Assessment of the Transferability of SWAT Model Parameters from Gauged to Ungauged Sub-Watersheds for Streamflow Simulation in the Upper Tana Watershed, Kenya

Abstract

Predictions in ungauged watersheds are regarded as some of the most challenging tasks in surface hydrology. A large amount of parameters and input data are required for the application of most hydrological models. Calibration of these models require high quality, sufficiently long term observation of streamflow and other variables, but observed data on both temporal and spatial scales of interest are always very limited. Due to the difficulty in direct implementation of hydrologic models in ungauged watersheds, alternative strategies for prediction are required. Prediction of streamflow in ungauged watersheds is performed through the transfer of hydrologic information (e.g., streamflow values, hydrologic indices, model parameters) from gauged to ungauged watersheds. The objective of this study was to assess the transferability of the Soil and Water Assessment Tool (SWAT) model parameters from gauged sub-watersheds for streamflow simulation in “ungauged” sub-watersheds of the Upper Tana Watershed. Three methods namely: spatial proximity, global averages and regression were evaluated as approaches for developing SWAT parameters values to enable estimation of daily streamflow for ungauged sub-watersheds with a certain degree of accuracy. In Upper Tana watershed, water is used for electricity generation by five main hydropower stations in Tana River, municipal water supply and for irrigation schemes. With increased demand of water to meet agricultural, domestic, municipal and industrial needs, there is an urgent need to manage water resources in the Upper Tana Watershed in a sustainable and integrated way. SWAT was calibrated at a daily time step in four sub-watersheds of the Upper Tana Watershed. Model calibration was first done manually and then automatically using the Sequential Uncertainty Fitting version 2 (SUFI-2) algorithm in the SWAT CUP software. The spatial proximity method was used to transfer parameters between neighbouring sub-watersheds. Global average parameters were determined by computing the mean of each of the parameters used in calibration. For the regression based transfer method, physical sub-watershed characteristics were derived from spatial data using GIS. Stepwise regression was used to develop equations which relate the sub-watershed characteristics to model parameters therefore enabling estimation of model parameters from sub watershed characteristics. The SWAT model performed well in simulating daily streamflow, attaining a coefficient of determination (R2) ranging from 0.57 to 0.69 and Nash-Sutcliffe efficiency (NSE) ranging from 0.51 to 0.67. The spatial proximity method yielded R2 ranging from 0.5 to 0.69. Global average parameters method attained R2 ranging from 0.54 to 0.67. For the regression based transfer method, R2 obtained ranged from 0.5 to 0.73. The spatial proximity method performed better than the global average and regression method. This was evident through the performance statistics and the simulation of the high and low flows. However, there is need to compare results from a different hydrological model in order to evaluate how the transfer approaches perform. The results of this study indicated that transfer of SWAT model parameters can be used to generate streamflow data in ungauged sub-watersheds for the purposes of water resources planning and management.