University of Twente Student Theses


Evaluation of spatial scale alternatives for hydrological modelling of the Lake Naivasha basin, Kenya

Meins, Frank Martijn (2013) Evaluation of spatial scale alternatives for hydrological modelling of the Lake Naivasha basin, Kenya.

[img] PDF
Abstract:To understand and predict the effects of anthropogenic interventions on the distribution of water, sediment and pollutants in a drainage basin hydrological models have been, and are still being developed. Most hydrological models use a water balance consisting of a change in water storage in a certain compartment over some time step as a function of rainfall, evapotranspiration, surface runoff and groundwater interactions. The difference between the available models lies in the way these components of the water balance are schematised. The time period over which processes related to the water balance occur ranges from minutes up to decades; while the areas in which they occur range from several square meters to thousands of square kilometres. Processes that might seem to behave in a certain manner at small scales might behave differently at larger scales, hence information obtained from experiments and observations at a small temporal or spatial scale cannot just be transferred directly to larger scales. Similarly large scale observations cannot directly be used for small scale simulations. This transfer of information from large scales to small scales and vice versa is called downscaling and upscaling respectively and problems associated with it are scale issues, which are studied in this research. The exact definition of scale used in this research is: “a characteristic time or length of a process, observation or model”. This refers to the difference in scale within one analytical dimension (e.g. millimetres, meters or kilometres). The focus is on spatial scale of model implementation. The objective is formulated as follows; “The objective of this study is to evaluate the effect of using different spatial scales for implementing a hydrological model of the upper Lake Naivasha basin, Kenya, on the accuracy of stream flow simulations” In this case ‘accuracy of stream flow simulation’ is defined as the agreement between observed and simulated monthly averaged stream flows (in m3/s). The study is performed by modelling the hydrology of the Malewa basin, which is a sub- basin of the Lake Naivasha basin, Kenya, and contributes approximately 80% to the surface runoff into Lake Naivasha. The hydrological model that was selected for modelling stream flows in this basin is the Soil Water Assessment Tool (SWAT). The model was considered to be suitable because once the data sets are prepared it is relatively easy to apply different spatial scales. SWAT divides a basin in sub-basins with each their own climate data and channel characteristics. For each of these sub-basins hydrological response units (HRUs) are then defined, which are areas with similar land use, soil and slope characteristics. 7 river gauging stations are available for the Malewa basin which can be used to calibrate the sub-basins. Combining these stations with the SWAT model structure resulted in the application of two types of spatial scales. Firstly three different basin delineations are be applied, with 1, 3 and 7 sub-basins that are generated based on the locations of the river gauging stations to ensure calibration of each sub-basin. Secondly multiple HRUs are applied using only one sub-basin that covers the entire Malewa basin. Additionally, sensitivity of the stream flow simulation to rainfall distribution was tested by applying a homogenous rainfall distribution to the case with 7 sub- basins. vi To test accuracy of stream flow simulation the Nash-Sutcliffe Efficiency (NSE) was calculated which explains correlation, bias and relative variability of simulated stream flow values as compared to observed stream flow values. Because of the poor data quality the NSE was calculated at a monthly time scale. When applying the three basin delineations mentioned before, the NSE of the most downstream basin outlet is higher for finer basin delineations. This means that when increasing the number of sub-basins in SWAT the accuracy with which stream flows are simulated increases. It must be noted that this only applies to the simulation of stream flows. Internal flows within the model such as surface runoff, lateral flow and groundwater flow were not included in the calibration procedure and in some cases assumed implausible values. Also, related to this, a number of model parameters adopted implausible values. When increasing the number of HRUs using only one sub-basin, no trend was observed in the accuracy with which stream flows are simulated. This can be attributed to a combination of two things. Firstly the effect of over-parameterization occurs more prominently when increasing the number of HRUs, because the number of parameters also increases with the number of HRUs while the number of variables used for calibration remains only one (the most downstream outlet). Secondly uncertainty in land use and soil data plays an important role when defining HRUs. Default SWAT parameters were used to represent the different land use types and the soil parameters used were uncertain, this introduces additional uncertainty in the resulting stream flows especially when the number of HRUs is increased. Because of these two things, improvements that were expected to occur when increasing the number of HRUs could not be observed. The model was found to be sensitive to rainfall and more specifically to the distribution of rainfall. This is because when applying homogenous rainfall to the case with 7 sub-basins, despite having the same rainfall sum, stream flows changed at the most downstream outlet. At sub-basin level rainfall sums did change when applying homogenous rainfall, which affected stream flow as well. In all cases, except for the most downstream one, a certain change in rainfall caused a much larger change in mean stream flow. This means that the model is very sensitive to changes in rainfall. It is concluded that a basin delineation with more sub-basins results in a more accurate simulation of stream flows when using SWAT. However, issues with data availability in combination with a large number of parameters used during calibration resulted in implausible internal model results despite good stream flow simulation results. This was especially observed when increasing the number of HRUs. Therefore, finer spatial scales of model implementation will improve accuracy of stream flow simulation, but only when data are available at the same spatial scale to ensure an accurate representation of the hydrological processes and to prevent over-parameterization by reducing the number of parameters that need to be calibrated.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Subject:38 earth sciences
Programme:Civil Engineering and Management MSc (60026)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page