Streamflow and Soil Moisture Forecast Skills as a Function of the Biophysical Attributes and the Dynamic Soil Moisture Conditions in the National Water Model

Abstract:

The study assesses the streamflow and soil moisture forecast skills of the state-of-the-art National Water Model (NWM) by comparing it with the USGS streamflow and Soil Moisture Active Passive (SMAP) soil moisture. The fundamental hypothesis of this research is that the forecast skill is a function of biophysical attributes. Hence, the uncertainties in the forecast skill at the ungauged basins can be determined as a function of its biophysical attributes, and forecast lead-time.

The NWM is an unprecedented effort by the NOAA to provide very high resolution (1km) soil moisture and streamflow forecast for 2.7 million streams in the U.S. The forecast data are archived only for two days that pose a challenge for the long-term evaluation and characterization of the forecast skills. In the project, we have downloaded and archived the NWM forecast data for 400 days from December 2018 to February 2020 and assess the forecast skills at 382 gauged locations in the states of Alabama and Georgia located in the humid subtropical climate of the Southeastern United States.

The forecast skills are assessed using the following metric: anomaly correlation coefficient (ACC) and normalized root mean square error (nRMSE), volumetric biases, base-flow ratio, and time to peak. We have used the following biophysical attributes: watershed area, soil texture, land use, and topography index (TI) to characterize the skills using the Classify and Regression Trees (CART) analysis. We found that the top one hundred large watersheds streamflow are more accurate than the smallest one hundred watersheds. The cultivated crop-land and forested watersheds show better accuracy than urban watersheds. Base flows are consistently underestimated in the coastal watersheds, and however, the base flow underestimation decreases as we move 100 km away from the coastline. The water moves slower in the NWM, resulting in the delayed time to peak compared to the USGS observations. Land-use type is the first order determinant, followed by the soil texture, and TI for the streamflow forecast skill. We will also present the evaluation of the soil moisture forecast skill, and effects of the dynamic soil moisture initial conditions on the streamflow forecast. Overall, this study provides a scientific basis for developing the uncertainty assessment for the ungauged basins.