准确且及时的获取农作物种植面积信息，对于粮食产量的预测和估算是至关重要的，对我国制定合理的粮食政策、经济计划和管理措施，确保国家粮食安全具有重要意义。遥感数据是一种覆盖范围广、时效性强、经济效益好的数据，可为区域面积总量估算提供客观有效的信息，具有传统调查数据不可比拟的优势。但受到遥感提取技术的限制，难以在较大空间范围或复杂景观格局下，利用遥感数据直接获得满足精度要求的结果。抽样技术与遥感技术的结合为获取农作物种植面积提供了行之有效的解决思路。本文以浙江省粮食大县单季晚稻种植为例，从抽样单元与总体、抽样方法、估计量和成本效益四个方面入手，综合对比遥感数据辅助下的5种抽样单元最优尺寸计算方法、2种抽样单元形态、13种抽样方法和13种估计量，同时建立基于景观格局指数的样本内调查成本模型，探讨浙江省粮食大县农作物种植面积遥感估算方案的优化设计，并从效益分析的角度优化野外调查。研究得到以下结论：（1）在抽样单元方面，遥感数据辅助下的5种抽样单元最优尺寸计算方法的结果说明使用100×100m2的抽样单元时，单元的全局空间相关性（Moran’s I 指数）最低，空间自相关性的最低，即抽样单元间的相互影响小，单元的独立性最好。通过规则抽样单元和不规则抽样单元估算结果的CV、RRMSE和变差的对比，可知当样本量小于300 时，使用规则抽样单元估算的精确度优于不规则抽样单元至少1%，即使用规则抽样单元更具有优势。因此，在浙江省粮食大县的农作物种植面积遥感调查方案设计中，推荐使用100×100m2方形抽样单元。（2） 在抽样方法方面，综合对比13种抽样设计方案，可知在简单抽样设计中，遥感分类面积作为分层指标的估算效果最佳，与简单随机抽样的相对效率为3.66；其次是以耕作比指标进行分层，它与前者具有相似的估计精确性和准确性，相对效率为3.41；在实践中可根据以上两种指标获取的难易程度进行选择。破碎度分层指标在此地区的适用性不佳，不建议使用。在两阶段复杂抽样中，两阶段空间系统抽样的精确性和准确性最好，在样本量为200 时，其与简单随机抽样的相对效率为5.97，略高于以耕作比为分层指标的两阶段分层随机抽样（相对效率5.49）。统观13种抽样设计方案，以单季晚稻遥感识别面积作为排序指标的系统抽样CV、RRMSE 在各样本量下均较低，相对效率最高，在样本量为200时，其与简单随机抽样的相对效率可达到12.73，且估计量的变差最小，是效果最佳的抽样设计方案。（3）在估计量方面，对于4种混淆矩阵估计量来说，直接估计量的估算精度均优于逆校正估计量，且逆校正估计量明显受到低样本量的限制，会直接导致估算失败。另外，基于网格的混淆矩阵估计量的估算效果优于基于像元的混淆矩阵估计量。然而混淆矩阵估计量受遥感数据的总体分类精度的影响显著，在低分类精度（接近60%时）的遥感数据辅助时，不建议使用混淆矩阵校正估计量。纵观 13 种估计量的估算效果，精度最高的四种为基于网格的混淆矩阵直接估计量、基于网格的混淆矩阵逆校正估计量、基于模型的比估计量和基于模型的差回归估计量。（4）在成本效益方面，建立调查成本效益分析模型，并发现 logistic 曲线可以很好地拟合破碎度这一景观格局指标与样本内调查成本，R2为 0.783。利用成本效益模型，对比评价 4 种无偏抽样方法和估计量组合，结果表明以遥感面积为排序指标的系统抽样与基于像元的混淆矩阵直接校正估计量组合相对效益最高，为 2.042。另外，在无人机调查样带的设计中，“十”字形调查样带能够更好地平衡样本代表性、调查难度和可操作性，相对效益比方形样方提高24.6 个百分点。

Precise and prompt acquisition of crops area information is vital to the prediction and estimation of crop yield, and will be of great importance in management of agricultural commodities, enactment of sound food policy and economic plan as well, which is obviously significant for national food security. Remote sensing data covers wide geography scope, and is timeless and economical. Although unable to provide precise information for a large coverage or complex landscape directly because of the limitation of classification accuracy, it has proven to be a useful tool to area estimation on account of describing the overview of the ground. Combining sampling technology and remote sensing data is feasible. This study analyses the impact of 3 factors on the efficiency of estimating the area of one season late rice in Pinghu county, Zhejiang Province, which are the size and shape of sampling unit, sampling methods, estimators, attempting to find out the optimal scheme for area estimation in county level. Meanwhile, a costing model with landscape pattern index is constructed and applied to probe the effectiveness of a new field investigation mode with unmanned aerial vehicle and interpretation, instead of manual field survey. From this study, some basic conclusions are as follows:(1)Sampling unitCompared 5 methods of optimizing the size of sampling unit with the assist of classification from satellite imageries, spatial autocorrelation, describing by Moran’s I index, is lowest at 100×100m2, that means the sampling units interact less in this particle size and have higher dependence. In addition, regular sampling unit (square segment) and irregular sampling unit (natural boundary of cropland patches) are used in estimation, and the former performs better in CV, RRMSE and range of estimate. Therefore, the square segment with the size of 100×100m2 is recommended.(2) Sampling methodFirst, 3 stratification indexes are constructed in stratified sampling, classification area, ratio of classification area and arable land area and heterogeneity (hereafter short for area index, ratio index and Heter). The area index provides the best estimation and the relative efficiency is 3.66, the ratio index takes the second place, however, Heter cannot give an acceptable estimation, even worse than random sampling method. Second, in 9 two-stage sampling methods, two-stage spatial system sampling performs well and gives a relative efficiency, 5.97, and two-stage stratified sampling using ratio index comes second. The overall comparison of 13 sampling methods shows that the relative efficiency of system sampling using sorted area index is pretty high, 12.73, and the range of estimation is pretty narrow.(3)EstimatorIn 4 confusion matrix calibration estimators, direct estimator (hereafter short for DIE) is more applicable than inverse estimator (INE), owing to its smaller CV, RRMSE, and less limitation of sampling size. However, this kind of estimators relies strongly on the overall accuracy of classification (OA). When OA is as low as 60%, these estimators are incapable to provide better estimate than simple estimator. The overall comparison of 13 estimators shows that in the case of guaranteeing the sampling size, DIE performs best, and INE, model-based ratio estimator, model-based difference regression estimator comes sequencely.(4)CostThis study shows that, logistic curve fit the relationship of Heter and cost per sample unit well (R2 0.783). Based on this finding, a conceptual costing-model is established and applied in practical cost analysis. Compared 4 unbiased compositions of sampling method and estimator, system sampling (area index) with DIE (based on pixel) gives highest effectiveness, 2.042. Moreover, UVA improve the efficiency of field survey, and cross strip rather than square segment is more appropriate for UVA’s inspections, owing to its shape and flexibility, and rise 24.6 percentage point.