Pinus patulan tuotosennusteiden tarkkuuden riippuvuus mallinnusaineiston koosta ja estimointitavasta : Tapaus Tansania Sao Hill

The aim of the thesis was to evaluate the effect of the data size and estimation method on the yield estimates of Pinus patula in Sao Hill forest plantation in Tanzania. The data consisted of even aged and nonthinned forest stands. The yield was estimated with a simultaneous system of equations (Eerikäinen 2002) that was originally developed for Pinus kesiya plantations in Zambia and Zimbabwe. The system of equations included non linear models for the stand dominant height, the number of stems per hectare, the basal area median diameter, the stand basal area and the total stand volume. Using the previously developed models could be useful in the case there is lack of time and the inventory/modelling budget is low. The system of equations was estimated from different sizes of samples of the original data in order to research the effect of the data size on the yield estimates. The models were estimated with a Two Stage Least Squares method (2SLS). In addition, two different kinds of model estimation methods were tested. Firstly, all the model parameters were estimated from the samples and, secondly, only the levels of the models were estimated from the samples and the parameter values for the model forms were derived from Eerikäinen´s (2002) research. This method was expected to be particularly efficient in the case that the amount of the estimation data would be small. The system of equations fitted quite well to the data of Sao Hill plantation as all the assumption related to the estimation were met quite well. On the other hand, the RMSE values were quite high compared to ones in Eerikäinen´s (2002) research. This could refer to quite a considerable variation in the data that the models could not explain very well. The yield estimates were more precise and accurate as the amount of data was increased in estimation. In general the 95 % confidence intervals, the ranges, the medians and means of the RMSE values and biases decreased as the amount of data increased. The estimation method where all the parameters were estimated from the data seemed in general better than the method where only the levels of the models were estimated from the data and the form parameters were given the values from Eerikäinen´s (2002) research. The differences between the estimation methods were, however, quite small in small sample sizes.