Monthly rainfall forecasting using temperature and climate indices through a hybrid method in Queensland, Australia

View More View Less
  • 1 Department of Civil and Construction Engineering, Swinburne University of Technology, Melbourne, VIC 3122, Australia.
© Get Permissions
Restricted access

Abstract

This paper presents applications of wavelet artificial neural networks (WANN) to forecast rainfalls one, three, six, and twelve months in advance using lagged monthly rainfall, maximum, minimum temperatures, Southern Oscillation Index (SOI), Inter-decadal Pacific Oscillation (IPO), and Nino3.4 as predictors. Eight input datasets comprised of different combinations of predictive variables were used for ten candidate climate stations in Queensland, Australia. Datasets were split as 1908 to 1999 for the training of the model and 2000 to 2016 for the verification of the model. Also, the conventional Artificial Neural Network (ANN) model was developed with the same input datasets to compare with WANN results. Moreover, the skillfulness of the WANN was investigated with the current climate prediction system used by the Australian Bureau of Meteorology (BOM), Australian Community Climate Earth-System Simulator–Seasonal (ACCESS–S) as well as climatology forecasts. The comparisons showed that the WANN achieved the lowest errors for three-month lagged prediction with an average Root Mean Square Error (RMSE) of 38.6mm. In contrast, for the same lag-period, the average RMSEs from ANN, ACCESS-S, and climatology predictions were 72.2mm, 102.7mm, and 72.2mm, respectively. It is also found that the ANN underestimates the peak values with an average value of 49%, 47%, 52%, and 53% at one, three, six, and twelve months lead times, correspondingly. However, the corresponding peak values underestimation through the WANN were 0%, 1%, 22%, and 39%, respectively. This research provides promising insights into using hybrid methods for predicting rainfall a few months in advance, which is extremely beneficial for Australia’s agricultural industries.

Abstract

This paper presents applications of wavelet artificial neural networks (WANN) to forecast rainfalls one, three, six, and twelve months in advance using lagged monthly rainfall, maximum, minimum temperatures, Southern Oscillation Index (SOI), Inter-decadal Pacific Oscillation (IPO), and Nino3.4 as predictors. Eight input datasets comprised of different combinations of predictive variables were used for ten candidate climate stations in Queensland, Australia. Datasets were split as 1908 to 1999 for the training of the model and 2000 to 2016 for the verification of the model. Also, the conventional Artificial Neural Network (ANN) model was developed with the same input datasets to compare with WANN results. Moreover, the skillfulness of the WANN was investigated with the current climate prediction system used by the Australian Bureau of Meteorology (BOM), Australian Community Climate Earth-System Simulator–Seasonal (ACCESS–S) as well as climatology forecasts. The comparisons showed that the WANN achieved the lowest errors for three-month lagged prediction with an average Root Mean Square Error (RMSE) of 38.6mm. In contrast, for the same lag-period, the average RMSEs from ANN, ACCESS-S, and climatology predictions were 72.2mm, 102.7mm, and 72.2mm, respectively. It is also found that the ANN underestimates the peak values with an average value of 49%, 47%, 52%, and 53% at one, three, six, and twelve months lead times, correspondingly. However, the corresponding peak values underestimation through the WANN were 0%, 1%, 22%, and 39%, respectively. This research provides promising insights into using hybrid methods for predicting rainfall a few months in advance, which is extremely beneficial for Australia’s agricultural industries.

Save