Sampling strategies for estimating forest cover from remote sensing-based two-stage inventories

摘要: Remote sensing-based inventories are essential in estimating forest cover tropical and subtropical countries, where ground cannot be performed periodically at a large scale owing to high costs inaccessibility (e.g. REDD projects) mandatory for constructing historical records that can used as baselines. Given the conditions of such inventories, survey area is partitioned into grid imagery segments pre-fixed size proportion measured within using combination unsupervised (automated or semi-automated) classification satellite manual (i.e. visual on-screen) enhancements. Because on-screen operations time expensive procedures, only sample selected first stage, while each segment estimated second stage from pixels segment. data arising may freely available Landsat imagery) over entire (wall-to-wall data) likely good proxies manually classified (sample data), they adopted suitable auxiliary information. The question how choose areas carried out. We have investigated efficiency one-per-stratum stratified sampling selecting pixels, carry out determine difference estimator exploiting information estimation level. performance this strategy compared with simple random without replacement. Our results were obtained theoretically three artificial populations constructed (forest/non forest) pixel level study located central Italy, assuming levels error rates imagery. exploitation map proves highly effective respect Horvitz-Thompson estimator, which no exploited. use provides relevant improvement many few - jointly estimate by remote inventories.