Today Xiaoyang updates the recently reappeared SCI papers, and I will introduce them in three parts. This article mainly explains DMSP/OLS data download and processing.
1. Introduction (Part)
Rapid economic development requires better monitoring and forecasting methods. Gross domestic product (GDP) is widely regarded as one of the main indicators to measure the sustainable economic development of a country or region. It provides a new basis for regional economic development and formulating sustainable development strategies to predict economic development at different scales. GDP data mainly comes from surveys by the National Bureau of Statistics or other administrative departments. Despite their authority, they have inherent limitations (eg, "water injection" and uncertainties introduced by statistical data methods). NTL imagery provides a unique opportunity to observe human activities directly from space, enabling many applications such as mapping urban areas, estimating population and urbanization, and monitoring disasters and conflicts. At the same time, it is found that there is a strong correlation between NTL and GDP, which makes it possible to use NTL to build a GDP forecasting model. Existing GDP forecasting models, researches tend to focus on the perspectives of provincial and municipal levels, and there are few studies on GDP forecasting at the national level. This study aims to fill this gap by developing GDP forecasts for all provinces in China.
2. Data download
- DMSP/OLS data download website:
- National vector border map (does not contain ten-dash lines)
Link: https://pan.baidu.com/s/1E1TcHUt2-dR_d4w0AtI_WQ
Extraction code: 0946
3. Data processing
Figure 1 shows the sum of DN values (totalDN value, TDV) of all bright value pixels in the uncorrected China region. As can be seen:
① The data of the same year can be acquired by different sensors, and the total number of bright value pixels acquired by different sensors is inconsistent, and the total DN value is inconsistent;
② There are abnormal fluctuations in the total number of pixels and total DN values acquired by the same sensor in different years;
③Due to the existence of the above two problems, the data between multi-phase night light images is discontinuous and incomparable;
④Due to the limitation of the spectral resolution of the sensor, the pixel values in the urban central area are clustered and saturated, and the maximum DN value is 63, making the difference in the urban central area insignificant.
The formula for calculating TDV is:
In the formula, DNi represents the DN value of i pixel; Ni represents the number of i pixel.
Figure 1 Uncorrected TDV in China
1. TDN data statistics
So how is the above data TDV calculated? We can use our old friend ModelBuilder. Extract the Chinese region by masking [Spatial analyst] - [Extract Analysis] - [Extract by Mask]. Then use [Projection Raster] to project it as Krassovsky_1940_Albers (more suitable for China). Since it is necessary to build an attribute table to count DN pixel values, it is necessary to use the [Display Zoning Statistics in Table] tool, but the premise of using this tool is to convert the data into integers. Therefore, you need to use [Analysis Tools] - [Mathematical Analysis] - [Convert to Integer] to achieve. And add a floating-point TDV field to the table.
Figure 2 Model Builder Statistics TDV
After the attribute table is constructed, use the field calculator to enter an expression to obtain TDV, and then perform statistics on the TDV field. According to the above method to complete the data statistics from 1992 to 2013.
2. Mutual correction and saturation correction
The 34 expected corrected stable light images from 1992 to 2013 and the data of the F16 sensor in 2006 in the radiometric calibration product data are mutually corrected and saturated corrected. Use the RSR correction model for correction, and the correction parameters are shown in the table below
In the formula, DN represents the brightness value before correction; a, b, c are regression coefficients; DN correct represents the DN value after correction.
| years |
a |
b |
c |
R2 |
years |
a |
b |
c |
R2 |
||
| F10 |
1992 |
0.036 |
0.337 |
2.581 |
0.85 |
F14 |
2001 |
0.032 |
0.56 |
0.487 |
0.869 |
| 1993 |
0.045 |
0.334 |
2.15 |
0.85 |
2002 |
0.023 |
0.631 |
0.356 |
0.888 |
||
| 1994 |
0.039 |
0.518 |
1.252 |
0.85 |
2003 |
0.02 |
1.173 |
-0.215 |
0.855 |
||
| F12 |
1994 |
0.039 |
0.599 |
0.277 |
0.854 |
F15 |
2004 |
0.024 |
1.146 |
-0.153 |
0.901 |
| 1995 |
0.034 |
0.513 |
0.485 |
0.851 |
2005 |
0.026 |
0.872 |
0.123 |
0.855 |
||
| 1996 |
0.037 |
0.479 |
0.511 |
0.862 |
2006 |
0.015 |
1.076 |
-0.081 |
0.894 |
||
| 1997 |
0.03 |
0.56 |
0.413 |
0.877 |
2007 |
0.018 |
0.977 |
0.086 |
0.892 |
||
| 1998 |
0.027 |
0.511 |
0.481 |
0.876 |
2004 |
0.022 |
0.809 |
0.231 |
0.866 |
||
| 1999 |
0.029 |
0.502 |
0.454 |
0.866 |
2005 |
0.022 |
0.919 |
0.096 |
0.887 |
||
| F14 |
1997 |
0.035 |
0.893 |
0.165 |
0.863 |
F16 |
2006 |
0.019 |
0.92 |
0.1 |
0.924 |
| 1998 |
0.036 |
0.731 |
0.229 |
0.865 |
2007 |
0.019 |
0.701 |
0.31 |
0.888 |
||
| 1999 |
0.028 |
0.901 |
0.124 |
0.863 |
2008 |
0.012 |
0.625 |
0.381 |
0.868 |
||
| 2000 |
0.024 |
0.606 |
0.346 |
0.873 |
2009 |
0.018 |
0.512 |
0.482 |
0.872 |
||
| 2001 |
0.021 |
0.78 |
0.188 |
0.862 |
F18 |
2010 |
0.021 |
0.146 |
1.525 |
0.851 |
|
| 2002 |
0.017 |
1.018 |
-0.113 |
0.876 |
2011 |
0.02 |
0.51 |
0.901 |
0.851 |
||
| 2003 |
0.017 |
1.096 |
-0.101 |
0.89 |
2012 |
0.017 |
0.483 |
0.508 |
0.852 |
||
| 2000 |
0.028 |
0.578 |
0.485 |
0.878 |
2013 |
0.016 |
0.271 |
2.68 |
0.85 |
Use [spatial analyst tools] - [Map Algebra] - [Raster Calculator] to calculate, as shown in the figure, take 2006 as an example for calculation.
3. Integration within the year
The data obtained by different sensors in the same year are inconsistent. In order to make full use of the data obtained by each independent sensor, and to solve the problem of discontinuity in the data obtained by the sensors, the research follows the following formula to fuse part of the images after mutual correction and saturation correction within the year . The years that need to be integrated within the year include: 1994, 1997-2007. Intra-year fusion includes: ①If the DN value of a certain pixel corresponding to the two images is both 0, then the DN value of the pixel is 0; ②Otherwise, take the average value of the two images as the DN value of the pixel.
In the formula, n = 1994, 1997, 1998, ..., 2007; DNa (n, i) and DNb (n, i) respectively represent the DN of pixel i acquired by two different sensors n years after mutual correction and saturation correction Value; DN(n, i) indicates the DN value of pixel i in n years after fusion correction within the image year. Still use [spatial analyst tools] - [Map Algebra] - [Raster Calculator] to calculate, as shown in the figure, take 2006 as an example to calculate, the correction formula:
Con(("xhjz_f10_1994"!=0) & ("xhjz_f12_1994"!= 0),("xhjz_f10_1994" + "xhjz_f12_1994") / 2,0)
4. Inter-annual correction
After intra-year fusion, there is still the phenomenon that the images of different years are not comparable, so inter-annual correction is performed. The correction basis is: the DN value of the pixel in the next year should not be less than the DN value of the pixel at the same position in the previous year. As shown in formula (4), the inter-annual correction includes: ① If the DN value of a certain pixel in the next year is 0, then the DN value of the pixel at the same position in the previous year is also 0; ② The DN value of a certain pixel in the previous year If the DN value of the cell is greater than the DN value of the pixel at the same position in the next year, the DN value of the previous year is assigned to the DN value of the pixel at the same position in the next year; ③Otherwise, the DN value of the pixel in the next year is itself DN value.
In the formula, n = 1992, 1993, 1994, ..., 2013; DN(n-1, i), DN(n, i), DN(n+1, i) represent the pixel i of the image at n-1 , n, and the DN value of n+1 years. As shown in the correction formula in 2006:
Con("rh1997" == 0,0,Con(("rh1997" > 0) & ("rh1997" > "rh1997"),"rh1997","rh1998"))；
Let us take 2006 as an example, and take a look at the changes after each step of correction.
Finally, according to the above statistical steps, the corrected data is counted, and it can be seen that the current 1992-2013 Daguang remote sensing data presents a fairly smooth rising pattern.
4. References
【1】Lu Xiu, Li Jia, Duan Ping, et al. Correction of DMSP/OLS nighttime light images in China ［J］．Surveying and Mapping Bulletin, 2019 ( 7 ): 127-131. DOI : 10.13474/j. cnki．11-2246．20190234．
【2】Preprocessing of DMSP luminous remote sensing data- Know about