Risk-Averse Importance Sampling of Tree Attributes in High-Risk Forested Areas

Abstract

Abstract This study develops the theory of risk-averse importance sampling and explains its potential application to forest inventory estimation through the use of a heuristic simulation. When the risk-producing elements of the landscape are known, a risk-averse sampling strategy can be created that results in fewer samples in high-risk areas. Our simulation shows that for certain high-risk populations, risk-averse importance sampling can be highly effective at reducing both risk to field crew members (requiring only 10% of the plot visits in the riskiest category) and sample variance relative to simple random sampling. The method is shown to be especially helpful when a population of values of interest decreases with increasing risk, with a reduction in mean square error (MSE) of 84% to 99% in these cases. The simulation also showed the opposite effect on MSE can be expected when values of interest increase with increasing risk. By increasing field crew safety, risk-averse importance sampling should also improve the frequency and accuracy of field observations, potentially leading to even bigger gains in estimate precision. We recommend risk-averse importance sampling any time hazardous conditions can result in a high number of missing observations and reasonably accurate characterizations of landscape risks can be developed.