Strategies for calculating confidence limits in forest inventories

Abstract

Accurate confidence intervals for volume are indispensable for planning and decision-making in sustainable forest management and environmental regulation. This study evaluated, through extensive simulations, the robustness of statistical methods for estimating confidence limits. The database comprised simulated datasets from 64 scenarios combining two vegetation typologies, two sample distributions (Normal and Log-normal), four sample sizes (5–20 units), and four variability levels. Generalization was assessed in 28 additional scenarios using Gamma and Weibull distributions to represent conditions in Ombrophilous Forests. Confidence limits (90% probability) were estimated using four methods: (A) classical t-Student; (B) Percentile Bootstrap; (C) Jackknife-z; and (D) a median-based variant of A. Our results demonstrate the superior robustness of the classical Method A. It consistently delivered coverage probabilities nearest the nominal 90% level across all distributions, including symmetric (Normal), positively skewed (Log-normal, Gamma), and negatively skewed (Weibull) conditions. Even for the smallest sample size (n=5), Method A maintained reliable coverage (85.5–96.0%), while resampling methods showed significant undercoverage (often ≤85%), and the median-based approach introduced substantial bias. We conclude that the classical t-based method provides the most reliable confidence limits for inventories with sampling constraints, proving robust under high variability and non-normal data without computationally intensive techniques. These findings, derived from a controlled simulation, provide a robust methodological framework; their application to field data should consider the specific characteristics of the target population.