A sensitivity assessment of the TOPKAPI model with an added infiltration module

SummaryIn this paper we extend the usefulness of the TOPKAPI model by adding a Green-Ampt infiltration module and make the model and source code freely available on the internet as PyTOPKAPI. Then, we investigate the sensitivity of the PyTOPKAPI hydrological model to systematic bias in the variables rainfall and evapotranspiration, as well as the physically based soil properties that describe the model behaviour. The model sensitivity is assessed in terms of relative changes in the Soil Saturation Index (SSI), which is defined as the percentage of soil pore space filled by water. The volumetric soil moisture content, can be calculated from SSI using location dependent soil properties, if required.The model sensitivity is calculated at 7200 sites in South Africa, for a 2.5 year simulation period with a time-step of three hours. This large spatial extent gives results for a wide array of climates and land properties. Overall, the sensitivity of the model turns out to be a closely linear function of, and the same order of magnitude as (or less than), the forcing/parameter bias. This indicates that the model is robust to errors in forcing/parameters.The results also show that the best estimates of soil water can be obtained by improving estimates of the storage parameters and rainfall forcing. However, the storage parameters must be obtained from static soil property data-sets and we show that there is value in making improvements to the rainfall forcing (in this case TRMM 3B42RT) for places where it is biased relative to observed rainfall.This work is particularly relevant for model application in ungauged basins, where the quality of forcing variables and physical parameters cannot be calibrated.

► Investigate sensitivity of PyTOPKAPI model to systematic bias in forcing/parameters.
► Sensitivity assessed over years for a wide array of climates and land properties.
► Model most sensitive to storage parameters and rainfall.
► The model is robust to errors in forcing/parameters and sensitivity is closely linear.
► Important for ungauged basins, where forcing/parameters cannot be calibrated.