Local rainfall modelling based on global climate information: A data-based approach

Modelling climate is complex due to multi-scale interactions and strong nonlinearities. However, climate signals are typically quasi-periodical and are likely to depend on exogenous-variables. Motivated by this insight, we propose a strategy to circumvent modelling complexity based on the following ideas. 1) The observed signals can be decomposed into non-stationary trends and quasi-periodicities through Dynamic-Harmonic-Regressions (DHR). 2) The main-frequencies and decomposed signals can be used for constructing a harmonic model with varying parameters depending on exogenous-variables. 3) The State-Dependent-Parameter (SDP) technique allows for the dynamical estimation of these parameters. The resulting DHR-SDP combined approach is applied to rainfall-monthly modelling, using global-climate signals as exogenous-variables. As a result, 1) the model yields better predictions than standard alternative techniques; 2) the model is robust regarding data limitations and useful for several-steps-ahead forecasting; 3) interesting relations between global-climate states and the local rainfall's seasonality are obtained from the SDP estimated functions.