%0 Articles
%T Estimating individual tree growth using non-parametric methods
%A Sironen, Susanna
%D 2009
%J Dissertationes Forestales
%V 2009
%N 94
%R doi:10.14214/df.94
%U http://dissertationesforestales.fi/article/1877
%X Information about the current condition, extent and quantity of forests that is provided by forest inventories, combined with forest growth models, is of great importance in forecasting the future development of forests. The ability to make reliable predictions has an important role as a tool of management planning, in evaluating silvicultural options, and ensuring sustainable forest management. In Finland, growth models are typically national models which may be markedly biased for a given stand or region. Non-parametric methods offer an alternative to the traditional regression methods. In non-parametric methods, the value of the variable of interest for a target observation is estimated often as a weighted average of the values of neighbouring reference observations, which are similar to the target observation in terms of the independent variables and weighted by their proximity to the target observation Locality can easily be described by non-parametric methods, if local data is available. The overall purpose of this thesis was to examine different non-parametric methods as a method for individual tree growth estimation. One of the main focuses was to test non-parametric methods in order to reduce the regional biases associated in the growth estimates.

The study material comprised temporary local sample plot data from Kuusamo in north-eastern Finland and nationwide permanent inventory growth plot data (INKA). The tree species considered were Scots and Norway spruce. The tested methods included k-nearest neighbour methods and generalized additive models. The different topics analysed in this thesis include local non-parametric growth estimation methods, localizing the non-parametric growth estimates, simultaneous estimation of individual tree diameter and height increment, and the effects of correlated observations on non-parametric growth estimation methods.

The results showed that non-parametric methods are suitable for estimation of growth, although the performance of the different methods varied depending on the purpose and the data used. The non-parametric methods were capable of reducing the regional biases. The most promising alternative to the means of localization was the sub-setting of the reference data by selecting the neighbours from a circle around the target tree. The levels of accuracy achieved in the estimation of individual tree growth were at least as good as those obtained by the parametric models at the tree, stand and regional levels. The methods presented in this thesis could be implemented in practical planning systems, although several issues still require further study and consideration, especially the issues concerning silvicultural treatments.