I have a long term times series of rainfall and NDVI, for many administrative units and I want to use them to improve crop yield forecasts, assuming that the crop yield regression models should a priori be related.
Dear Riad, i didn't get your question as enough, but as you explained I assume you have 5 factors for analysis which all are continuous. Rainfall, NDVI, time, and your spatial coordinates. If and if the area under study is not to large you can use multiple regression, however, some of your relationships seem non-linear. Besides, if the time series are too long you may see a trend in rainfall data which should be removed.