资源环境科技发展态势分析平台

Eucalyptus process-based models are the most promising tools available for assisting foresters and companies in the pursuit of appropriate planning, long-term management and for assessing the impact of climate variability and change on the forest sector. Reducing errors of Eucalyptus yield simulations is of high importance since it increases the accuracy and credibility of decision making in forest systems. Furthermore, information on ensemble approach for Eucalyptus yield estimations is still scarce. The aims of this study were: to calibrate and evaluate the performance (goodness-of-fit) of the FAO, APSIM and 3PG models for different major Brazilian Eucalyptus clones; to perform an intercomparison of the structural features of Eucalyptus simulation models; and to assess the performance of an ensemble approach of these models. To these ends, experimental yield data from 2012 to 2017 from eight Eucalyptus clones distributed over 23 locations with several and contrasting environmental conditions in Brazil were used. During the evaluation phase, the R-2 ranged from 0.81 to 0.88, the modelling efficiency index (E) ranged from 0.75 to 0.81 and the root mean square error (RMSE) ranged from 37.1 to 42.3 m(3) ha(-1), when the models were individually assessed. The multimodel ensemble approach improved the performance of Eucalyptus yield estimations, with higher R-2 (0.89), E (0.85) and lower RMSE (33.1 m(3) ha(-1)), compared to the individual performance of the models. Forest simulation models were able to estimate Eucalyptus yield across a wide range of climates and soil types when properly calibrated. The use of a multimodel ensemble reduced the errors of Eucalyptus yield estimations.