Incorporating vegetation time series to improve rainfall-runoff model predictions in gauged and ungauged catchments

Abstract

Conceptual lumped rainfall-runoff models are widely used for surface runoff predictions in gauged and ungauged catchments because they are simple, easy to calibrate, and give relatively accurate runoff predictions. Lumped rainfall-runoff model inputs are generally daily precipitation and potential evapotranspiration (or air temperature). Vegetation processes are seldom considered in these models but they can play an important role in controlling runoff production in mid-latitude catchments. The use of surface vegetation information in rainfall-runoff modeling may allow better estimation of water balance components, evapotranspiration and water storage change resulting in more accurate estimates of runoff. Most studies calibrate the rainfall-runoff models against gauged streamflow data and use regionalisation methods to specify parameter values to model runoff in the ungauged catchments. This study investigates the potential to improve runoff and soil moisture prediction by incorporating vegetation time series data into lumped rainfall-runoff modelling. The modelling experiments are carried out using a daily conceptual lumped rainfall-runoff model SIMHYD. Daily rainfall, meteorological and streamflow data, NOAA-AVHRR monthly remotely sensed leaf area and TRMM-TMI daily microwave surface soil moisture for 470 unregulated catchments (50-5,000 km 2) across Australia over the period of 1981 to 2006 are used. The SIMHYD model is adapted to incorporate leaf area index series and land cover types by modifying the evapotranspiration sub-model (called SIMHYD-ET), with an additional one parameter for SIMHYD (total of 10 parameters). The original and modified versions of the SIMHYD models are then calibrated against daily streamflow in each of the 470 catchments. The model's ability to predict runoff and soil moisture in 'ungauged' catchments is then assessed by using parameter values from the geographically closest gauged catchment. The calibration and prediction results of the SIMHYD and SIMHYD-ET models are then evaluated using the Nash-Sutcliffe Efficiency (NSE) of daily runoff, Water Balance Errors (WBE) percentage and correlation coefficient between modeled daily soil moisture and TRMM-TMI soil moisture. The modelling results indicate that the daily runoff series and total runoff volume modelled by the SIMHYDET model are similar to (or only very marginally better than) those simulated by the original SIMHYD model. The SIMHYD-ET model, however, performs noticeably better than the SIMHYD model in soil moisture predictions for both gauged and ungauged catchments. It is possible that better prediction skills can be achieved by modifying the lumped SIMHYD-ET model into a gridded model to take advantage of gridded/spatial rainfall and remote sensing data (leaf area index and land cover types) inputs.