Quantifying the Trade-off Between Cost and Precision in Estimating Area of Forest Loss Using Probability Sampling in Guyana

Accurately mapping forest loss through statistical sampling is less expensive than wall-to-wall mapping. The Guyana Forestry Commission (GFC) has used 5m RapidEye imagery to map forest loss at the national scale from 2011-2014. Guyana is classified as a 'High Forest cover Low Deforestation' country with very rare, small scale forest loss events. The wall-to-wall data created by the GFC allow us to evaluate the precision of different sample-based strategies for estimating area of forest loss. Estimates can be produced at a greatly reduced cost relative to wall to wall mapping. We divided Guyana into 374 blocks (24km x 24km) in alignment with the 3A image tiles provided by RapidEye. We compared the standard errors of area estimators obtained from simple random, stratified random and systematic sampling. Two 30m resolution maps were compared for constructing strata, a global forest loss map, and a national forest loss map produced specifically for Guyana. For each map, several options for defining stratum boundaries and allocating sample size were evaluated. All stratified design options reduced the standard error of the estimates relative to simple random sampling. The Dalenius-Hodges and Jenks methods for choosing stratum boundaries were more precise than the Equal Area method. Optimal and equal allocation of sample size led to substantially better precision than proportional allocation. Small improvements in precision resulted by increasing the number of strata from three to five. Incorporating the global and national forest loss maps in a regression estimator reduced standard errors respectively by 80% and 50% relative to the standard error of the simple random sampling direct estimator. Stratified designs were better than simple random with regression. RapidEye data cost USD800/tile, so for USD75200, a stratified random sample of 94 tiles could yield a CV of 7-8% for forest loss area compared to a cost of USD274000 for complete coverage of Guyana. This study demonstrates the utility of the global forest loss product for use as a stratification input at the national level where both forest loss and the associated degradation are very rare and fine-scaled.