中国1986-2021年30米年度耕地数据集（China’s Annual Cropland Dataset, CACD）

精确、详细且及时的耕地范围信息对于粮食安全保障和环境可持续性至关重要。然而，由于农业景观的复杂性和足够训练样本的缺乏，在大范围下进行高时空分辨率的耕地动态监测仍然具有挑战性，尤其是对于农业用地发生剧烈变化的地区。在这里，我们开发了一种新颖而高效的年度耕地制图框架，该框架集成了时间序列Landsat影像、自动化训练样本生成以及机器学习和变化检测技术。我们将所提出的方案实施到谷歌地球引擎的云计算平台上，并首次以30米空间分辨率生成了中国年度耕地数据集（CACD）。结果表明，我们的方法能够跟踪不同农业区域的动态耕地变化。CACD年度地图和变化地图的像素尺度F1分数分别为0.79±0.02和0.81。更进一步，跨产品间的精度评估、与统计数据的相关性以及空间细节方面显示出CACD相比于其他数据集的精确性和稳健性。根据我们的估计，从1986年到2021年，中国的耕地总面积扩大了30, 300平方公里（1.79%），在2000年之前经历了增长，但在2000年至2015年期间总体下降，之后略有恢复。耕地扩张集中在西北部，而东部沿海地区经历了大量的耕地流失。此外，我们观察到在研究期间至少有419, 342平方公里（17.57%）的耕地至少发生了一次弃耕。一致而高分辨率的CACD数据集可以被用于各种研究应用中，以促进可持续农业利用和粮食生产进展。

Precise, detailed and timely cropland extent information is critical for ensuring food security and environmental sustainability. However, due to the complexity of agricultural landscapes and the lack of sufficient training samples, dynamic cropland monitoring at large scales with high spatiotemporal resolution remains challenging, especially in regions where agricultural land undergoes dramatic changes. Herein, we developed a novel and efficient annual cropland mapping framework that integrates time-series Landsat imagery, automated training sample generation, as well as machine learning and change detection technologies. We deployed the proposed framework on the Google Earth Engine cloud computing platform, and for the first time generated the China Annual Cropland Dataset (CACD) at 30-meter spatial resolution. The results show that our method can track dynamic cropland changes across diverse agricultural regions. The pixel-scale F1-scores of the annual CACD maps and change detection maps are 0.79±0.02 and 0.81, respectively. Furthermore, cross-product accuracy assessments, correlations with official statistical data, and spatial detail analyses demonstrate the superior accuracy and robustness of CACD relative to other existing datasets. Based on our estimates, from 1986 to 2021, the total cropland area in China expanded by 30,300 square kilometers (1.79%). It experienced growth prior to 2000, then saw an overall decline between 2000 and 2015, before recovering slightly in the subsequent years. Cropland expansion was concentrated in the northwestern region, while the eastern coastal areas experienced substantial cropland loss. Additionally, we observed that at least 419,342 square kilometers (17.57%) of cropland experienced at least one instance of abandonment over the course of the study. This consistent, high-resolution CACD dataset can be utilized for a wide range of research applications to advance sustainable agricultural utilization and food production outcomes.

搜集汇总

背景与挑战

背景概述

该数据集提供了中国1986-2021年期间30米空间分辨率的年度耕地信息，通过集成Landsat影像和先进的变化检测技术，能够精确跟踪耕地动态变化。数据集显示中国耕地总面积在2000年前增长，2000-2015年下降，之后略有恢复，且耕地扩张集中在西北部，东部沿海地区则经历了大量耕地流失。