Hydrological simulation aided by numerical weather prediction model

In many water resources and hydrological projects, it is not always possible to get access to in-situ long-term time series weather measurements, especially for ungauged catchments. Even with gauged catchments, it is common that only rain gauge and river level data are available; other weather variables such as solar radiation, wind speed, surface temperature, surface air pressure and relative humidity are usually missing and if available are generally not in continuous form. These weather variables are basic building blocks of the global hydrological cycle that includes evapotranspiration (ET 0) and runoff estimation. The ET 0 and runoff can be estimated from the Penman-Monteith equation and rainfall runoff modeling respectively. This thesis explored a potential application of downscaled global reanalysis weather data using Pennsylvania State University-National Center for Atmospheric Research Mesoscale Model 5 (MMS). MMS is able to downscale the global weather data down to a much finer resolution in space and time for use in local hydrological investigations. The exploration of downscaling the ERA-40 reanalysis data to the Brue catchment in Southwest England and the assessment of the relevant weather variables in comparison with those measured at the ground was described in the thesis. However, there is a problem in using these selected weather variables in hydrological processes due to uncertainties obtained from the mesoscale modelling. Therefore, this thesis focused on the improvement of the weather variables from the dynamical downscaling and statistical modeling. The improvement of dynamic downscaling with the MMS cumulus parameterization schemes (CPSs) by changing the horizontal and vertical resolutions are presented in this thesis for rainfall estimation. Meanwhile, the error correction with statistical models is an attempt to hybridize MMS with two regression models ( the multiple linear regression (MLR) and the nonlinear regression (NLR)) and two artificial intelligence systems (the artificial neural networks (ANNs) and the support vector machines (SVMs)). This exploration is to tackle the errors between the MMS downscaled and observed data in addition to other MMS derived hydro- meteorological parameters. The hold-out validation with a forward selection method was employed as an input variable selection procedure to examine the model generalization errors in these statistical models. Upon the implementation of the error correction technique of weather variables, a comparative study of runoff simulation via the PDM model was completed between the MMS downscaled, corrected and observed data. This thesis also presents a sensitivity analysis of six weather variables to ET 0 estimation and runoff simulation through various combinations of the Penrnan-Monteith equation and Probability Distributed Model (PDM}inputs. Finally, by this assessment of several case studies in this thesis, it has shown that the enhanced MMS modeling scheme with the correction approaches substantially improves the forecasted weather variables over the study area which is important for the hydrological processes.EThOS - Electronic Theses Online ServiceGBUnited Kingdo