[Truncated abstract] Generally hydrologic and ecologic models operate on arbitrary time and space scales, selected by the model developer or user based on the availability of field data. In reality rainfall is highly variable not only annually, seasonally and monthly but also the intensities within a rainfall event and infiltration properties on semi-arid hillslopes can also be highly variable as a result of discontinuous vegetation cover that form mosaics of areas with vegetation and areas of bare soil. This thesis is directed at improving our understanding of the impacts of the temporal representation of rainfall and spatial heterogeneity on model predictions of hydrologic thresholds and surface runoff coefficients on semi-arid landscapes at the point and hillslope scales. We firstly quantified within storm rainfall variability across a climate gradient in Western Australia by parameterizing the bounded random cascade rainfall model with one minute rainfall from 15 locations across Western Australia. This study revealed that rainfall activity generated in the tropics had more within storm variability and a larger proportion of the storm events received the majority of rain in the first half of the event. Rainfall generated from fontal activity in the south was less variable and more evenly distributed throughout the event. Parameters from the rainfall analysis were then used as inputs into a conceptual point scale surface runoff model to investigate the sensitivity of point scale surface runoff thresholds to the resolution of rainfall inputs. This study related maximum infiltration capacities to average storm intensities (k*) and showed where model predictions of infiltration excess were most sensitive to rainfall resolution (ln k* = 0.4) and where using time averaged rainfall data can lead to an under prediction of infiltration excess and an over prediction of the amount of water entering the soil (ln k* > 2). For soils susceptible to both infiltration excess and saturation excess, total runoff sensitivity was scaled by relating drainage coefficients to average storm intensities (g*) and parameter ranges where predicted runoff was dominated by infiltration excess or saturation excess depending on the resolution of rainfall data were determined (ln g* <2). The sensitivity of surface runoff predictions and the influence of specific within storm properties were then analysed on the hillslope scale. '...' It was found that using the flow model we still get threshold behaviour in surface runoff. Where conditions produce slow surface runoff velocities, spatial heterogeneity and temporal heterogeneity influences hillslope surface runoff amounts. Where conditions create higher surface runoff velocities, the temporal structure of within storm intensities has a larger influence on runoff amounts than spatial heterogeneity. Our results show that a general understanding of the prevailing rainfall conditions and the soil's infiltration capacity can help in deciding whether high rainfall resolutions (below 1 h) are required for accurate surface runoff predictions. The results of this study can be considered a contribution to understanding the way within storm properties effect the processes on the hillslope under a range of overall storm, slope and infiltration conditions as well as an improved understanding of how different vegetation patterns function to trap runoff at different total vegetation covers and rainfall intensities.

[Truncated abstract] During the last fifty years mathematical models of catchment hydrology have been widely developed and used for hydrologic forecasting, design and water resources management. Most of these models need large numbers of parameters to represent the flow generation process. The model parameters are estimated through calibration techniques and often lead to ‘unrealistic’ values due to structural error in the model formulations. This thesis presents a new strategy for developing catchment hydrology models for representing streamflow and salinity generation processes. The strategy seeks to ‘learn from data’ in order to specify a conceptual framework that is appropriate for the particular space and time scale under consideration. Initially, the conceptual framework is developed by considering large space and time scales. The space and time scales are then progressively reduced and conceptual model complexity systematically increased until ultimately, an adequate simulation of daily streamflow and salinity is achieved. This strategy leads to identification of a few key physically meaningful parameters, most of which can be estimated a priori and with minimal or no calibration. Initially, the annual streamflow data from ten experimental catchments (control and cleared for agriculture) were analysed. The streamflow increased in two phases: (i) immediately after clearing due to reduced evapotranspiration, and (ii) through an increase in stream zone saturated area. The annual evapotranspiration losses from native vegetation and pasture, the ‘excess’ water (resulting from reduced transpiration after land use change), runoff and deep storage were estimated by a simple water balance model. The model parameters are obtained a priori without calibration. The annual model was then elaborated by analysing the monthly rainfall-runoff, groundwater and soil moisture data from four experimental catchments. Ernies (control, fully forested) and Lemon (53% cleared) catchments are located in zone with a mean annual rainfall of 725 mm. Salmon (control, fully forested) and Wights (100% cleared) are located in zone with mean annual rainfall of 1125 mm. Groundwater levels rose and the stream zone saturated area increased significantly after clearing. From analysis of this data it was evident that at a monthly time step the conceptual model framework needed to include a systematic gain/loss to storage component in order to adequately describe the observed lags between peak monthly rainfall and runoff.

[Truncated abstract] The use of uncertainty analysis is gaining considerable attention in catchment hydrological modeling. In particular, the choice of an appropriate model structure, the identifiability of parameter values, and the reduction of model predictive uncertainty are deemed as essential elements of hydrological modelling. The chosen model structure must be parsimonious, and the parameters used must either be derivable from field measured data or inferred unambiguously from analysis of catchment response data. In this thesis, a long-term water balance model for the Susannah Brook catchment in Western Australia has been pursued using the ?downward approach?, which is a systematic approach to determine the model with the minimum level of complexity, with parameter values that in theory are derivable from existing physiographic data relating to the catchment. Through the analysis of the rainfall-runoff response at different timescales, and the exploration of the climate, soil and vegetation controls on the water balance response, an initial model structure was formulated, and a priori model parameter values estimated. Further investigation with the use of auxiliary data such as deuterium concentration in the stream and groundwater level data exposed inadequacies in the model structure. Two more model structures were then proposed and investigated through formulating alternative hypotheses regarding the underlying causes of observed variability, including those associated with the absence of a contribution of deep groundwater flow to the streamflow, which was indicated by deuterium concentration and internal dynamics characterized by the observed groundwater levels. ... These differences are due to differences in the time delay between rainfall and recharge between upland and riparian regions. The ages of water recharging the groundwater and discharging from the catchment were estimated by assuming a piston flow mechanism. In the deeper, upland soils, the age of recharging water was considerably larger than the unsaturated zone delay would suggest; a recharge response 16 days after an infiltration event may involve water as much as 160 days old. On the other hand, the delay and the age of recharging water were much lower in the shallow riparian zone. Where the upland zone contributes significantly to discharge, the predicted difference between the rainfall-discharge response time and the average age of discharging water can be significant.