On the Importance of Soil Moisture for Streamflow Forecasting
2017-02-09T05:18:40Z (GMT) by
Streamflow forecasting is essential for improving efficiencies in water use through reduced water losses on irrigation orders, and enhancing water management operations based on better information on inflows and off-takes in time and space. In addition, it provides valuable information on flood events for the dissemination of flood warnings with sufficient accuracy and lead time. Hydrologic forecasting models are used extensively in simulation of river flows in both flood and non-flood events. Quantitative Precipitation Forecasts (QPFs) from Numerical Weather Prediction (NWP) models are the primary source of rainfall data for input into hydrologic forecasting models, other than a forecaster’s intuition. <br> Soil moisture is a key factor controlling the hydrological behaviour of a catchment, particularly for flood modelling, as it controls transformation of rainfall into infiltration or runoff. Advances in remote sensing technologies have provided a variety of opportunities for improved hydrologic prediction, including the observation of land surface states such as soil moisture through time and across large areas. However, there has been limited effort to utilise such remote sensing information in hydrological modelling, especially in the context of operational applications. <br> The principal objectives of this thesis are i) evaluation of QPFs from the Australian forecast system product, ii) understanding the impact of soil moisture on streamflow prediction skill when used in the hydrologic model calibration stages, iii) assessment of satellite-based soil moisture observation constraint of the hydrologic model and its subsequent streamflow generation, and iv) the overall impact on the streamflow forecast skill when putting all three components together. <br> The NWP QPFs from the Australian Community Climate Earth-System Simulator (ACCESS) are evaluated against rainfall observations from a weather radar, to understand the uncertainties transferred to the streamflow forecasting model. The radar observations are first calibrated to remove the expected bias in the data according to in-situ rainfall observations. The QPFs evaluation indicates that significant rainfall uncertainty is expected to be propagated into the streamflow forecasting in this research. <br> Next, the ground-based measurement of soil moisture from research monitoring stations are used to calibrate and evaluate the soil moisture predictive capability in two rainfall-runoff models, Génie Rural 4 paramètres Horaire (GR4H) and Probability Distributed Model (PDM), and its subsequent effect on the streamflow predictions. Two calibration methods are tested; calibration to streamflow alone and joint-calibration using both streamflow and soil moisture observations. The results suggest that the GR4H model be used in Australia, in preference to PDM, and that soil moisture observations be used in the calibration process. <br> To investigate the impact of ongoing soil moisture constraint on streamflow forecasting, the root-zone soil wetness is first estimated from Soil Moisture and Ocean Salinity Mission (SMOS) satellites near-surface soil moisture retrievals. According to the comparisons with in-situ soil wetness data in the study area of this thesis, the exponential filtering technique is selected as the best approach. The hydrologic models are then constrained with the satellite-based root-zone estimates using a nudging approach, and the results are benchmarked against ground-based soil moisture data. It is shown that the effectiveness of soil moisture constraint depends on both catchment characteristics and the selected model for coupling soil moisture and runoff generation. <br> Finally, soil moisture constrained streamflow forecasts are assessed in the context of a real-time forecasting scenario, utilising both satellite-based estimates of root-zone soil moisture and NWP forecast rainfall. It is demonstrated that even with the degraded rainfall information, soil moisture constraint typically improve the streamflow forecasts, especially for moderate sized events, while for major events the forecasts are only improved for longer lead times.