Real-time estimation of rainfall: A dynamic spatio-temporal model
Williams, Talithia D.
Ensor, Katherine B.
Doctor of Philosophy
Real-time estimation of rainfall and the resulting runoff is the primary task of Flood Alert Systems (FAS). Urban areas such as Houston, rely on FAS for intelligent management of frequently occurring floods. Estimation of rainfall is a critical component of an effective Flood Alert Systems (FAS). This thesis develops a dynamic spatio-temporal state space model for improved strategies of rainfall estimation. Typical devices used to measure rainfall, namely rain gauges and weather radars, are prone to measurement error and systematic bias. Problems often ignored in current methods are addressed in our technology, namely, adjusting for storm intensity while accounting for spatial and temporal nonstationarity. We develop a state-space model consisting of a spatio-temporal model for the underlying rainfall process coupled with observation equations for both the rainfall and radar observations series. A dynamic space-time Kalman filter in conjunction with the estimation-maximization algorithm allows us to obtain parameter estimates for the process. Furthermore, our approach addresses the change of support issues which arise between the point observations of rainfall represented by the gauges and the block estimates of rainfall represented by the radar. A result of our modeling efforts is improved prediction of rainfall as well as improved estimation of the systematic bias in the radar observations.
Statistics; Environmental science