学习笔记2018-8-28（读论文MODIS-TERRA & MODIS-Chla）

In contrast toMODISA,MODIS-Terra (MODIST) has been regarded as not being able to provide science quality data for water applications (Franz et al., 2007; Franz et al., 2008) until 2010 when NASA initiated an ocean color reprocessing of the full MODIST mission based on improved calibration of the MODIST visible (VIS)–NIR bands.

MODIS terra:上午星 AQUA：下午星

This is because MODIST SWIR bands contain more noise than MODISA SWIR bands, and MODIST SWIR bands have not been vicariously calibrated.Furthermore, both atmosphere and coastal/inland waters may change quickly; so much so that MODISA (afternoon passes) and MODIST (morning passes) may face very different cloud and water conditions.

MODISA SWIR波段存在垂直条纹噪声，MODISA SWIR带中的这种垂直条带噪声可以通过特定于各子帧的校准来校正。但是MODIST SWIR中既存在水平噪声，又存在垂直噪声，这使得噪声很难去除，A 3 × 3 median filter has shown effectiveness in removing the horizontal striping noise，but vertical striping noise remains a challenge.

提出了一个新的用于去除垂直条带噪声的方法：Average of Median Filtering of Even rows and Odd rows（偶数行和奇数行中值滤波的平均值，AMFEO），步骤如下：
(1) Fill the odd numbered columnswith invalidvalue (NAN), and apply a n × n median filter;
(2) fill the even numbered columns with invalid value (NAN), and apply a n × n median filter;
(3) calculate the average of the two filtered images. Considering that Taihu Lake is less homogenous than open ocean, n was selected to be 3.
但其实他并没有改变物理上的分布，只是改变了条纹的形状。

VIS-NIR：可见光-近红外

辐射定标与大气校正的关系
传感器定标就是将图像的数字量化值（DN）转化为辐射亮度值或者反射率或者表面温度等物理量的处理过程。其中反射率又分为大气外层表观反射率和地表实际反射率，后者又属于大气校正的范畴，有的时候也会将大气校正纳入传感器定标的一种途径。传感器定标可分为绝对定标和相对定标。绝对定标是获取图像上目标物的绝对辐射值等物理量；相对定标是将图像目标物辐射量归一化某个值范围内，比如以其他数据作为基准。目的 是消除传感器本身产生的误差。
大气校正：就是将辐射亮度或者表观反射率转换为地表实际反射率，目的是消除大气散 射、吸收、反射引起的误差。

接下来是对低质量的MODIST数据做处理的主要步骤：
1.短波红外波段的噪声估计与去除（AMFEO法）
2.区域辐射定标的修正：（1）Calibration adjustment of SWIR bands based on the well-calibrated NIR ocean bands
（2）Calibration adjustment of VIS-NIR land bands based on calibrated SWIR bands and in situ Rrs
3.MODIT大气校正与评价

*对于实测点数据的取舍问题：利用同步的卫星遥感和水面实测数据进行匹配的时候，由于水面实测是一个点，而图像的一个像元是500*500m，如果实测点周围不均匀，那么跟图像对应就存在问题。
关于这个问题，国际上比较常用的fan方法是： (1)以实测点对应的像元为中心，提取3*3窗口的像元；（2）计算这9给像元的变化率=标准差/均值；（3）变化率如果大于40%的，剔除这个点；（4）变化率小于40%的，求3*3的中值，作为和实测数据匹配的图像像元值。*

MODIS-CHLA
AURE and rURMSE are important and are often used when both MODIS-derived and field measured values contain large errors, when Chla concentrations vary between the time that MODIS data and in-situ measurements were obtained, or when water properties are not homogenous around the sampling site.