Parameter recovery vs. parameter prediction for the Weibull distribution validated for Scots pine stands in Finland

Abstract

The moment-based parameter recovery method (PRM) has not been applied inFinland since the 1930s, even after a continuation of forest standstructure modelling in the 1980s. This paper presents a general overviewof PRM and some useful applications. Applied PRM provided compatibilityfor the included stand characteristics of stem number (N) and basal area(G) with either mean (D), basal area-weighted mean (DG), median (DM) orbasal area-median (DGM) diameter at breast height (dbh). A two-parameterWeibull function was used to describe the dbh-frequency distribution ofScots pine stands in Finland. In the validation, PRM was compared withexisting parameter prediction models (PPMs). In addition, existingmodels for stand characteristics were used for the prediction of unknowncharacteristics. Validation consisted of examining the performance ofthe predicted distributions with respect to variation in stand densityand accuracy of the localised distributions, as well as accuracy interms of bias and the RMSE in stand characteristics in the independenttest data set. The validation data consisted of 467 randomly selectedstands from the National Forest Inventory based plots. PRM demonstratedexcellent accuracy if G and N were both known. At its best, PRM providedaccuracy that was superior to any existing model in Finland - especiallyin young stands (mean height < 9 m), where the RMSE in total and pulpwood volumes, 3.6 and 5.7%, respectively, was reduced by one-half ofthe values obtained using the best performing existing PPM (8.7-11.3%).The unweighted Weibull distribution solved by PRM was found to becompetitive with weighted existing PPMs for advanced stands. Therefore,using PRM, the need for a basal area weighted distribution provedunnecessary, contrary to common belief. Models for G and N were shown tobe unreliable and need to be improved to obtain more reliabledistributions using PRM.