在绿潮遥感业务化监测中, 250 m分辨率的MODIS卫星数据是主要数据源, 归一化差值植被指数(NDVI) 是绿潮卫星遥感信息提取的主要方法。研究发现, 由于MODIS空间分辨率较低, 存在大量的混合像元, 导致提取的绿潮覆盖面积明显偏大。针对该问题, 本文在MODIS绿潮NDVI计算的基础上, 首先对大于NDVI阈值的像元进行混合像元分解, 得到MODIS NDVI混合像元分解后的绿潮面积, 然后以准同步的30 m分辨率HJ-1 CCD影像提取的绿潮覆盖面积为真值, 建立了MODIS NDVI混合像元分解得到的绿潮面积与HJ-1提取的绿潮面积之间的关系模型, 以实现绿潮面积的精细化提取。与传统的NDVI阈值法和混合像元分解法相比, 该方法提取的绿潮覆盖面积更接近于“真值”, 面积约为“真值”的96%, 而传统的NDVI阈值法和混合像元分解方法提取的面积分别为“真值”的2.96倍和45%。另外, 与传统的NDVI阈值法相比, 新方法对NDVI阈值变化不敏感, 在相同的NDVI阈值变化区间内, 前者提取的绿潮覆盖面积变化了41%, 而新方法的变化仅为11%。本文的工作在很大程度上解决了MODIS空间分辨率低导致的绿潮监测结果不准确的问题, 为精细化的绿潮卫星遥感业务监测提供了参考。

In the green tide operational remote sensing monitoring, MODIS data with 250 m spatial resolution is the commonly adopted satellite data. Normalized difference vegetation index (NDVI)is the primary method used for extraction of green tide information. However, it has been found that, because the MODIS spatial resolution is low, there are a lot of mixed pixels, resulting in severe overestimation of the extracted green tide area. To addressthis issue, we appliedthe mixed pixel decomposition methodto the MODIS NDVI image, and the derived green tide area was then calibrated with thatextracted from the quasi-synchronous 30 m resolution HJ-1 CCD image. The validation shows thatcompared with the traditional methods, the new model provides more accurate result andis less sensitive to the uncertainty of NDVI threshold.