Integration of Rain Gauge and Doppler Radar Data Using Bayesian Non-Parametric Approach
Abstract
Precipitation is an essential hydrologic process. Accurate representation of the spatial and temporal distribution of rainfall intensity is critical to development of robust calibrated hydrologic models. Rainfall data are commonly collected using rain gauges and Doppler radar. Rain gauge data are more accurate but they only yield information at point locations. Radar data provide continuous spatial information pixel-by-pixel but they are less accurate. This paper presents a Bayesian non-parametric approach for integrating gauge and radar data to develop more accurate and continuous rainfall interpretation. In a Bayesian framework, the resulting probability distribution of rainfall intensity (posterior distribution) at a location at a time step is computed from a prior distribution and a likelihood function. In this paper, the prior distribution is estimated by applying geostatistical methods to rain gauge data. The likelihood function is calculated based on the mismatch errors between the rainfall radar and rainfall gauge data where they overlap. A non-parametric approach allows rainfall spatial structures to be intensity dependent. At each time step, a range of rainfall threshold levels is considered. For each threshold level, rain gauge and radar data are encoded into indicator values with 1 denoting rainfall intensity greater than the threshold level. Radar data are used to characterize the correlation structure of the indicator field. Indicator Kriging using the resulting correlation model is applied to gauge indicator data to compute the prior estimate of the probability of exceeding the rainfall threshold. Fault table based on comparison of gauge and radar indicator values is used to compute the likelihood at a location. The resulting posterior estimate of the probability of exceeding the rainfall threshold represents the cumulative probability density function value corresponding to the rainfall threshold at the location. Available Rain gauge and radar data in southwest Florida were used to demonstrate this data integration approach and to benchmark its performance.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2012
- Bibcode:
- 2012AGUFM.H33K1480M
- Keywords:
-
- 1854 HYDROLOGY / Precipitation