基于时空数据融合的实际灌溉面积遥感监测方法研究

Research on remote sensing monitoring methods for actual irrigated area based on spatiotemporal data fusion

摘要

【目的】充分挖掘遥感影像的时序信息，动态识别灌溉面积及空间分布，为提升灌溉用水监管能力提供数据支撑。【方法】以内蒙古河套灌区永济灌域为研究区，基于增强的时空自适应反射率融合模型(Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model, ESTARFM)对MODIS和Sentinel-2数据进行时空融合，计算并构建逐日的再修正垂直干旱指数时间序列。针对时空融合数据出现的灌溉特征坦化现象，基于滑动窗口的双向长短期记忆网络(Bi-directional Long Short-Term Memory, Bi-LSTM)识别灌溉事件，动态监测灌区内的灌溉面积及空间分布。【结果】2024年4—6月永济灌域的实际灌溉面积分别为323.88 km²,462.67 km²,500.57 km²,灌溉事件识别结果的平均总体精度为89.82%,平均kappa系数为0.77,有效体现了灌域内灌溉事件时空分布的动态变化。【结论】时空融合为实际灌溉面积监测提供了更为连续的影像数据基础。基于滑动窗口的Bi-LSTM方法，有效捕捉了灌溉过程中土壤含水量的时序变化，解决了不同时空分辨率数据融合中土壤含水量变化过程坦化带来的灌溉识别难题，提高了实际灌溉面积连续动态监测的能力。

Abstract

[Objective] To fully utilize the temporal information of remote sensing imagery, dynamically identify irrigated areas and their spatial distribution, and provide data support for enhancing irrigation water management capabilities. [Methods] Taking the Yongji Irrigation Area in the Hetao Irrigation District of Inner Mongolia as the study area, spatiotemporal fusion of MODIS and Sentinel-2 data was performed using the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model(ESTARFM) to calculate and construct a daily time series of the Re-modified Perpendicular Drought Index(RPDI). To address the flattening of irrigation characteristics in spatiotemporal fused data, a Bi-directional Long Short-Term Memory(Bi-LSTM) network based on sliding window was used to identify irrigation events and dynamically monitor the irrigated area and its spatial distribution. [Results] From April to June in 2024, the actual irrigated areas in the Yongji Irrigation Area were 323.88 km², 462.67 km², 500.57 km², respectively. The irrigation event identification achieved an average overall accuracy of 89.82 % and an average kappa coefficient of 0.77, effectively reflecting the dynamic spatiotemporal variation of irrigation events within the irrigation area. [Conclusion] Spatiotemporal fusion provides a more continuous image data foundation for monitoring actual irrigated areas. The Bi-LSTM method based on sliding window effectively captures the temporal variations in soil moisture content during irrigation. It addresses the challenge of irrigation identification caused by the flattening of soil moisture changes in the fusion of different spatiotemporal resolution data, thereby improving the capability for continuous and dynamic monitoring of actual irrigated areas.

关键词

实际灌溉面积 / 遥感监测 / 时空数据融合 / RPDI时间序列 / Bi-LSTM / 永济灌域 / 影响因素

Key words

actual irrigated area / remote sensing monitoring / spatiotemporal data fusion / RPDI time series / Bi-LSTM / Yongji Irrigation Area / influencing factors

基金资助

国家重点研发计划项目(2024YFC3213600)