On the use of neural networks for dendroclimatic reconstructions

This paper investigates if the use of neural networks can improve the accur acy of tree-ring based paleoclimatic reconstructions with respect to some o f the commonly used methods. A three layers feedforward model of neural net work is shown to be very efficient in explaining a high percentage of the v ariance of the instrumental climatic record both for calibration and valida tion. Some traditional statistics have been estimated to evaluate the accur acy of the reconstruction. These results have been finally compared with th ose of a regression-based model, showing the higher accuracy of the neural network reconstruction.