Distributed streamflow predictions based on precipitation data from sparse ground networks and the North American Land Data Assimilation System in Northwest Mexico

There is a critical need to improve surface water supply forecasts in arid and semiarid regions of the Southwest United States and Northwest Mexico. Streamflow plays a significant role in sustaining a majority of economic activities; However, operational forecasting is a challenge due to the strong spatiotemporal variability of precipitation and landscape characteristics in this region and sparse climatic records.

In this study, a distributed hydrologic model, the TIN-Based Real Time Basin Simulator (tRIBS), was set up and parameterized to evaluate how rainfall variability and landscape characteristics impact streamflow generation in a large un-gauged watershed in northwest Mexico. The sparse forcing data in the region prompted us to explore alternative methods to provide meteorological conditions for streamflow predictions in the basin. We utilized rainfall and meteorological observations from: (1) a sparse network of ground-based stations (hourly resolution), (2) raw model products (12 km pixel, hourly resolution) from the North American Land Data Assimilation System (NLDAS) and (3) the NLDAS product adjusted using available rain and meteorological observations. Our distributed approach divided the large watershed (~9500 km2) into 291 subbasins. For each subbasin, simulations were conducted using recorded data from June 1, 2007 to May 31, 2008. A three year spin up period was used during the model initialization.

Our results indicate that the NLDAS raw model forcing estimates underestimated streamflow magnitudes generated from the sparse network data, but greatly improved the spatial variability in the model response. After adjusting the NLDAS data, the streamflow estimates improved in their overall magnitude in comparison to the sparse network estimates. Prior to adjusting the NLDAS data, precipitation did not exceeded the maximum saturated hydraulic conductivity (Ksat) in any of the subbasins; after adjusting NLDAS data 130 subbasins exceeded the maximum ‘Ksat' value. The result was that streamflow estimates from the adjusted NLDAS product exhibited interesting spatiotemporal patterns. We explore the spatiotemporal model response to identify the value added by the NLDAS products and their local corrections, such as creating high resolution runoff coefficient maps. Our results provide an example of the usefulness of NLDAS as forcing for hydrological models in real-world and international settings with sparse ground networks.