The effect of discretization of hydrologic response units on the performance of SWAT model in simulating flow and evapotranspiration

The approach to hydrological modelling has changed over time in the face of a changing climate, land use and development and management of water resources, putting more emphasis on quantification of water balance components within the catchment than flow at catchment/sub-catchment levels. This has created a need to have an almost explicit representation of landscape heterogeneity, which can be handled by distributed hydrological models. A disadvantage of these distributed models is that they are highly parameterized and thus, attempts are made to simplify the model through lumping of some parameters and losing some of the fine details in the landscape heterogeneity. Various studies have investigated the effects of aggregation of model input parameters on model outputs. These effects vary with the chosen spatial scale (model input and calibration data), key processes simulated in the model and the types of model outputs considered. In this study, we investigate at a much finer scale, the impact of the number and size of hydrologic response units (HRUs) on flow and evapotranspiration. We applied the Soil and Water Assessment Tool (SWAT), a semi-distributed hydrological model, to an irrigated catchment in south east Australia. Three models were compared with different number of HRUs and parameters. These are i) 11-HRU, ii) 98-HRU, and iii) 11- HRUCrV (i.e. model with 98-HRU parameters applied to the 11-HRU model). This comparison was important because the number of remotely sensed actual evapotranspiration (ETrs) data points for calibration of the SWAT model depended on the total number of HRUs in the catchment, in addition to flow. After a one year warm-up, the SWAT model was calibrated using flow and ETrs. In general, the model catchment and within-catchment outputs (flow and evapotranspiartion) exhibited different sensitivities to the scale of calibration i.e coarse versus fine HRU representation. Monthly flow was calibrated with a reasonable accuracy. Catchment ET was not affected by the number of HRUs used in the model. This is because in the 11-HRU the dominant soil and land uses were used which together covered more than 70% of the catchment. In addition, the simulated catchment ET was not affected by the transfer of parameters between the models. On the other hand, flow seemed to be influenced both by the number of HRUs (mainly due to land use and irrigation management) and model parameterization. The 11-HRU model was seen to perform better than the 98-HRU in simulating catchment flow and ET. However, within- catchment water fluxes were found to be variable within a model as well as between the models, and so a more detailed catchment discretization would provide a better quantification of the water balance components. Greater variability of evapotranspiration observed across land use than the soil type suggests that fine discretization of land use data than of the soil type could improve within catchment water flux results in this catchment. The different sensitivities of flow and evapotranspiration (daily and monthly) to the scale of calibration can provide a guide as to which model output has a higher priority according to the objectives of the study.