文摘
Statistical sampling is a practical alternative to complete coverage mapping of land cover when the objective is to estimate aggregate properties such as landscape pattern metrics. The bias and precision of sample-based estimators of twelve landscape metrics were evaluated for one point in time and for change over time using land-cover data from four regions representing diverse landscapes. Shape and edge metrics were estimated with very small biases,whereas patch and heterogeneity metrics were estimated with moderate to large biases. For all pattern metrics,very small biases were observed for estimating change. For simple random sampling,the sample size required to achieve a specified relative error varied greatly depending on the metric. Stratified random sampling improved precision relative to simple random sampling. The 10 km by 10 km block size generally yielded larger bias but smaller variance than 20 km by 20 km blocks. Keywords: Bias,variance,probability sampling,design-based inference,Horvitz-Thompson estimator.