中文摘要：

为了获取城市尺度组分温度,实现城市水热平衡的高精度反演,探索了一种多波段热红外遥感影像的城市尺度组分温度反演算法。算法选取了植被、土壤和不透水表面等3种组分,并且针对ASTER数据,利用线性混合像元分解方法获取像元平均比辐射率,以MODIS近红外数据估算大气水汽含量和大气透过率,采用牛顿迭代法获取大气平均温度,并用最小二乘原理获取地表组分温度。最后,应用长沙市区的实验影像进行了实验研究,通过纯净像元上组分温度反演结果与分裂窗算法反演结果的对比分析,以及组分温度反演结果与实测数据的对比分析,对算法的精度进行了验证,结果表明：（1）纯净像元上,组分温度反演结果与分裂窗算法反演结果具有较好的相关性,植被组分相关性最高,达0.9796,2种结果平均绝对偏差值为0.36℃;（2）组分温度反演结果与实测组分温度绝对偏差范围为0.2～1.4℃,植被组分温度与实测值偏差相对较小,不透水表面组分温度与实测值偏差相对最大。

英文摘要：

Land surface component temperature has more significant physical meaning, and it reflects the actual distribution of temperature more significantly. Meanwhile, its retrieval algorithms have no need for hypothesis that components in pixels have the same temperature. Although the multi-angle retrieval algo- rithm of component temperature has become mature gradually, its application in the studies on urban thermal environment is restricted due to the difficulty in acquiring urban-scale mum-angle mermm m~l~u uo ta. Therefore, based on the existing multi-band remote sensing data, access to appropriate urban-scale component temperature is an urgent issue to be solved in current studies on urban thermal infrared remote sensing. In this paper, a new algorithm to retrieve land surface component temperature for urban area had been proposed. It took advantage of ASTER data, and evaluated mean emissivity of pixels based on linear spectral unmixing, retrieved atmospheric water vapor content from MODIS NIR bands, and used Newton ＂s iterative method to obtain atmosphere average temperature. Finally, an experimental study of this algo- rithm had been conducted and the retrieval result had been validated using some measured data. The re- sults showed that. （1） the results of component temperature retrieval algorithm and split window algo- rithm of pure pixels have high correlation coefficient and the correlation coefficient of vegetation is the highest; （2） compared with the measured data, biases of the retrieval result ranged between 0.2 and 1.4℃, and the vegetation component temperature among different components had the smallest bias value.