基于多源长时序遥感影像云计算的逐月灌溉面积监测研究

Study on monthly irrigated area monitoring based on multi-source long time-series remote sensing image cloud computing

摘要

【目的】灌溉面积精准核算对指导农业灌溉和用水量调配具有重要意义。然而,受到光学遥感影像可用性不足、灌区地面观测数据不足、公开数据更新频次较低等限制,在区域尺度开展灌溉面积快速识别仍然面临诸多困难。为支撑灌溉面积的多源卫星快速监测,【方法】基于光学和雷达长时序遥感影像,在Google Earth Engine云计算平台上提出了一种基于已发布数据集的月度历史样本获取方法,并通过融合雷达、光谱、水热、环境、时序等多源特征,实现了美国华盛顿州区域尺度2016—2017年逐月合成的灌溉面积快速识别。【结果】结果表明,研究区核心地区10%~17%土地在4—10月作物生长季发生了明显灌溉行为,灌溉初期面积达到800~1 100 km²,生长旺季灌溉面积超过1 600 km²,年度灌溉总面积达到2 200 km²。研究区逐月灌溉面积识别的总体精度为84.76%(Kappa=0.694 4)。【结论】相比公开发布的基于遥感影像的灌溉数据集和区域年度农业普查数据,利用提供的多源影像云计算方法,能更有效合理地识别逐月灌溉面积分布,提高灌溉信息更新频率并减少农业普查带来的灌溉面积低估风险,为作物需水量研究、灌溉用水评估提供重要依据。

Abstract

[Objective] The accurate accounting of irrigated area is of great significance for guiding agricultural irrigation and water consumption allocation. However, due to limited availability of optical remote sensing imagery, insufficient ground observation data in irrigated area, and low update frequency of public data, there are still many challenges in achieving rapid identification of irrigated area at the regional scale. The aim of the study is to support the rapid monitoring of irrigated area using multi-source satellite data. [Methods] Based on optical and radar long time-series remote sensing images, a method for acquiring monthly historical samples from published datasets was proposed on the Google Earth Engine cloud computing platform. By integrating multi-source features, including radar, spectral, hydrothermal, environmental, and temporal data, rapid identification of monthly synthesized irrigated area at the regional scale was achieved for Washington State, USA, from 2016 to 2017. [Results] The result showed that 10% ~17% of the land in the core area of the study area had obvious irrigation activities during the growing season from April to October, with the irrigated area at the initial stage reaching 800 km² to 1 100 km², the irrigated area during peak growing season exceeding 1 600 km², and the total annual irrigated area reaching 2 200 km². The overall accuracy of monthly irrigated area identification in the study area was 84. 76%(Kappa coefficient = 0. 694 4). [Conclusion] Compared with the publicly available remote sensing irrigation datasets and regional annual agricultural census data, the proposed multi-source image cloud computing method enables more effective and accurate identification of monthly irrigated area distribution. Meanwhile, it can improve the frequency of irrigation information updates and reduce the risk of underestimating irrigated area caused by agricultural census, thereby providing a crucial basis for research on crop water demand and irrigation water consumption assessment.

关键词

灌溉面积 / Google Earth Engine / 长时序遥感 / 多源特征 / 逐月灌溉 / 合成孔径雷达 / 遥感云计算 / 灌溉用水管理

Key words

irrigated area / Google Earth Engine / long time-series remote sensing / multi-source features / monthly irrigation / Synthetic Aperture Radar(SAR) / remote sensing cloud computing / irrigation water management

基金资助

国家重点研发计划项目(2021YFB3900604)

国家自然科学基金(42171078)