The profound impact of soil temperature (TS) on crucial environmental processes, including water infiltration, subsurface movement, plant growth, and its influence on land–atmosphere dynamics, cannot be undermined. While satellite and land surface model-based data are valuable in data-sparse areas, they necessitate innovative solutions to bridge gaps and overcome temporal delays arising from their dependence on atmospheric and hydro–meteorological factors. This research introduces a viable technique to address the lag in the Famine Early Warning Network Land Data Assimilation System (FLDAS). Notably, this approach exhibits versatility, proving highly effective in analyzing datasets characterized by significant seasonal trends, and its application holds immense value in watershed-scaled hydrological research. Leveraging the enhanced state-space (SS) method for forecasting in the FLDAS, this technique harnesses TS datasets collected over time at various depths (0–10 cm, 10–40 cm, and 40–100 cm), employing a multiplicative SS model for modeling purposes. By employing the 1-step, 6-step, and 12-step-ahead models at different depths and 2 locations in Quebec, Canada, the outcomes showcased a performance with an average coefficient of determination (R2) of 0.88 and root mean squared error (RMSE) of 2.073 °C for the dynamic model, R2 of 0.834 and RMSE of 2.979 °C for the 6-step-ahead model, and R2 of 0.921 and RMSE of 1.865 °C for the 12-step-ahead model. The results revealed that as the prediction horizon expands and the length of the input data increases, the accuracy of predictions progressively improves, indicating that this model becomes increasingly accurate over time.
Read full abstract