The application of artificial intelligence for monthly rainfall forecasting in the Brisbane Catchment, Queensland, Australia

Abstract

Concurrent relationships between climate indices and Australian spring rainfall have been used extensively to explain weather events. In order for climate indices to be useful for rainfall forecasting there must be relationships between their lagged values and rainfall. The methods currently used by the Australian Bureau of Meteorology for seasonal weather forecasting have limited capacity to exploit the often non-linear relationships that potentially exist between the lagged values for these indices and rainfall. This paper reports on the application of a method of forecasting based on the use of neural networks, a form of artificial intelligence. Neural networks facilitate the input of multiple variables simultaneously. The variables most useful for determining rainfall are elucidated by application of algorithms during the optimisation process.Brisbane, the capital of Queensland, Australia, has flooded periodically and catastrophically. The neural network described in this study was used to forecast rainfall for three locations in the Brisbane River catchment one to three months in advance, including the 2011 flood event. Results are compared on the basis of root mean square error with output from the Australian Bureau of Meteorology's general circulation model, POAMA. The Neural Network model shows considerable more skill. The Neural Network incorporates lagged values for key climatic indices and also rainfall and atmospheric temperatures.

Improving monthly rainfall forecasts using artificial neural networks and single-month optimisation: a case study of the Brisbane catchment, Queensland, Australia

“…Two approaches were used for ANN optimization. With the first approach, designated as "all-month optimization", data for all 12 months of the year was included as input and optimised together, as in our previous studies [18][19][20][21]. With the second approach, designated as "single month optimisation", forecasts corresponding to each calendar month were performed individually, so that 12 optimisations were carried out to produce monthly rainfall forecasts for the entire year.…”

“…ANNs can accommodate non-linear relationships, and test multiple inputs, particularly important when the influence of climate indices may vary geographically and temporarily in poorly understood ways [13]. ANNs have been applied to produce seasonal and monthly rainfall forecasts in many parts of the world [15][16][17], including Australia [18][19][20][21].…”

“…Our first attempt at using ANN models to forecast rainfall at two key locations in the Brisbane catchment was reported at the Seventh River Basin Management Conference [21]. In the present study, the ANN approach has been extended to investigate the application of a novel method of optimising the ANN models, in particular instead of optimising a given set of inputs for all months simultaneously, the program is provided with data for only a single month, i.e.…”

The potential benefits of using artificial intelligence for monthly rainfall forecasting for the Bowen Basin, Queensland, Australia

“…Other studies of flood forecasting, specific to south-east Queensland, suggest that the neural network technique can be easily extended to a forecast with a three month lead-time (i.e. a forecast three months in advance) without a significant deterioration in skill [11].…”

Using lagged and forecast climate indices with artificial intelligence to predict monthly rainfall in the brisbane catchment, Queensland, Australia

“…In other words, when using statistical models, it is only relationships between future rainfall and lagged values of climate indices, or rainfall and future predicted values of climate indices that can practically serve as valid model input. This was explained in a conference paper [10] that was the catalyst for the development of the method for independently forecasting climate indices used in this study.…”