Skip to Main Content

The selection of input variables or independent variables (x1, x2, x3, …, xn) is a fundamental step in the development of data-driven models like ANN. All independent variables may not have the same level of importance, because some of them may be correlated with others or noisy or possess no predictive powers (Bowden et al. 2005). The importance of variable selection and selection methodologies for ANN models have been summarized by Maier & Dandy (2000), Guyon & Elisseeff (2003), Bowden et al. (2005), May et al. (2011), Nourani & Parhizkar (2013) and Nourani et al. (2015). The most commonly used methods are variable ranking method, mutual information, Gamma test, self-organizing map, and ANN modeling. For this study, the variable ranking method was adopted. Table 1 shows that 88 potential independent variables relevant for this study were tested for each season of rainfall. The lead time was defined as the time difference in months between the last month of the forecasted season and the last month of the independent variable's season. For example, if SST(t−4) is a predictor for MJJ, it means the season for SST is JFM (January–February–March) and the lead time is four months. As a result, MJJ can be forecasted at the beginning of April, which is one month prior to the MJJ season.

Table 1

Potential independent variables for each season of rainfall

Season of rainfallMJJ (t), ASO (t), NDJ (t), FMA (t)
Independent variables (xSAT(t−a), SLP(t−a), SXW(t−a), SYW(t−a), PW(t−a), RH(t−a), SST(t−a), Rainfall(t−b) 
Lead time of independent variable (month) a = 4, 5, 6, 7……………, 15 
b = 6, 9, 12, 15 
Season of rainfallMJJ (t), ASO (t), NDJ (t), FMA (t)
Independent variables (xSAT(t−a), SLP(t−a), SXW(t−a), SYW(t−a), PW(t−a), RH(t−a), SST(t−a), Rainfall(t−b) 
Lead time of independent variable (month) a = 4, 5, 6, 7……………, 15 
b = 6, 9, 12, 15 

Close Modal

or Create an Account

Close Modal
Close Modal