Stafford Joshua D, Reinecke Kenneth J, Kaminski Richard M, Gerard Patrick D
Department of Wildlife and Fisheries, Box 9690, Mississippi State University, Mississippi State, MS 39762, USA.
J Environ Manage. 2006 Mar;78(4):353-61. doi: 10.1016/j.jenvman.2005.04.029. Epub 2005 Sep 9.
Large-scale sample surveys to estimate abundance and distribution of organisms and their habitats are increasingly important in ecological studies. Multi-stage sampling (MSS) is especially suited to large-scale surveys because of the natural clustering of resources. To illustrate an application, we: (1) designed a stratified MSS to estimate late autumn abundance (kg/ha) of rice seeds in harvested fields as food for waterfowl wintering in the Mississippi Alluvial Valley (MAV); (2) investigated options for improving the MSS design; and (3) compared statistical and cost efficiency of MSS to simulated simple random sampling (SRS). During 2000-2002, we sampled 25-35 landowners per year, 1 or 2 fields per landowner per year, and measured seed mass in 10 soil cores collected within each field. Analysis of variance components and costs for each stage of the survey design indicated that collecting 10 soil cores per field was near the optimum of 11-15, whereas sampling >1 field per landowner provided few benefits because data from fields within landowners were highly correlated. Coefficients of variation (CV) of annual estimates of rice abundance ranged from 0.23 to 0.31 and were limited by variation among landowners and the number of landowners sampled. Design effects representing the statistical efficiency of MSS relative to SRS ranged from 3.2 to 9.0, and simulations indicated SRS would cost, on average, 1.4 times more than MSS because clustering of sample units in MSS decreased travel costs. We recommend MSS as a potential sampling strategy for large-scale natural resource surveys and specifically for future surveys of the availability of rice as food for waterfowl in the MAV and similar areas.
在生态学研究中，通过大规模样本调查来估计生物及其栖息地的数量和分布变得越来越重要。由于资源的自然聚类，多阶段抽样（MSS）特别适合大规模调查。为了说明其应用，我们：（1）设计了一种分层多阶段抽样方法，以估计收获田地里作为密西西比河冲积平原（MAV）越冬水禽食物的水稻种子的晚秋数量（千克/公顷）；（2）研究了改进多阶段抽样设计的选项；（3）将多阶段抽样的统计效率和成本效率与模拟简单随机抽样（SRS）进行了比较。在2000 - 2002年期间，我们每年对25 - 35个土地所有者进行抽样，每个土地所有者每年抽取1或2块田地，并在每块田地里采集的10个土壤样本中测量种子质量。对调查设计每个阶段的方差成分和成本分析表明，每块田地采集10个土壤样本接近11 - 15个样本的最优值，而每个土地所有者抽取>1块田地带来的益处很少，因为同一土地所有者不同田地的数据高度相关。水稻数量年度估计值的变异系数（CV）在0.23至0.31之间，且受土地所有者之间的变异和抽样的土地所有者数量限制。代表多阶段抽样相对于简单随机抽样统计效率的设计效应在3.2至9.0之间，模拟表明简单随机抽样的成本平均比多阶段抽样高1.4倍，因为多阶段抽样中样本单元的聚类降低了出行成本。我们推荐多阶段抽样作为大规模自然资源调查的一种潜在抽样策略，特别是用于未来对MAV及类似地区作为水禽食物的水稻可获得性的调查。
