一种DMSP/OLS稳定夜间灯光影像中国区域的校正方法

随着遥感技术的快速发展,国防气象卫星计划作战线扫描系统（DMSP/OLS）夜间灯光图像开始更多的应用于人文经济研究中。由于原始影像存在较多问题,如DN值饱和、年际不连续问题等,因此在使用之前需要对影像进行校正。“传统不变区域法”是应用较为广泛的校正方法,但仍存在部分问题,如未考虑目标区域长时间尺度上的微弱变化以及连续校正前基准年份的选取。本文对传统不变目标区域法进行改良,以黑龙江鹤岗市作为不变目标区域,选取3期辐射定标影像作为参考影像对DMSP/OLS稳定夜间灯光影像进行饱和校正,通过对比各年份影像饱和校正情况,选出最为合理的基准年份,从而对饱和校正后的影像进行连续性校正。为验证影像校正精度,本文从国家像元DN值、省级GDP与电力消费量、地级市GDP和县域GDP 4个层面与对应DN值进行线性回归检验,结果显示经改进方法校正后DMSP/OLS稳定夜间灯光影像TDN与市级GDP的拟合度（R2）平均值为0.85,远大于传统方法校正TDN与市级GDP拟合度（R2）平均值的0.53,且随着时间推移,传统不变目标区域法校正后拟合度逐渐降低至2013年的0.40,而利用改进后方法进行校正的拟合度未出现递减现象,2013年其R2仍为0.88,表明与传统不变区域法相比校正精度明显提高,饱和问题得到较大改善。

Correction of DMSP/OLS Stable Night Light Images in China

With the rapid development of remote sensing technology, the night light image of Defense Meteorological Satellite Program Scanning System (DMSP/OLS) has been more and more applied in the research of humanist economics. Since there are many problems in the original images, such as the saturation of DN value and inter-annual discontinuity, it is necessary to correct those data before using them. "Traditional invariant region method" is a widely-using correction method, but there are still some problems, such as not taking into account the weak changes from the prospect of a long-time scale of the target region and the selection of the reference year in a continuous correction. In this paper, the traditional invariant region method was improved. In order to reduce the impact on correction results, caused by the small changes of the target region in a long time scale, and to improve the saturation of the original images, the stable night light images of DMSP/OLS were saturation corrected with the help of selecting three group data (F12_1999, f14-15_2003 and F16_2006) as reference images. By the comparison between saturation corrected images and reference images, the reasonable reference year, F15_2006 (after the saturation correction), with the minimum error was selected, so as to carry out the continuous correction of the saturation corrected images. To verify the correction precision of images, in this paper, pixel DN value test was carried out at the national level. And linear regression test was carried out in the corresponding TDN value between GDP at the provincial level, electricity consumption at the provincial level, GDP at the prefecture-level and at the county level, at the same time, compared with the other published correction results such as Cao Z Y^[25], results shew that the average fit R2 between the improved DMSP/OLS stable night light images TDN and the city's GDP was 0.85 while the average fit R2 between the results with the traditional invariant region method and with city GDP was only 0.53. As time goes by, the fitting degree of the traditional invariant region method gradually decreased to 0.40 in 2013. However, the fitting degree using the improved correction method did not decrease. In 2013, the R2 was 0.88, which indicated that compared with the traditional invariant region method, the correction accuracy was significantly improved and the saturation problem was greatly improved. At the same time, the fitting degree of TDN and GDP on the county scale was about 0.6, indicating that the night light image data set on the county scale also had certain applicability. However, this approach did not completely solve the problem of pixel saturation. How to solve this problem perfectly is the core in the application research of noctilucent data in the future.