Deep learning neural network: A machine learning approach for monthly rainfall forecast, case study in eastern region of Thailand

Accurate monthly rainfall forecasting is essential for efficient watershed management, particularly for the current situation with high variation of rainfall due to global climate change. A variety of researchers attempted to develop more sophisticated models to enhance model capability to capture uncertainty due to high variation in rainfall both in time and space. The objective of this study is to investigate capability of a Deep Learning Neural Network (DNN) in forecasting monthly rainfall. A river basin in eastern region of Thailand, where a high increase in water demand is expected in next 20 years due to the national development plan, is selected as the study area. In this study LAV with different atmospheric layers, such as air temperature, geopotential height, meridonal wind, omega, outgoing longwave radiation, relative humidity, specific humidity, sea level pressure, sea surface temperature, zonal wind, precipitation rate and precipitable water, were selected as inputs to the DNN model. Monthly rainfall at Pluak Deang station from 1991 to 2010 were used for the training process in the DNN model. Monthly rainfall from 2011 to 2016 were used for model validation. Results of forecasting revealed that DNN is able to predict monthly rainfall from one month up to 12 months in the future, however, accuracy of forecasting decreases when the forecast time horizon increases. The most practical time of forecast is one month into the future yielding a forecast where around 70% of the forecasted values are within the range of one standard deviation from the observed values.

Abstract licence: CC BY-NC-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.