National forest inventories are crucial for assessing and monitoring forest resources, facilitating their effective management, and informing environmental policies. These efforts require substantial coordination, personnel, and financial resources, which are often limited, particularly in developing countries. Therefore, deciding where to establish inventory plots is essential for optimizing national forest monitoring designs. Using Mexico as an example, one challenge in completing its national forest inventory every five years is determining which forest plots to include while ensuring accurate data amid unpredictable budget constraints. In collaboration with the Comisión Nacional Forestal , this study develops a geostatistical framework using autocorrelated conditioned Latin Hypercube Sampling to optimize the design of the national forest inventory. We compared the baseline design (i.e., inventory from 2009 to 2014 with 19,251 plots) with alternative designs that retained 47%, 24%, and 12% of the original plots. We demonstrate that selecting fewer, yet representative, plots preserves univariate and multivariate probabilities, spatial dependencies, and upscaling capabilities for predicting forest biomass when using an optimization approach. The 24% design achieved 94.6% representativeness and predicted forest biomass with low error (1.22%) and uncertainty (1.50%) compared to the baseline design. Furthermore, the 24% design maintained data quality while enabling substantial reductions in sampling costs. This co-produced optimization approach demonstrates that the sample size can be reduced in budget-limited scenarios while preserving geostatistical properties. Moreover, it is broadly applicable and customizable for any forest inventory and target variable (e.g., biomass, height, basal area), meeting reporting and budget expectations. • We developed a geostatistical framework for optimizing forest monitoring designs. • The acLHS method preserves spatial and statistical representativeness with fewer plots. • The 24% sampling design reproduced 94.6% representativeness of the full inventory. • Forest biomass estimates showed <1.3% error and 76% cost savings vs. baseline. • This operational workflow is transferable for other forest monitoring systems.
Díez-Pastor et al. (Fri,) studied this question.