Real Time Rainfall Estimation From Satellite and Model Data Using Artificial Neural Networks
Abstract
In this work a case-study of an application of Artificial Neural Networks (ANN) to real time monitoring of rainfall is presented. The Zambia region is chosen where rain- fall estimation from satellite imagery is the only feasible approach. The ANN inputs provide different parameters and the weighting assigned to each parameter is deter- mined by a training process. During the training the observed parameters are compared with the desired output of many different examples. The novel feature of this work is the combined use of Meteosat Thermal Infra-Red imagery and model fields from the ECMWF forecast model and the long term aim is to produce daily rainfall estimates for operational hydrological forecasting on a river catchment scale. The input data of the ANN are the codified satellite information and the PC (principal components) of the humidity field, surface horizontal wind and vertical velocity at different heights. The output is verified using gauge data from African stations. The ANN method is compared with a more standard one used at the moment in those regions.
- Publication:
-
EGS General Assembly Conference Abstracts
- Pub Date:
- 2002
- Bibcode:
- 2002EGSGA..27.1701C