Identifying spatial and temporal profiles from forest nutrition data : Master's thesis, Helsinki University of Technology, Department of Engineering Physics and Mathematics

Helsinki

2003

70 S.

Monographie

114274

In this study, the nutrient concentrations of pine and spruce needles were analyzed using different data analysis methods. The data was collected from Finland and Austria in 1987-2000. The tested analysis methods included a few spatial statistics methods: semivariograms and interpolation, clustering of the self-organizing map with some simple temporal analysis of the clusters and actual time series modeling with the hidden Markov model. The aim of the study was to analyze the spatial and temporal distribution of the nutrient concentrations and simply try to find out what kind of internal structure there is in the data and how the different data analysis methods perform with this kind of data. It was found that semivariance, a spatial statistic, is a reasonably usable measure for analyzing the factors that affect the nutrient concentrations on a local scale. With the data used in this study, the semivariograms showed some trends in the similarity of nearby stands. The problem with the graphs was that they were rather noisy and therefore not very easy to interpret. It was also noted that interpolation of the measurements makes it possible to draw figures that are both visually appealing and help understand the geographical structure of the data. The hidden Markov model used in the time series analysis did not yield much information about the temporal structure of the data. When compared to the clustering method, the results of time series modeling were clearly inferior. The classification of the data into two groups did not give any especially interesting information about the connections between the measurements. The two states were simply so similar to each other that no remarkable conclusions could be drawn considering the possible source that could have generated the measurements. Apparently, the basic hidden Markov model was not the optimal time series model to be used with this kind of data. The VS clustering algorithm based on the self-organizing map provided new information about the relations of the nutrients between different years and locations. With the clustering method, it was possible to divide the measurements into six groups. In each group, the growth of the needles and the amounts of the nutrients were different, i.e. different groups represented different types of growing conditions. Forest experts were able to construct a model that characterizes the development of the condition of forests in Finland using the result of the clustering method.