JP1.9
Predicting rainfall in the Lake Okeechobee watershed based on historic rainfall records and climate indicators: an Artificial Neural Network modeling approach

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Monday, 30 January 2006
Predicting rainfall in the Lake Okeechobee watershed based on historic rainfall records and climate indicators: an Artificial Neural Network modeling approach
Exhibit Hall A2 (Georgia World Congress Center)
Reinaldo Garcia-Martinez, Univ. of Miami, Coral Gables, FL; and S. Ahmad and F. Miralles-Wilhem

Lake Okeechobee (LO) is the main source of fresh water used for agriculture and urban water supply in South Florida. The prediction of the quantity and quality of water into the lake is very important for water resources management. We present an Artificial Neural Network (ANN) model to predict rainfall in the LO watershed, based on historic rainfall data and climatic indices. The predicted rainfall will be used as an input to a rainfall-runoff model to predict water discharges into the lake.

The ANN model uses a multilayer feed-forward neural network architecture and has been trained with available rainfall data, and various combinations of climatic related indices such as the Southern Oscillation Index (SOI), the Atlantic Ocean Thermohaline index and the Atlantic Multi-decadal Oscillation. We have tested a number of ANN architectures and input data sets to train the network. The resulting network structure includes input data for rainfall and climatic indices for the previous three to six months. Output from the ANN model is the cumulative rainfall for next month or for next three months. The ANN development was tested by separating the historic data records in two periods: 1933-1983 for training and 1984-2004 for validation. Using previous months rainfall data, the model was able to satisfactorily predict next month rainfall with an in the range of 15-20% RMS error.

Links of El Niņo and La Niņa phenomena with rainfall patterns in the Florida peninsula have been previously studied. However, the cause-effect relationship is not always a clear one. We also evaluated the influence of climate anomalies like El Niņo and La Niņa events on rainfall in the Lake Okeechobee watershed. The ANN model results indicate that rainfall is mostly influenced by the effect of combined indices.

We are presently extending the model to incorporate prediction of water quality of the inflow to the LO. It is expected that the model water quality and rainfall predictions will be valuable for LO managers to assess the impact of climate anomalies in proposed operational schedules that are being evaluated.