Landsat 8 TIRS热红外光谱数据定标准确性的分析

Landsat系列卫星的热红外数据一直是获取地球表面温度的重要数据源，而新一代Landsat 8卫星的TIRS热红外传感器数据进一步延续了这一重要使命。但该卫星发射以来，其热红外传感器的定标参数不断发生变化，致使美国地质调查局(USGS)不得不在2014年2月对所有已获取的Landsat 8卫星数据进行重新处理。为了考察新处理数据的定标准确性，利用定标精度很高的Landsat 7 ETM+的3幅热红外影像来对同日过空的Landsat 8 TIRS热红外影像进行对比, 以查明TIRS热红外数据的定标准确性。结果表明，尽管Landsat 8 TIRS与Landsat 7 ETM+的热红外光谱数据很接近，但是，二者之间也存在着差别。与ETM+6波段反演的大气顶部温度相比，TIRS 10波段表现为高估，幅度最大为1.37 K，而TIRS 11波段则表现为低估，幅度可达-3 K。可见，Landsat 8 TIRS热红外光谱数据的定标参数精度仍不稳定，且以TIRS 11波段表现得更明显。进一步分析发现，TIRS数据的误差会随着地表植被和裸土覆盖比例的不同而发生变化。表现在TIRS 10波段的高估会随着植被比例的下降而加大，而TIRS 11波段的低估则会随着植被比例的下降而减少。因此，虽然USGS提倡用TIRS 10单波段来反演温度，但TIRS 10波段在低植被高裸土区的反演精度却远不及TIRS 11波段，所以在低植被高裸土区可能不宜一味地采用TIRS10波段，在没有把握的情况下，在低植被覆盖区也可尝试采用TIRS 10和11波段温度的均值，它可将误差缩小在<0.5 K范围以内。

A Comparative Study on the Calibration Accuracy of Landsat 8 Thermal Infrared Sensor Data

The satellite thermal infrared image has been an important data source for the acquisition of the earth ’s surface tem‐perature .The thermal infrared sensor (TIRS) Landsat 8 satellite newly launched onboard has added valuable data for this mis‐sion .However ,the calibration parameters for the two bands of the TIRS ,i .e .,TIRS Bands 10 and 11 ,had been modified sev‐eral times since its launch .This finally led the United States Geological Survey (USGS) to reprocess all achieved Landsat 8 data starting from February 2014 .In order to examine the calibration accuracy of the reprocessed TIRS data ,this paper crossly com‐pares Landsat 8 TIRS data with synchronized ,well‐calibrated Landsat 7 ETM + thermal infrared data .A total of three date‐coin‐cident image pairs of western United States ,downloaded from USGS Earth Explorer website ,were used for the cross compari‐son .Three test sites were selected respectively from the three image pairs for the comparison ,which representing moderate veg‐etation‐cover area (test site 1) ,low vegetation‐cover area (test site 2) ,and bare soil area (test site 3) .The thermal infrared data of the three image pairs of both sensors had been firstly converted to at‐sensor temperature .A band‐by‐band comparison and a regression analysis were then carried out to investigate the relationship and difference between the two sensor thermal data .The results show a very high degree of agreement between the three compared Landsat 8 TIRS and Landsat 7 ETM + thermal infrared image pairs because the correlation coefficients between the retrieved at‐sensor temperature of the two sensors are generally grea‐ter than 0.95 .Nevertheless ,the cross comparison also reveals differences between the thermal infrared data of the two sensors . Compared with retrieved at‐sensor temperature of Landsat 7 ETM + Band 6 ,TIRS Band 10 shows an overestimation ,which can be up to 1.37 K ,whereas TIRS Band 11 underestimates the temperature ,with a difference reaching to - 3 K .This suggests that in spite of the reprocessing of Landsat 8 thermal infrared data ,the calibration parameters for the satellite’s TIRS data are still unstable ,especially for TIRS Band 11 .It was found that the at‐sensor temperature difference between ETM + Band 6 and TIRS Band 10 was enhanced with the decrease in vegetation coverage from test site 1 to test site 3 .The at‐sensor temperature difference of test site 1 is 0.07 K and increased to 1.37 K in test site 3 ,a net increase by 1.3 K .While the at‐sensor temperature difference between ETM + Band 6 and TIRS Band 11 had an inverse performance .With the decrease in vegetation coverage from test site 1 to test site 3 ,the at‐sensor temperature difference was reduced from ~ - 3.0 to - 0.4 K .Therefore ,in bare soil dominated test site 3 ,the temperature difference was 1.37 K for TIRS Band 10 and - 0.4 K for TIRS Band 11 .The RMSE of TIRS Band 11 is also much lower than that of TIRS Band 10 .This suggests that TIRS Band 11 can perform batter in bare soil area than TIRS Band 10 though the latter shows an overall batter performance than TIRS Band 11 .The study also found that in low vegetation cover areas like in test sites 2 and 3 ,taking an averaged at‐sensor temperature of TIRS Bands 10 and 11 ,the difference between the two sensors’ at‐sensor temperature can be reduced to less than - 0.5 K .