Real-time multi-source information assimilation improves multi-layer soil moisture for real-time flood forecasting

Accurate acquisition of the initial soil moisture (M_S) values and states within a basin is essential for enhancing real-time flood forecasting accuracy. Based on the physical parameters of the Xinʹanjiang model, a point-sub-basin conversion model (WKNN) was developed. This model facilitates the transition of real-time M_S data from point to sub-basin scale, utilizing both point M_S measurements and CLDAS M_S data. The resulting sub-basin scale data served as observed data for assimilating the M_S process forecasted by the Xinʹanjiang model using the ensemble Kalman filtering method, enabling real-time flood forecasting in the middle area of the Wuqiangxi basin. Application results demonstrate that the WKNN effectively captures M_S relationships across different soil layers and scales, from point to sub-basin, exhibiting robust applicability as a scale conversion tool. Implementation of a multi-layer assimilation scheme resulted in a 12.67% decrease in average runoff error and a 41.82% increase in average deterministic coefficient. Real-time assimilation of M_S efficiently reduced forecast errors in model state variables, consequently improving flood forecast accuracy. Notably, substantial reductions in relative runoff errors were observed during flood events at the onset of the flood season, with the multi-layer assimilation scheme outperforming its single-layer counterpart.