全球微波陆表发射率产品质量评估及优化

Quality evaluation and optimization of global microwave land surface emissivity products

全球微波陆表发射率(MLSE)地图集是目前卫星资料同化观测算子中的重要初猜值或模拟值,但由于缺乏“真实”的观测MLSE,已有的基于多种微波传感器制作的全球MLSE产品质量可靠性未知,而且还缺少1套质量较好频率覆盖范围较宽的全球MLSE数据集。选择大气快速辐射传输模型(RTTOV)中使用的3套MLSE地图集(SSMI/S、AMSU-A/B、ATMS)、TELSEM2工具背景数据集、2套AMSR-E数据集(AMSR-E1和AMSR-E2)和1套FY-3D数据集,基于统计分析技术开展7套MLSE产品的全球时、空一致性评估,并选择6种典型土地覆盖类型开展各产品MLSE随土地覆盖类型和频率变化一致性评估,并在MLSE产品优选的基础上对TELSEM2发射率产品进行优化,新建1套频率为6.925—150.0 GHz的全球MLSE数据集(CoTELSEM2)。研究结果显示:AMSR-E2几乎不可用,AMSR-E1、TELSEM2、SSMI/S、AMSU-A/B、ATMS、FY-3D月MLSE之间具有较好的时、空一致性,平均空间相关系数为0.887—0.928,其中TELSEM2最高(0.928),FY-3D略低(0.914);平均绝对偏差为0.031—0.041,其中TELSEM2最低(0.031),FY-3D最高(0.041);传感器相同扫描方式较不同扫描方式的空间一致性更好,圆锥和跨轨扫描方式分别以TELSEM2和AMSU-A表现更好;ATMS的51.7 GHz MLSE存在系统性高估,AMSR-E1和FY-3D的23.8、89.0 GHz MLSE在高植被覆盖地区也存在系统性高估,而FY-3D MLSE则存在一些明显偏高或偏低问题。总体上,TELSEM2和AMSU-A/B的质量可靠性较高,FY-3D质量可靠性较差。新的CoTELSEM2发射率产品具有较好的时、空一致性和频率依赖一致性,且全球MLSE的季节变化不确定性存在明显的土地覆盖类型依赖特征。

The global microwave land surface emissivity(MLSE)atlas is currently an important initial guess or simulation value in satellite data assimilation observation operators.However,due to the lack of"real"observation of MLSE,the quality and reliability of existing global MLSE products based on multiple microwave sensors are unknown,and there is also a lack of global MLSE datasets with good quality and wide frequency coverage.This study selects three sets of MLSE atlases(SSMI/S,AMSU-A/B,ATMS)used in RTTOV(Radiative Transfer for TOVS)as well as the background dataset included in the TELSEM2 MLSE estimation tool,two sets of AMSR-E datasets(AMSR-E1 and AMSR-E2),and one set of FY-3D dataset.Based on statistical analysis techniques,global spatiotemporal consistency of the seven MLSE products is evaluated.Furthermore,six typical land cover types(Amazon tropical rainforest,Northern coniferous forest,North China plain farmland,Qingzang plateau grassland,Sahara desert,and Greenland ice cover)are selected to conduct a dependency assessment for each product based on land cover type and frequency.Based on the selection of emissivity products,the TELSEM2 emissivity products are optimized,and a new set of global MLSE data with a frequency coverage of 6.925—150.0 GHz is created(named CoTELSEM2).The results show that there is a significant inversion error in the AMSR-E2 product,making it almost unusable.The MLSEs on the monthly scale of AMSR-E1,TELSEM2,SSMI/S,AMSU-A/B,ATMS and FY-3D have a good spatiotemporal consistency with an average spatial correlation coefficient(R)of 0.887—0.928,with TELSEM2 having the highest value(0.928)and FY-3D having a slightly lower value(0.914).The mean absolute deviation(MAD)is 0.031—0.037,with TELSEM2 having the lowest(0.031)and FY-3D having the highest(0.041).The spatial consistency is better with the same scanning method than with different scanning methods.The cone and cross-track scanning methods perform better with TELSEM2 and AMSU-A/B products,respectively.ATMS 51.7 GHz channel MLSE has a systematic overestimation,and 23.8 GHz and 89.0 GHz channel MLSE for AMSR-E1 and FY-3D products also have a systematic overestimation in areas with high vegetation coverage,while FY-3D MLSE shows significant overestimation or underestimation at different times,regions,and frequencies.The seasonal variability uncertainty of MLSE increases with increasing frequency.Little uncertainty is found in the emissivity of the Sahara and Amazon tropical rainforests,while the uncertainty in the northern coniferous forest is the greatest.Overall,TELSEM2 and AMSU-A/B have higher quality reliability,followed by SMMI/S and ATMS,and AMSR-E1 and FY-3D have relatively poor quality reliability.The optimized new CoTELSEM2 emissivity product shows a good dependence on land cover type and frequency,while the uncertainty of global MLSE seasonal change has a strong dependence on land cover type,of which the uncertainty of Sahara desert and Amazon tropical rainforest emissivity is very small,while the uncertainty of northern coniferous forest is the largest,and the high frequency is significantly greater than the low frequency.