weather forecasts are assumed perfect (observations) in order to assess the potential improvement from the SNOTEL/CDEC data assimilation only. We focus here on river basins with relatively short concentration times, typically less than 5 days. In the reference run, no assimilation is performed. In the experimental runs, spatially distributed SWE conditions are assimilated from SNOTEL/CDEC observations at days 6 and 3 prior to and on the forecast day. The spatial distributions for the assimilation are generated using different weights. Bias and predictability of the control and experiment deterministic forecasts are evaluated over the Feather River basin, CA for the period 1985-2008.