Artificial neural networks modelling of streamflow and water quality in ungauged watersheds: Investigating the potential of remote sensing information

Most of the currently available models for watershed modelling are limited in practice because of the extensive requirement for landscape data. A class of models that can simulate the response of ungauged watersheds, without being ground-based data collection and time intensive, was developed to provide the necessary information for responsive watershed management practices.; A class of watershed models that are less reliant on ground-based measurements by using remote sensing (RS) information instead was devised. The focus was on formulating streamflow (Q) and total phosphorus (TP) concentrations models, which are only reliant on the currently available meteorological information, as well as publicdomain free-of-cost Moderate Resolution Imaging Spectroradiometer (MODIS)-derived RS information. A number of Q and TP models were devised and applied to a number of watersheds (5 to 130 km2 in basin area). The thesis presented: (1) the first effort to compare autoregressive moving average with exogenous input (ARMAX) modelling to artificial neural network (ANN) modelling for TP predictions and confirmed that the ANN approach is superior to the ARMAX in modelling time-correlated gapped data; (2) a step-by-step guideline to ANN modelling of time-correlated variables that can account for data hystereses; (3) an ANN modelling algorithm that relies only on lowcost, readily available meteorological data, and careful time series manipulation prior to model building for Q predictions and, thus, is suitable for modelling streamflow in ungauged watersheds; (4) a new remotely-sensed hydrologic similarity measure that provided a successful indicator of basins similarity; (5) a successful modelling algorithm that can rely on a dynamic suite of RS vegetation indices (VIs) for predicting TP concentrations and; (6) the first attempt to address the impact of watershed subdivision on a water quality parameter using an ANN modelling technique.; The results from this exercise demonstrated the applicability of the ANN modelling approach, and the usefulness of the MODIS-derived VIs in simulating Q and TP dynamics. Such models can potentially serve as valuable tools for watershed-scale forest management.