Abstract:

Land surface temperature(LST) is an indispensable parameter for studying the exchange of matter and energy between the earth surface and the atmosphere, the changes of surface processes, geothermal detection and precursors of seismic thermal anomalies, etc. However, cloud cover phenomenon results in a large number of null values in MODIS LST products, which limits the wide application of LST. In this study, a MODIS LST reconstruction algorithm based on LST diurnal variation correlation of similar land features at different time is proposed. Taking Hotan, Xinjiang as the research area, and using MODIS 8-day LST products from 2003 to 2015 as experimental data, based on the correlation of LST between similar features at different times of a day, with the help of the surface cover type, the regression models of morning, afternoon, evening and early morning surface temperature were established respectively. The LST fitting of the three was regressed to the LST of the early morning, and then the optimal combination of the fitting results of morning, afternoon and evening was taken to achieve the two-step reconstruction of the early morning LST. The experimental results show that the minimum error of this method is 0.57K, the errors are all below 1.2K, and the average error is 0.92K. It is proved that this method has also a good complement effect when applied to the reconstruction of surface temperature of the remaining three time periods. Compared with the existing LST complement methods, the proposed method achieves higher accuracy and rate of complement with a small amount of auxiliary data, which can provide a solid data basis for the research of surface process based on temperature and seismic thermal anomaly detection. The main conclusions of this study are verified as follows:
(1)By establishing regression models for MOD11A2 morning and evening and MYD11A2 afternoon and MYD11A2 early morning surface temperature of different land cover types, it can be seen that second-order polynomial fitting effect is better and the model has better stability.
(2)For the study area, the accuracy of using night data to complement the early morning LST is the highest, but the spatio-temporal continuity of reconstructed LST cannot be completely guaranteed. If morning and afternoon LST data are added, the complement rate can be further improved.
(3)The reconstruction data of land surface temperature based on this method has better regional consistency and no step change. Compared with the LST compensation method proposed in recent five years, it is found that this method uses less auxiliary data and has relatively high accuracy of compensation. The errors after compensation are all below 1.2K, with an average error of 0.92K and a minimum error of 0.57K, which can meet the requirements of surface temperature compensation.
(4)At the same time, the missing data of MODISLST data in the other three time periods of a day are reconstructed respectively in this paper, with high reconstruction accuracy. Based on this method, the accuracy of data reconstruction in the early morning and evening is higher than that in the morning and afternoon. Therefore, in the future anomaly extraction work, the evening or early morning data can be given priority as the main data source, so as to achieve accurate anomaly extraction.
(5)In order to verify the universality of this method in different regions, two sub-regions with different land cover types and locations in the study area were randomly selected to carry out regional usability evaluation, and the results showed that the accuracy of reconstruction in both regions was high, and missing value reconstruction could be achieved.
In conclusion, the proposed method in this paper provides a new idea and method for MODIS land surface temperature reconstruction. Compared with the existing LST complementary method, the proposed method achieves higher complementary accuracy and complementary rate with a small amount of auxiliary data, which can provide a solid data foundation for the study of surface processes based on temperature. In addition, the complete surface temperature data after reconstruction can provide data support for geothermal anomaly detection and extraction, and lay a theoretical foundation for the interpretation of seismic thermal anomaly mechanism.

Key words: LST, MODIS, different time, data reconstruction, land cover type

摘要：

地表温度(LST)是研究地表与大气之间物质和能量交换、 地表过程变化以及地热探测与地震热异常前兆等方面不可或缺的重要参数, 然而云覆盖现象导致MODIS LST产品存在大量空值, 限制了LST的广泛应用。文中提出了一种基于同类地物不同时刻LST日变化相关性的MODIS LST重建算法。以新疆和田为研究区, 使用2003-2015年MODIS 8d合成地表温度产品为实验数据, 根据一天中同类地物不同时刻LST之间的相关性, 以地表覆盖类型产品为依据, 分别创建各类地物上午、 下午和晚上与凌晨的LST回归模型, 将三者的LST拟合回归至凌晨时刻的LST, 然后取上午、 下午和晚上拟合结果的最优组合以实现对凌晨LST的两步重建。实验结果表明, 该方法的最小误差为0.57K, 误差均在1.2K以下, 平均误差为0.92K。经验证, 将该方法应用于其余3个时刻的地表温度重建工作中仍可得到较好的补值效果。与现有的LST补值方法进行对比可知, 本方法以少量辅助数据实现了较高的补值精度和补值率, 可为基于温度的地表过程研究和地震热异常检测等研究提供坚实的数据基础。

关键词: LST, MODIS, 非线性回归, 数据重建, 地表覆盖类型