Accessing the spatial distribution of aboveground biomass in tropical mountain forests using regression kriging simulation: a geostatistical approach for local-scale estimates

Abstract Background Accurate measurements of aboveground biomass (AGB) are essential for understanding the planet’s carbon balance. The Atlantic Forest of the Serra do Mar in southeastern Brazil contains large areas of well-preserved remnants, characterized by mountainous terrain with significant orographic contrasts along its elevation gradient. This diverse landscape creates a variety of biophysical factors that strongly influence the spatial distribution of AGB. This study aims to estimate AGB using a hybrid geostatistical methodology, regression kriging simulation (RKS), to analyze AGB spatial distribution at a local scale (84 plots, each 0.01 ha) across a small forest fragment covering the entire tree-covered area (8777 ha). Building on traditional regression kriging method, this study introduces an innovative approach by incorporating Gaussian simulation to interpolate residuals, allowing RKS to account for uncertainties in the estimation process and create new results. This allows us to clearly distinguish exogenous ecological processes from endogenous ones before reaching the model’s final estimate. Results Four regression kriging models were created, and the best-performing model used the Enhanced Vegetation Index and direct solar radiation (DSR), achieving an R 2 of 55%. A Gaussian simulation was performed to interpolate the residuals of this model. The final results indicate that RKS provides accurate AGB estimates (RMSE = 1.333 Mg/0.01 ha and R 2 of 77%). Additionally, the inclusion of DSR as a new predictor variable enhances the precision of AGB estimates. The analysis showed that 63% of the sample pairs exhibited measurable spatial dependence. Conclusions Regression kriging simulation is proposed using Gaussian simulation, altering the classical application of regression kriging. For this, a case study was conducted in the Atlantic Forest of Serra do Mar to estimate the spatial distribution of tree biomass in a forest fragment of this region. We demonstrate that the proposed method better captures the heterogeneity of the region and produces more comprehensive results than regression kriging. Regression kriging simulation estimates tree biomass by considering the actual fluctuations of the spatial distribution of tree biomass in the region, taking into account exogenous and endogenous ecological processes, addressing random noise, and allowing the creation of dynamic maps for use by environmental managers.