Monday, 7 January 2013: 11:00 AM
Room 18A (Austin Convention Center)
Handout (4.6 MB)
Rainfall in the Sahel region of Africa is prone to large interannual variability and has exhibited a recent multidecadal drying trend. This variability is of scientific interest, and it is also a matter of considerable humanitarian importance: the social and environmental impacts of precipitation anomalies in the Sahel are dramatic and well-documented. Recognizing the need for improved understanding and prediction of Sahelian precipitation variability, a number of researchers have proposed statistical models of seasonal precipitation anomalies. These models are based on the principle that atmospheric processes relevant to seasonal precipitation are influenced by anomalies in large scale patterns of sea surface temperature (SST), surface pressure, surface air temperature (SAT), and other variables, and that this influence exists with enough lead time to inform useful seasonal predictions. These statistical models have demonstrated some skill in predicting precipitation, but nearly all have adopted conventional statistical modeling techniques—most commonly generalized linear models—to associate predictor fields with precipitation anomalies. Here, we present the results of an artificial neural network (ANN) machine learning algorithm applied to predict summertime (July-September) Sahel rainfall anomalies on the basis of springtime (April-June) SST and SAT anomalies for the period 1900-2011. The regularization of the ANN was based on weight decay method. Principal component analysis (PCA) was used to remove multicollinearity between predictor variables, and predictive accuracy was assessed using repeated k-fold random holdout and leave-one-out cross-validation methods. It was found that the ANN achieved predictive accuracy superior to that of eight alternative parametric and non-parametric statistical methods tested in this study, and also superior to that of previously published predictive models of summertime Sahel precipitation. These results point to the value of ANN techniques for seasonal precipitation prediction in the Sahel, and they suggest that ANN may be of use for seasonal prediction in other regions of high rainfall variability as well.
Supplementary URL: http://blaustein.eps.jhu.edu/~zaitchik
![](images/flag.png)
![](images/blue_ribbon.png)