Developing Tree Mortality Models Using Bayesian Modeling Approach

Abstract

The forest mortality models developed so far have ignored the effects of spatial correlations and climate, which lead to the substantial bias in the mortality prediction. This study thus developed the tree mortality models for Prince Rupprecht larch (Larix gmelinii subsp. principis-rupprechtii), one of the most important tree species in northern China, by taking those effects into account. In addition to these factors, our models include both the tree—and stand—level variables, the information of which was collated from the temporary sample plots laid out across the larch forests. We applied the Bayesian modeling, which is the novel approach to build the multi-level tree mortality models. We compared the performance of the models constructed through the combination of selected predictor variables and explored their corresponding effects on the individual tree mortality. The models precisely predicted mortality at the three ecological scales (individual, stand, and region). The model at the levels of both the sample plot and stand with different site condition (block) outperformed the other model forms (model at block level alone and fixed effects model), describing significantly larger mortality variations, and accounted for multiple sources of the unobserved heterogeneities. Results showed that the sum of the squared diameter was larger than the estimated diameter, and the mean annual precipitation significantly positively correlated with tree mortality, while the ratio of the diameter to the average of the squared diameter, the stand arithmetic mean diameter, and the mean of the difference of temperature was significantly negatively correlated. Our results will have significant implications in identifying various factors, including climate, that could have large influence on tree mortality and precisely predict tree mortality at different scales.