Empirically-based modeling of radar-rainfall uncertainties

There are large uncertainties associated with radar-based rainfall estimates, including both systematic errors and random effects from numerous sources. Their propagation through all models for which radar-rainfall is used as input or for which is used as an initial condition is necessary to enhance our understanding and interpretation of the obtained results. Despite the relevance of the topic, there is no comprehensive statistical characterization of these errors.;The specific goals of this dissertation include: (1) development of an empirically-based radar-rainfall error model; (2) investigation of the sensitivity of the components of the error model to different radar setups, spatial and temporal resolutions, and geographic areas; (3) development of a generator of ensembles of probable true rainfall fields, conditioned on a given radar-rainfall map; (4) application of the error model to: (i) the flash flood forecasting problem; (ii) the evaluation of satellite-based rainfall products; and (iii) the impact of radar-rainfall uncertainties on the scaling properties of rainfall;;According to the approach proposed in this study, a realistic parameterization of the relationship between true rainfall and radar-rainfall can be achieved with a model described by two elements: a systematic distortion function and a random component. These two components are identified using a non-parametric approach, and rain gauges are used as an approximation of the true ground rainfall. This model has the flexibility to account for range from the radar, different spatio-temporal scales, rain regime, and space and time dependency of the errors. The results of this study are based on large samples of radar and rain gauge data from Oklahoma, United States, and south-west Great Britain.;One of the assumptions in this study is that the true ground areal rainfall can be approximated by rain gauge measurements, introducing sampling discrepancies between a radar pixel and a rain gauge. Therefore, the uncertainties associated with the approximation of an areal estimate with a point measurement (spatial sampling errors) will be investigated for different spatial and temporal scales by means of an empirically-based error model.