Effect of antecedent rainfall on daily flow forecasting using a soil moisture accounting algorithm

Studies have shown both rainfall and soil moisture have a noticeable impact on daily runoff generation. In many cases, measured soil moisture data are unavailable, and soil moisture is instead estimated by various proxies, including the sum of precipitation over a number of days. Here we test the predictive value of antecedent rainfall for daily flow forecasting, using the Kuhesookhteh Watershed in Iran as a test case. A 20-year runoff time series was simulated using the Soil Moisture Accounting Algorithm of HEC-HMS. The results showed a Nash-Sutcliffe Efficiency of 0.67 for the calibration period (2000-2015) and 0.53 for the validation period (2015-2020).

Comparisons of daily simulated and observed flows show that the soil moisture accounting algorithm did not forecast the high values of streamflow well. We found a non-linear relationship between antecedent precipitation and the residuals of the flow simulation. Flow simulations substantially improved (i.e., residuals substantially decreased) when up to 4-5 days of antecedent rainfall were used as soil moisture proxies; further extending this antecedent rainfall interval to 7 days resulted in only minor further improvement. Since antecedent rainfall can be considered as a proxy for soil moisture, we infer that soil moisture acts as a system memory that retains information for at least 4-5 days. This inference is also supported by a data-driven, model-independent technique (Ensemble Rainfall-Runoff Analysis), applied to quantify the nonstationary runoff response of the Kuhesookhteh Watershed under different levels of antecedent rainfall

Keywords: soil water balance, surface abstraction, effective rainfall, water budget