Smoothed Prediction of the Onset of Tree Stem Radius Increase Based on Temperature Patterns
Mikko Korpela, Harri Mäkinen, Mika Sulkava, Pekka Nöjd and Jaakko Hollmen
Proceedings of the 11th International Conference on Discovery Science (DS-2008)
Lecture Notes in Artificial Intelligence
ISBN 978 3 540 88410 1
Possible changes of the growing season of trees would have significant consequences on forest production. Predicting the onset of tree growth on the basis of climate records can be used for estimating the magnitude of such changes. Conventional methods for estimating the onset of tree growth use cumulative temperature sums. These estimates, however, are quite coarse, and raise questions about making better use of the weather information available. We approach the problem of predicting the onset of tree growth with a predictor based on a combination of a k-nearest neighbor regressor and a linear regressor. The inputs are weighted sums of daily temperatures, where the weights are determined by a subset of Bernstein polynomials chosen with a variable selection methodology. The predictions are smoothed for consecutive days to give more accurate results. We compare our proposed solution to the more conventional approach. The proposed solution is found to be better.