Analyzing the effects of stand thinning on microclimates with semiparametric smoothing splines

• Published in: Canadian journal of forest research Vol. 36; no. 7; pp. 1641 - 1648
• Main Authors: Guan, B.T, Weng, S.H, Kuo, S.R, Chang, T.Y, Hsu, H.W, Shen, C.W
• Format: Journal Article
• Language: English
• Published: Ottawa, Canada NRC Research Press 01.07.2006; National Research Council of Canada; Canadian Science Publishing NRC Research Press
• Subjects: Analysis; Biological and medical sciences; Environmental monitoring; equations; forest ecology; Forest management; forest stands; forest thinning; forest trees; Forestry; Forests; Fundamental and applied biological sciences. Psychology; Growth; Microclimate; monitoring; Objectives; Plant populations; Spline theory; statistical models; Studies; Thinning; Trees; Variables; Taiwan; Forestry
• ISSN: 0045-5067; 1208-6037
• DOI: 10.1139/x06-057

Monitoring the effects of stand thinning on microclimates is an integral part of any thinning experiment. It is through its modifications of microclimates that thinning alters important ecological processes. An efficient analysis of microclimate-monitoring data should address both the effects of thinning regimes on, and the temporal response trends of, microclimates. Probably because of the difficulties in modeling temporal trends parametrically, an examination of the existing literature on thinning showed that only a few studies have attempted to address the second aspect. We propose the use of semiparametric smoothing splines to analyze monitoring data from thinning experiments. First, the concept of a smoothing spline is briefly described. We then provide an example in which semiparametric mixed-effects smoothing-spline models were used to analyze microclimate-monitoring data from a thinning experiment. The proposed approach not only successfully detected the effects of thinning, but also revealed interesting temporal trends. For each of the microclimatic variables, we also compared the performance of the fitted semiparametric model with that of a parametric model. In general, the semiparametric model performed better than its parametric counterpart. We also addresse some concerns in using the proposed approach.