传感器光谱响应函数的差异对棉花反射信息的影响

The Impact of Differences in Sensor Spectral Response Functions on Cotton Reflectance Information

摘要: 棉花作为全球最重要的天然纤维作物，在农业经济、“减排降碳”和国际贸易中具有关键地位。准确刻画棉花多源遥感反射信息的一致性，是开展精细化监测、资源管理和多源数据融合的前提。近年来多源中高分辨率遥感快速发展，为区域尺度棉花信息提取提供了有效技术途径，但针对传感器光谱响应函数差异对棉花反射率及植被指数估算的系统影响仍缺乏量化认识。因此，本文面向国内外多种主流中高分辨率光学传感器，利用2346条不同物候期棉花冠层高光谱观测，结合各传感器SRF，模拟生成多源多光谱反射率及植被指数，并从“波段–指数–物候期”三维视角解析由SRF内在差异向地表观测传递的不确定性。结果表明，蓝、绿及近红外波段的模糊贴近度普遍高于0.95，NDVI和SAVI的平均绝对相对偏差整体控制在0.2%~0.35%，而红波段局部贴近度可降至约0.89，EVI的偏差可达约3.8%，成为多源棉花反射与指数不一致的敏感波段和指数；具备SWIR1波段的L7_ETM+、L8_OLI与Sentinel-2A/B在LSWI上的一致性极高。由此表明，在棉花植被遥感研究中若忽略SRF诱发的差异，可能在特定物候期和敏感指数上引入数个百分点甚至更高的系统误差，因此在多源数据融合与跨传感器对比时必须显式考虑SRF差异并设计合理的数据组织与误差校正策略，以保证研究结论的科学可靠与可比。

Abstract: As the most important natural fiber crop worldwide, cotton plays a pivotal role in the agricultural economy, carbon emission reduction, and international trade. Accurately characterizing the consistency of multi-source remote sensing reflectance information for cotton is a prerequisite for refined monitoring, resource management, and multi-source data fusion. In recent years, the rapid development of multi-source medium- to high-resolution remote sensing has provided an effective technical means for extracting cotton information at regional scales, yet a quantitative understanding of the systematic impacts of differences in sensor spectral response functions (SRFs) on cotton reflectance and vegetation index estimation is still lacking. Therefore, this study targets multiple mainstream medium- to high-resolution optical sensors from China and abroad, and uses 2346 canopy hyperspectral measurements of cotton at different phenological stages, combined with the SRFs of each sensor, to simulate multi-source multispectral reflectance and vegetation indices. From a three-dimensional perspective of “band-index-phenological stage”, we analyze how intrinsic SRF differences propagate into uncertainties in surface observations. The results show that the fuzzy similarity in the blue, green, and near-infrared bands is generally higher than 0.95, and the mean absolute relative bias of NDVI and SAVI is overall constrained within 0.2%~0.35%. In contrast, the local similarity in the red band can drop to about 0.89, and the bias of EVI can reach approximately 3.8%, making them the most sensitive band and index responsible for inconsistencies in multi-source cotton reflectance and vegetation indices. Sensors equipped with an SWIR1 band, such as L7_ETM+, L8_OLI, and Sentinel-2A/B, exhibit extremely high consistency in LSWI. These findings indicate that neglecting SRF-induced differences in cotton vegetation remote sensing studies may introduce several percentage points or more of systematic error at specific phenological stages and for certain sensitive indices. Consequently, SRF differences must be explicitly considered, and appropriate data organization and error-correction strategies must be designed in multi-source data fusion and cross-sensor comparisons to ensure the scientific reliability and comparability of research conclusions.