VIIRS/DNB夜间灯光月度产品插补方法对比——以北京为例

新一代VIIRS/DNB(Visible Infrared Imaging Radiometer Suite Day/Night Band)夜间灯光数据因其具备更精细的时空分辨率、数据不存在饱和现象、不同年份数据可比性强等优点而迅速代替DMSP/OLS(Defense Meteorological Satellite Program's Operational Linescan System)夜间灯光数据成为新的研究热点。但由于杂散光的污染,VIIRS/DNB夏季数据数值缺失严重,致使数据在空间和时间上不连续,因此,插补缺失数据成为后续应用的前提。鉴于此,论文从插补结果异常值、与参照值对比、计算性能三个方面,系统比较了三次样条插值(样条)、三次Hermite插值(Hermite)、灰色预测模型(GM)、三次指数平滑法(指数)4种插补方法的适用性,以期为插补VIIRS/DNB夜间灯光数据提供方法选择的依据。研究结果表明：①异常值比较方面,Hermite法未出现异常值,另外3种算法仅出现少量异常值(0.02%~1.34%);②与参照值的对比方面,Hermite法与参考值接近程度最高,GM接近程度最低,样条法和指数法介于两者中间;③算法性能比较方面,4种方法都具备计算简单、容易编程的特点,但指数法的算法时长是另外3种方法的10倍以上。因此,综合多方表现,当插补月份前后两侧均有足够长的原始数据时,插补效果好、计算速度快、不会出现过冲现象的Hermite法最适宜,样条法次之;当插补月份仅单侧有足够长的数据时,适宜采用指数法(插补效果好、计算速度较慢)或GM(插补效果偏低、计算速度快)进行插补。

Interpolation methods comparison of VIIRS/DNB nighttime light monthly composites: A case study of Beijing

Comparing with nighttime light data acquired by the Defense Meteorological Satellite Program's Operational Linescan System (DMSP/OLS), nighttime light data sensed by the Visible Infrared Imaging Radiometer Suite Day/Night Band (VIIRS/DNB) have a higher spatial resolution and finer temporal resolution. VIIRS/DNB nighttime light data also have a substantial number of improvements in terms of accuracy and in-flight calibrations. As a result, VIIRS/DNB nighttime light data become a new research hotspot rapidly. Even so, VIIRS/DNB nighttime light data are vulnerable to stray light and contain a large number of distorted values in mid and high latitudes, especially in summer. Therefore, this study took Beijing as an example and adopted cubic spline interpolation (spline), cubic Hermite interpolation (Hermite), gray model (GM), and triple exponential smoothing (exponent) to interpolate default data of May to July 2015, and then compared the results of these four interpolation algorithms. The result shows that: 1) With regard to abnormal values, Hermite does not produce any abnormal value, while the other three algorithms generate few such values (0.02%~1.34%). 2) Comparing with the reference data—the Visible Infrared Imaging Radiometer Suite Cloud Mask Stray Light (VCMSL) version, the interpolation result of Hermite is closest to the reference, and the GM result is least close to the reference. 3) In terms of computing time, all of these four algorithms are easy to be programmed and calculated, but the exponential smoothing method has to calculate smoothing parameter repeatedly and therefore it will spend much more time than the other three algorithms. In conclusion, a comprehensive assessment shows that when the two time periods before and after the interpolation months both have enough original data, Hermite will be the best choice because of its great interpolation performance, no overshoots, and fast calculation speed. Spline takes the second place. When only one side of the interpolation months has adequate data, GM and exponent methods both can be used. The GM calculation runs fast but the interpolation result is not optimal, and exponent calculation runs slow but the algorithm interpolates well.