Towards optimized sampling design for forest monitoring: A moving window analysis of scale-sensitive indices

Accurate assessment of forest ecosystems requires choosing appropriate sampling scales that capture spatial heterogeneity. To determine the minimum sampling scale needed to assess key forest structural indices, we applied a moving window approach across twenty 1-ha primary broad-leaved Korean pine forest plots in Heilongjiang, China, systematically increasing window sizes from 5 m × 5 m to 95 m × 95 m and sliding them at 5 m intervals within each plot. At each window, we calculated seven indices representing three ecological dimensions: spatial structure (neighborhood comparison, angular scale, mingling), productivity (biomass, basal area), and species diversity (Simpson and Shannon indices). These values were compared to plot means to identify the minimum spatial scale at which index values stabilized within deviation thresholds of 10% or 20%. Then, we used a random forest model to investigate the drivers of variability in minimum sampling scale, incorporating plot-level means of each index and their coefficients of variation across 5 m × 5 m windows as explanatory variables. Results revealed varying scale sensitivities across indices. At the 10% deviation threshold, neighborhood comparison and angular scale stabilized at ≤ 50 m, species diversity at ≤ 75 m, and mingling and productivity at ≤ 85 m, which was identified as the minimum scale capturing all structural indices. Random forest analysis showed that while influencing factors varied across indices, the coefficient of variation was consistently the most important. This study provides a framework for tailored sampling strategies, supports forest inventory enhancements, and guides large-scale monitoring and conservation in temperate mixed forests.