Improving hydrological and vegetation modelling using regional model calibration schemes together with remote sensing data

Remotely sensed data are widely used for estimating hydrological variables, such as land surface soil moisture, land surface evapotranspiration and catchment runoff because they provide temporally dynamic and spatially explicit information on land surface characteristics. Passive microwave observations have been used to infer surface soil moisture information because they are not affected by cloud cover and there is a physical relationship relating emissions to soil water. Remote sensing vegetation cover types and leaf area index time series data have been used as inputs into distributed, semi-distributed and lumped hydrological models (Liu et al., 2007). This paper investigates the potential to improve runoff, soil moisture and vegetation dynamics predictions in ungauged catchments using a land surface hydrological model, AWRA-L, together with remotely sensed leaf area index measurements from NOAA-AVHRR and surface soil moisture measurements from TRMM-TMI. The study is conducted in 579 unregulated catchments across Australia. The AWRA-L model was regionally calibrated (i.e. a single set of parameters optimised) for half the catchments in four experiments: (1) against daily recorded streamflow data (Exp1); (2) against daily recorded streamflow together with monthly NOAAAVHRR leaf area index data (Exp2); (3) against daily recorded streamflow together with daily TRMM-TMI soil moisture data (Exp3); and (4) against all three data sets (Exp4). Next, the four optimised parameter sets obtained from the four regional calibration schemes were applied to the remaining half of the catchments for validation to evaluate the modelling skills for daily runoff and soil moisture predictions in independent catchments. This validation gives an indication of the abilities of the different calibration schemes to provide predictions in ungauged or poorly gauged catchments. The results here show that (1) it is technically feasible (i.e. use of advanced scientific computing, such as CSIRO GPU cluster) to use regional model calibration for hydrological modelling for continental Australia; (2) the incorporation of remotely sensed data into the calibration objective function marginally improves the daily runoff estimates, but noticeably improves the leaf area index and soil moisture estimates in the validation catchments; (3) the biggest benefit comes from Exp4 calibrating against recorded runoff and remotely sensed leaf area index and soil moisture observations. This study is being extended to investigate regional calibration over hydroclimate regions (rather than across the whole of Australia) and in a gridded modelling application to better use the spatial remotely sensed data and to represent rainfall gradients within catchments. It is likely that this, together with adaptation of surface hydrological models to make better use of remotely sensed data, will improve runoff estimates across large regions and the impact of climate and land use changes on runoff. It is noted that the global optimiser, the genetic algorithm toolbox built in MATLAB® did not found global optimum for the regional model calibration scheme one. Nevertheless, this should not noticeably impact the comparison results between the four regional calibration schemes in the validation catchments. This is an ongoing study. It needs to re-configure the optimiser to for obtaining better regional model calibrations.