Evaluation of sampling strategies for estimating area of landcover and land-cover change

Monitoring land-cover area provides critical information to land managers and policy makers. Data from land-cover maps are often incorporated in model-assisted estimators such as difference and post-stratified estimators to make monitoring more cost effective. The objectives of this research were to evaluate these model-assisted estimators over a broad set of populations to quantify differences in precision for cluster sampling designs and to quantify the relative contributions of among- and within-cluster variance. Based on an assessment of over 100 populations distributed globally, model-assisted estimators reduced standard errors by 10-20% relative to a direct estimator that uses no auxiliary information. The gain in precision increased as map accuracy increased. Two-stage sampling resulted in standard errors generally only 5-10% higher than one-stage cluster sampling. A hybrid estimator constructed for use when the difference estimator had a high standard error was unbiased and yielded precision comparable to or better than the difference estimator.