Sampling Approaches for Monitoring Urban Forests Inventories: Sample Size, Costs and Accuracy of Estimates

ALESSO, C.A.; CASTRO D.C.

Conferencia; Conference on Applied Statistics in Agriculture and Natural Resources; 2021

The University of Florida's Institute of Food and Agricultural Sciences

Management of urban forest requires accurate and updated information about forest structure and condition. Although very accurate, complete surveys are very expensive and have low time resolution for management purposes. Sampling methods are an affordable alternative for monitoring forest attributes with a desired level of uncertainty. Sample size needs, attainable accuracy and costs can vary due to the nature of the target variable, its distribution and the spatial configuration of the forests. However, only general one-fits-all recommendations based on the size of the population are available for designing sampling plans. Using real data from inventories of 40k to 370k trees, we evaluated simple random sampling (SRS) and cluster random sampling (CRS) in terms of sample size, accuracy and costs (time) computing the minimum sample size by Cochran?s and Tortora?s methods for categorical attributes. In general, standard recommendations resulted in larger sample sizes than those obtained by theoretical formulas. For small cities, travel time reductions using CRS with city blocks as clusters did not outperformed the larger sample size required. In contrast, for large cities, SRS increased the sampling costs compared to CRS