基于实测值的Landsat 8水面温度反演算法对比——以太湖为例

水面温度是影响湖泊物理、化学、生物和生态过程的关键因素，遥感反演可以扩展获取水面温度的空间尺度，现有研究多选择表层水温或其他温度产品而非实际水面温度来验证遥感反演结果。为获取太湖水面温度的最佳反演算法，首先量化水面温度与水下20 cm和50 cm水温的差异，再基于太湖四季典型日期的Landsat 8热红外数据，利用辐射传输方程法（radiative transfer equation, RTE算法）、覃志豪单窗算法（mono-window algorithm, MW算法）、Offer Rozenstein劈窗算法（split window, SW1算法）和Jiménez-Mu?oz劈窗算法（SW2）反演水面温度，以实测值验证反演结果。结果表明：太湖水面温度与表层水温存在显著差异，白天水面温度分别比20 cm和50 cm深处水温高3.6 K和5.6 K，Landsat 8过境时刻水面温度分别比20 cm和50 cm深处水温高2.7 K和2.9 K；以实测值为准，SW2算法对太湖水面温度的反演效果最佳，绝对误差范围为0.1~1.4 K，MW和RTE算法次之，SW1算法反演效果最差，在低温/高温时的反演值较实测值偏低/高2.0~3.0 K；四种算法均能反演出太湖水面温度的时间变化，但难以量化其空间格局。

Comparison of algorithms to retrieve water surface temperature using Landsat 8 image with in-situ observations in Lake Taihu

Water surface temperature plays an important role in lake physical, chemical, biological and ecological processes. Remote sensing retrieval can expand the spatial observation scale of water surface temperature. However, most of current studies choose surface layer water temperature or satellite products instead of actual water surface temperature to verify remote sensing retrieval results. In order to select the most suitable algorithm to retrieve water surface temperature in Lake Taihu, the difference between the observational water surface temperature and the surface layer (20 cm and 50 cm depth) water temperatures was quantified firstly. Then, water surface temperature on typical dates in four seasons were retrieved using the radiative transfer equation algorithm (RTE), Qin Zhihao mono-window algorithm (MW), Offer Rozenstein split window algorithm (SW1) and Jiménez -Mu?oz split window algorithm (SW2), respectively. The retrieval performance of four algorithms was evaluated with in-situ observations from Taihu eddy flux network. The results show that there is significant difference between water surface temperature and surface layer water temperature. The maximum surface temperature is 3.6 K and 5.6 K higher than that at 20 cm and 50 cm depth, respectively during the daytime. At Landsat 8 overpass time, the surface temperature is 2.7 K and 2.9 K higher than that at 20 cm and 50 cm depth, respectively. Among the four algorithms, the SW2 algorithm ranks best with an absolute error range of 0.1 - 1.4 K, followed by the MW and RTE algorithms. The SW1 algorithm shows the worst performance with retrieval values 2.0 - 3.0 K lower / higher than observations in cold / warm season, respectively. The temporal variation in water surface temperature of Lake Taihu can be well reproduced by the four algorithms. However, the spatial pattern in water surface temperature is difficult to be quantified with any one of them.