Skip to Main Content

The ANN technique was employed to develop a forecast model for each season. At the first stage, a trial and error method was used to select the best combination of input variables and the best network architecture. The best networks and their respective best variable combinations were selected according to the performance criteria: r and RMSE among 2n − 1 combination cases of ‘n’ input variables. Then, the best network was checked for a testing with ‘out of sample dataset’ (2008–2012). The best network architectures and corresponding combination cases for each nCr (where n is the total number of variables, and r is the number of variables used for each combination case) have been summarized for the case of ASO rainfall forecast in Table 3. The selected networks for further study are in bold letters. Following the same procedure, the selected networks and predictors for each season of rainfall were obtained and summarized in Table 4. The proposed models for MJJ and FMA are able to forecast one month prior to each season and the models for ASO and NDJ are able to do so two months prior to each season according to the lag time of the best predictors.

Table 3

Best combination cases as an example for ASO rainfall season

Training
Testing
nCrInput variablesHidd. N.RMSE/mmrRMSE/mmr
SLP(t−8) 67 0.51 37 0.68 
SLP(t−8)SYW(t−9) 13 52 0.74 59 0.80 
SLP(t−8)SLP(t−13)RH(t−5) 50 0.77 70 0.80 
SLP(t−8)SLP(t−13)RH(t−5)SYW(t−9) 41 0.75 38 0.75 
5 SLP(t−8)SLP(t−13)SLP(t−10)RH(t−5)SYW(t−9) 9 32 0.86 68 0.83 
SLP(t−8)SLP(t−13)SLP(t−10)RH(t−5)SYW(t−9)SAT(t−12) 14 58 0.71 64 0.77 
Training
Testing
nCrInput variablesHidd. N.RMSE/mmrRMSE/mmr
SLP(t−8) 67 0.51 37 0.68 
SLP(t−8)SYW(t−9) 13 52 0.74 59 0.80 
SLP(t−8)SLP(t−13)RH(t−5) 50 0.77 70 0.80 
SLP(t−8)SLP(t−13)RH(t−5)SYW(t−9) 41 0.75 38 0.75 
5 SLP(t−8)SLP(t−13)SLP(t−10)RH(t−5)SYW(t−9) 9 32 0.86 68 0.83 
SLP(t−8)SLP(t−13)SLP(t−10)RH(t−5)SYW(t−9)SAT(t−12) 14 58 0.71 64 0.77 
Table 4

Summary of selected networks for further study

Season of rainfallNetwork architecture (in-hidden-out)Selected combination case
FMA 5–20–1 SYW(t−8)PW(t−4)SLP(t−6)SLP(t−12)SXW(t−14) 
MJJ 4–19–1 SYW(t−12)SXW(t−4)SYW(t−10)SAT(t−12) 
ASO 5–9–1 SLP(t−8)SLP(t−13)SLP(t−10)RH(t−5)SYW(t−9) 
NDJ 4–8–1 SXW(t−6)SYW(t−11)SXW(t−5)SYW(t−10) 
Season of rainfallNetwork architecture (in-hidden-out)Selected combination case
FMA 5–20–1 SYW(t−8)PW(t−4)SLP(t−6)SLP(t−12)SXW(t−14) 
MJJ 4–19–1 SYW(t−12)SXW(t−4)SYW(t−10)SAT(t−12) 
ASO 5–9–1 SLP(t−8)SLP(t−13)SLP(t−10)RH(t−5)SYW(t−9) 
NDJ 4–8–1 SXW(t−6)SYW(t−11)SXW(t−5)SYW(t−10) 

Close Modal

or Create an Account

Close Modal
Close Modal