| Summary, etc |
The process of choosing a suitable hydrological model, calibrating it, and evaluating its performance poses intricate challenges for a modeler. The process of model selection is frequently influenced by the modeller's familiarity and expertise with specific models, which tends to be highly context dependent. Consequently, within a given field, the hydrological community typically relies on a limited number of preferred models. The process of model calibration, which involves determining the optimal parameter values for a given model, is an intricate mathematical optimization procedure that can potentially yield parameter values that are not realistic. There exists a necessity to streamline the calibration procedure and enhance the modellers' confidence in the outcomes of their models. This can be achieved through the implementation of a catchment classification scheme that not only provides guidance on the selection of parameters for calibration, but also specifies the appropriate calibration scheme to be employed. As a result, the computational resources and data prerequisites for the calibration process are minimized. Typically, model calibration is commonly conducted for streamflow at the outlet of a river basin, but this approach does not guarantee satisfactory performance in the upstream catchments. When conducting model calibration at multiple sites, this approach can effectively guarantee optimal performance across all sections of the river basin, as opposed to solely focusing on downstream areas. Typically, a multi-site calibration is conducted solely for streamflow, neglecting other hydrological variables that are integral components of the hydrological cycle. Additionally, it should be noted that a model that has been calibrated specifically for streamflow may not necessarily exhibit satisfactory performance when applied to other variables, such as evapotranspiration or root zone soil moisture. When additional variables are incorporated into a multi-site multi-variate model for calibration and validation purposes, the resulting calibrated and validated model can produce dependable simulations not only for streamflow, but also for various other hydrological variables. The evaluation of model performance and the selection of an appropriate model calibration scheme are typically conducted based solely on performance measures, without considering the various sources of uncertainties and the robustness of the model. There exists a necessity to formulate a comprehensive framework for the selection of model evaluation and calibration schemes. This framework should consider not only the values of performance measures but also take into account various sources of uncertainties and the robustness of the model. The implementation of an evaluation framework that incorporates uncertainty enhances the reliability and utility of model outcomes, facilitates more informed decision-making, and contributes to the continuous advancement of hydrological models and their practical applications. This thesis comes under the purview of the development of a model evaluation and calibration scheme selection framework. The thesis can be broadly divided into three parts. This first part evaluates and compares two models: the Water Systems Integrated Modelling framework (WSIMOD), an integrated water systems model that is applied outside of the UK for the first time, and the Soil and Water Assessment Tool (SWAT), a hydrological model that is widely employed in India. This assessment is done through a physiography-based catchment classification scheme for high, medium, and low flows in a large river basin in India, the Cauvery River Basin having a catchment area of 85,626 square kilometres. It establishes a connection between catchment characteristics and model performance and provides guidelines for "informed calibration". The results demonstrate that neither model gives satisfactory performance in an uncalibrated condition; however, WSIMOD has superior performance compared to SWAT in its default configuration, particularly when simulating average flows. WSIMOD requires more input data preparation compared to SWAT but has lesser calibration complexity. SWAT’s performance is greatly influenced by the catchment characteristics, making multi-site calibration crucial while WSIMOD has a lesser dependency on catchment characteristics, allowing regions within the same agro-ecological zones to have the same parameter values. SWAT calibration must include parameters related to topography, precipitation, and soil, whereas, in WSIMOD, infiltration rate, residence times of water for baseflow, surface, and subsurface runoff are more important. Catchment classification as an extra step before model calibration informs model parameterization and mitigates calibration complexity. The subsequent section of the thesis evaluates the efficacy of five multi-site multi-variate model calibration scenarios by analysing performance measure values. Three variables (streamflow, actual evapotranspiration, and/or root zone soil moisture) are used for the calibration of SWAT model for 18, 40, and 39 subbasins respectively, in a large river basin in India, the Mahanadi River Basin, having an area of about 1,41,589 square kilometres. The five calibration scenarios are: simultaneous streamflow calibration, sequential streamflow calibration, streamflow plus actual evapotranspiration calibration, streamflow plus root zone soil moisture calibration, streamflow plus actual evapotranspiration with root zone soil moisture calibration. The incorporation of additional variables, beyond streamflow, during the calibration procedure facilitates the parameterization of subbasin areas lacking in-situ streamflow measurements. The calibration of multi-site streamflow models exhibits reduced performance for upstream outlets in comparison to downstream outlets. The simultaneous calibration approach yields favourable results in terms of the performance measure, specifically the Kling-Gupta Efficiency, for the upper Mahanadi basin. Conversely, the sequential calibration method, which is executed in a stream-order-wise manner, demonstrates superior performance for the middle and lower Mahanadi basins. The incorporation of additional variables in the calibration process has been observed to enhance the model's performance in simulating variables beyond streamflow exclusively. In the context of multi-site multi-variate calibration, the approach that incorporates all three variables, namely streamflow, actual evapotranspiration, and root zone soil moisture, during the calibration process yields the most favourable outcomes in terms of performance measure values. The third part of the thesis presents the development of an evaluation framework that incorporates various factors, including performance measure values, model robustness, parametric, observational, model simulation, aggregated, and sampling uncertainty of performance measures. Simultaneous streamflow calibration results in reduced aggregated uncertainty for the upstream outlets, while sequential streamflow calibration demonstrates superior performance for the downstream outlets. Calibration for all three variables improves the observational, aggregated, and sampling uncertainty of performance measures. The inclusion of variables other than streamflow in model calibration leads to an increase in the model simulation uncertainty. When prioritizing model simulation uncertainty, the use of actual evapotranspiration over soil moisture must be considered. Conversely, when prioritizing observational uncertainty, the use of soil moisture over actual evapotranspiration must be preferred for calibration purposes, in conjunction with streamflow. The framework and findings presented in this study offer valuable guidance for calibration efforts focused on mitigating various sources of uncertainties in modelling heterogeneous river basins. |
