面向风云静止卫星地表温度产品的缺失数据修复方法对比  

A comparison of missing data reconstruction methods for Feng Yun geostationary satellite land surface temperature products

作　　者：薛兴盛[1] 吴艳兰[1] 

机构地区：[1]安徽大学资源与环境工程学院安徽省地理信息工程中心,合肥230601

出　　处：《安徽农业大学学报》2017年第2期308-315,共8页Journal of Anhui Agricultural University

基　　金：安徽大学引进人才科研启动基金(J01006037)资助

摘　　要：静止卫星地表温度数据是研究昼夜温度变化规律及全球气候变化的重要参数。但常见的静止卫星地表温度数据,如风云二号F星地表温度产品(FY-2F LST)由于受到云、雾霾和气溶胶等大气因素的影响,往往会出现数值缺失的现象。针对该问题,一系列基于温度昼夜变化模型(diurnal temperature cycle,DTC)的静止卫星地表温度产品空值数据修复方法被提出,如VAN2006方法;此外,常见的三次样条插值(Cubic spline)和智能的反向传播网络(back-propagation network,BP)利用像元时间连续性原理,通过温度预测实现地表温度产品空值数据修复在理论上也是可行的。模拟不同类型的像元缺失情况分析和比较了3种方法的温度修复结果。研究表明,在空值数量较少的情况下,3种方法修复结果都比较理想,其中Cubic spline方法效果最好;随着空值数量增加,3种方法修复效果都不同程度变差,其中VAN2006衰退较缓慢,Cubic spline衰退最快;当空值数目继续增多且达到一定数量时,VAN2006仍可以较为真实的还原地表温度数据,并且优于BP,其中Cubic spline修复结果最差,难以反映实际地表温度。对14种不同空间大小空值区域实验研究进一步说明3种方法对地表温度空值修复具有可行性,但修复效果与空值区域大小无关,时间序列中空值数量是最主要的影响因素。Land surface temperature （LST） derived from the geostationary satellites （GEO-LST） plays an im- portant role in studying the diurnal variation of temperature and the global climate change. However, the presence of clouds, haze, aerosol and other atmospheric disturbance generates numerous missing and abnormal values which affect the application of LST data. To solve this problem, a series of approaches for reconstructing geostationary sat- ellite LSTs based on diurnal temperature cycle （DTC） were proposed, such as the model VAN2006. In addition, the common cubic spline interpolation method and intelligent back propagation network （BP） repair missing data of geostationary satellites （GEO-LST） by temperature prediction used as time pixel continuous principle is feasible in theory. In this paper, different types of missing pixels were simulated to analyze and compare the three missing data reconstructed methods. The results showed that the three methods have ideal reconstructed results under circum- stances of few missing data. The Cubic Spline can effectively reconstruct the missing LST data, which is better than the other two methods. With an increase of missing data, the analysis of recession of three methods showed different degrees of variations in which the VAN2006 was the least sensitive to missing data, whereas the Cubic Spline showed the highest sensitive to missing data. When the missing data continued to increase and reached to a certain number, the VAN2006 still could show relatively real reconstructed results, which was better than BP; the Cubic Spline showed the worst data restoration, and it was difficult to reflect the actual surface temperature. The results further showed that the three methods could estimate LST smoothly with high accuracy, and the reconstructed results were independent of the missing data size, which was mainly determined by the number of missing samples in the entire day.

关 键 词：风云2号F星 地表温度 温度日变化方法 反向传播网络 三次样条 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]