Modeling and spatial analysis of carbon stock and forest attributes using mixed-effects models and artificial intelligence techniques

Abstract

Forests provide numerous ecosystem services, such as regulation of biogeochemical cycles, pollution control, food supply and the sequestration and storage of atmospheric carbon. These services are crucial, as they act directly in the mitigation of global warming and are of strategic importance in mitigating climate change. In this context, the quantification of the carbon stock present in the most varied types of forests constitutes an important tool for monitoring this ecosystem service. The estimation of carbon stock by indirect methods makes use of modeling and simulation techniques. Historically, the modeling of forest attributes has relied on approaches based on statistical models. These approaches now share space with computational approaches of artificial intelligence/machine learning, such as artificial neural networks, support vector machines, decision trees, among others, which have been gaining ground as tools for forest data analysis, modeling, estimation of variables and production prognosis. These tools have provided gains in the quality of estimates and predictions. In this work, we analyzed the spatial distributions of the carbon stock in a tropical forest and evaluated the performance of models extracted from artificial intelligence techniques to model the carbon stock in tropical forests; in addition to the use of artificial intelligence and mixed models with the adoption of a structure in the variance and covariance matrix for volumetric estimates. The total estimated carbon stock was 267.52 Mg·ha-1

, of which 35.23% was in aboveground biomass, 63.22% in soil, and 1.54% in roots. In the soil, a spatial pattern of the carbon stock was repeated at all depths analyzed, with a reduction in the amount of carbon as the depth increased. The carbon stock of the trees followed the same spatial pattern as the soil, indicating a relationship between these variables. In the fine roots, the carbon stock decreased with increasing depth, but the spatial gradient did not follow the same pattern as the soil and trees, which indicated that the root carbon stock was most likely influenced by other factors. The techniques performed satisfactorily in modeling, with homogeneous distributions and low dispersion of residuals. The

quality analysis criteria indicated the superior performance of the mixed model with a Huynh- Feldt structure of the variance and covariance matrix, which showed a decrease in mean relative

error from 13.52% to 2.80%, whereas machine learning techniques had error values of 6.77% (SVM) and 5.81% (ANN). This study confirms that although fixed-effects models are widely used in the Brazilian forest sector, there are more effective methods for modeling dendrometric variables.