Predicting deadwood densities of Cryptomeria japonica and Chamaecyparis obtusa forests using a generalized linear mixed model with a national-scale dataset

Predicting the stocks of deadwood in planted stands is important for controlling biodiversity, carbon content, and nutrient cycling in forest ecosystems and for managing deadwood accumulation because of conservation activities for biodiversity and economic reasons. To calculate the carbon stock in deadwood, the volume, carbon content, and appropriate wood density for the decomposition state must be known. To develop a predictive model for estimating deadwood density, we used a generalized linear mixed model and a national-scale dataset with large variations in different deadwood factors related to wood properties and climatic conditions. Our samples of fallen logs were obtained by noncommercial thinning of Cryptomeria japonica D. Don (C. japonica) and Chamaecyparis obtusa (Sieb. et Zucc.) Endl. (C. obtusa) planted forests throughout Japan, covering various climatic conditions. The wood density model for C. japonica was Dw = 0.342 − 0.010y − 0.001d + 0.001Sa + 0.012C − 0.028Fc1 − 0.006 Fc2 + re and that for C. obtusa was Dw = 0.431 − 0.015y − 0.001d + re, where d is the diameter, Sa is the stand age, C is contact with the forest floor, Fc1 and Fc2 are the climatic factors, and y is the years since death with random site effect re. Fc1 and Fc2 are the principal component scores showing Japanese climatic characteristics that were calculated through principal component analysis using temperature, precipitation, and snow depth data. The negligible variance in the random site effect for both tree species suggested that the models could be applied to other sites. In this study, our models showed accurate predictions of deadwood densities comparable to single exponential decay models, as shown by values for root-mean-square error and root-mean-square relative error.

► We develop a prediction model for estimating the deadwood density. ► The model was used a generalized linear mixed model and a national-scale dataset. ► The models generated more accurate predictions of the densities than previous ones.