ABSTRACT With a population of 267 million, Indonesia faces the significant challenge of inaccurate rice production data leading to a flawed national rice import policy and supply problems. Its 2018 rice production and harvest area data were generated through the Area Sampling Frame (ASF) method which incurred high labour and financial costs as well as failing to optimize accuracy; hence, an alternative method needs exploration. This study compares ASF and remote sensing Synthetic Aperture Radar (SAR) methods to calculate rice growth stages (RGS) using Indramayu Regency, the highest rice producer in West Java Province, as the study area. The SAR-based method used time-series of Vertical Horizontal (VH) polarization of Sentinel-1A data that employed a combination of k-means clustering, hierarchical cluster analysis (HCA), a visual interpretation and support vector machine (SVM) classifier. Both SAR and ASF methods can generate results on a monthly basis, although remote sensing satellite time revisits can be shortened (every 12 days). Whilst the ASF, a basic technique for collecting agricultural statistics, was easy to implement in large-scale areas its accuracy depended on the quantity and representativeness of the samples. This study applied the ASF by simulating a sample size of 1.7%, 3.3% and 5% of a rice field area with unmanned aerial vehicles (UAVs) data as a reference. Whilst remote sensing SAR methods involve complex data processing the image classification process can be conducted automatically and cost-effectively (data and its software are free of charge). Moreover, it yields not only statistical data on RGS but also determines the spatial planting patterns and the RGS distribution at 10 m pixel resolution. This method showed more accurate results with overall accuracy of image classification of 81.89% and a kappa coefficient (κ) of 0.73. The comparative result was relatively small, i.e., 4,094.89 ha more than the ASF results (3.5% difference), since this study covered a limited research area. Nonetheless, with evidence of more accurate results remote sensing holds the potential for replication across the country’s 416 regencies, so enabling government to develop more appropriate policies that minimize the risks of either a surplus or shortage in the nation’s most important food supply.

中文翻译：

---

印度尼西亚西爪哇省因德拉马尤县水稻种植面积估算中的遥感与面积抽样框法

摘要 印度尼西亚拥有 2.67 亿人口，面临着大米生产数据不准确的重大挑战，导致国家大米进口政策存在缺陷和供应问题。其 2018 年的水稻产量和收获面积数据是通过面积抽样框架 (ASF) 方法生成的，该方法产生了高昂的人力和财务成本，并且未能优化准确性；因此，需要探索一种替代方法。本研究比较 ASF 和遥感合成孔径雷达 (SAR) 方法以计算水稻生长阶段 (RGS)，以西爪哇省最大的水稻生产商 Indramayu Regency 作为研究区域。基于 SAR 的方法使用 Sentinel-1A 数据的垂直水平 (VH) 极化时间序列，采用 k 均值聚类、层次聚类分析 (HCA)、视觉解释和支持向量机 (SVM) 分类器。SAR 和 ASF 方法都可以按月生成结果，尽管可以缩短遥感卫星时间重访（每 12 天）。虽然 ASF 是一种收集农业统计数据的基本技术，但在大范围地区很容易实施，但其准确性取决于样本的数量和代表性。本研究以无人机 (UAV) 数据为参考，通过模拟 1.7%、3.3% 和 5% 的稻田区域的样本大小来应用 ASF。虽然遥感 SAR 方法涉及复杂的数据处理，但图像分类过程可以自动且经济高效地进行（数据及其软件是免费的）。而且，它不仅产生 RGS 的统计数据，而且还确定了空间种植模式和 10 m 像素分辨率下的 RGS 分布。该方法显示出更准确的结果，图像分类的总体准确率为 81.89%，kappa 系数 (κ) 为 0.73。比较结果相对较小，即比 ASF 结果多 4,094.89 公顷（差异 3.5%），因为该研究涵盖的研究领域有限。尽管如此，有了更准确结果的证据，遥感技术仍有可能在全国 416 个地区进行复制，从而使政府能够制定更合适的政策，将国家最重要的食品供应过剩或短缺的风险降至最低。该方法显示出更准确的结果，图像分类的总体准确率为 81.89%，kappa 系数 (κ) 为 0.73。比较结果相对较小，即比 ASF 结果多 4,094.89 公顷（差异 3.5%），因为该研究涵盖的研究领域有限。尽管如此，有了更准确结果的证据，遥感技术仍有可能在全国 416 个地区进行复制，从而使政府能够制定更合适的政策，将国家最重要的食品供应过剩或短缺的风险降至最低。该方法显示出更准确的结果，图像分类的总体准确率为 81.89%，kappa 系数 (κ) 为 0.73。比较结果相对较小，即比 ASF 结果多 4,094.89 公顷（差异 3.5%），因为该研究涵盖的研究领域有限。尽管如此，有了更准确结果的证据，遥感技术仍有可能在全国 416 个地区进行复制，从而使政府能够制定更合适的政策，将国家最重要的食品供应过剩或短缺的风险降至最低。