Calibration, validation, parameter indentifiability and uncertainty analysis of a 2 – parameter parsimonious monthly rainfall-runoff model in two catchments in Zimbabwe

Hydrologic modelling lies at the core of hydrology and water resources management. Attempts at gaining a holistic grasp on model robustness, hydrologic theory and processes have inadvertently led to models that are not-well structured or too complex to apply in arid and semi-arid catchments and in Africa, in particular. In view of this, this paper reports on the application of a monthly parsimonious hydrologic model in two catchments in Zimbabwe, the Nyatsime and Upper Save river catchments. The two (2) parameter monthly parsimonious GR2M model was applied. The inputs were rainfall and potential evapotranspiration. Measured discharge was used for calibration and validation. Calibration and uncertainty analysis were done using the Differential Evolution Adaptive Metropolis (DREAM) algorithm. The performance of the GR2M model was evaluated using ten (10) model performance metrics. Parameter indentifiability was analysed on the basis of the shape of the posterior distribution of parameters. Parameter and total uncertainty were analysed in the context of the formal Bayesian DREAM approach. The 10 performance evaluation metrics showed that the model performed satisfactorily during calibration and validation in terms of the overall fit of observed and simulated stream flows, low flows and the runoff volumes. The Nash-Sutcliffe efficiency (NSE) was >0.85, the Kling–Gupta Efficiency (KGE) was >80% and Volume Efficiency was >59% during calibration. Slight performance drops were noted during validation except for the NSE in Nyatsime catchment whilst the KGE remained relatively high. The validation NSE was >0.65, the Kling–Gupta Efficiency (KGE) was >71% and Volume Efficiency was >55%. Calibrated parameters values showed good time-stability and were well identifiable with posterior parameter distributions having Gaussian shapes. Parameter uncertainty, in relation to total uncertainty was low. Parameter uncertainty constituted about 7% of the total uncertainty region. It was concluded that, although the model only had two parameters, the model performed quite satisfactorily in the simulation of monthly flows which makes it a good tool for operational hydrology and water resources modelling, planning and management especially in regions with inadequate data.