Estratégias para o cálculo do limites de confiança em inventários florestais

Abstract

This work aimed to evaluate different strategies for calculating confidence limits in forest inventories. Dataset was derived from computational simulations based on combinations of two vegetation types (Typical Cerrado and Seasonal Semideciduous Forest), two sample distributions (Normal and Log-normal), four sample sizes (5, 10, 15, and 20 sampling units), and four levels of volumetric variability (coefficients of variation of 5, 10, 20, and 30%). A total of 64 scenarios were simulated, each consisting of 200 samples. Confidence limits ( ) for each sample were estimated using the following methods: (A) classical approach, based on the t-Student distribution and the mean as a central tendency measure; (B) Percentile Bootstrap; (C) Jackknife-z; and (D) a variation of Method A, replacing the mean with the median. Method A yielded the most precise limits, with a higher percentage of samples containing the population mean volume within confidence intervals. On average, this percentage was 90.1% for normally distributed samples and 89.8% for log-normally distributed samples. It is concluded that Method A is robust for estimating confidence limits in samples following a Normal or Log-normal distribution, even under high variability and small sample sizes.