Figure 1.
DEM data from the study area of Zhaotong City, Yunnan Province. The red frame shows the location of Yunnan Province in China, and the green frame shows the location of Zhaotong City in Yunnan Province.
Figure 1.
DEM data from the study area of Zhaotong City, Yunnan Province. The red frame shows the location of Yunnan Province in China, and the green frame shows the location of Zhaotong City in Yunnan Province.
Figure 2.
Zhaotong City’s land cover types.
Figure 2.
Zhaotong City’s land cover types.
Figure 3.
The technical flow chart of this paper.
Figure 3.
The technical flow chart of this paper.
Figure 4.
Image distribution required to fully cover Zhaotong City. (a) Sentinel-2 image distribution; (b) Landsat-8 OLI image distribution.
Figure 4.
Image distribution required to fully cover Zhaotong City. (a) Sentinel-2 image distribution; (b) Landsat-8 OLI image distribution.
Figure 5.
The overlapping area of the benchmark image and the image to be corrected in different situations. (a) Images from the same sensor (Sentinel-2) on different days; (b) Images from different sensors (Sentinel-2 MSI and Landsat-8 OLI) on different days.
Figure 5.
The overlapping area of the benchmark image and the image to be corrected in different situations. (a) Images from the same sensor (Sentinel-2) on different days; (b) Images from different sensors (Sentinel-2 MSI and Landsat-8 OLI) on different days.
Figure 6.
Spectral reflectance curves of different land cover types in the NIR, red and green bands.
Figure 6.
Spectral reflectance curves of different land cover types in the NIR, red and green bands.
Figure 7.
Apparent reflectance value of near-infrared images on a certain day, where the pixels in the red frame are under the shadow of clouds.
Figure 7.
Apparent reflectance value of near-infrared images on a certain day, where the pixels in the red frame are under the shadow of clouds.
Figure 8.
Time-consistent image composite for April 2021: (a) Benchmark image; (b) vector data; (c) correction image; (d) time-consistent image, synthesized from (a,c).
Figure 8.
Time-consistent image composite for April 2021: (a) Benchmark image; (b) vector data; (c) correction image; (d) time-consistent image, synthesized from (a,c).
Figure 9.
An example of the process of eliminating the erroneous pixels in a Sentinel-2 image. (a) Original image, where the pixels in the red frame are under the shadow of clouds; (b) experimental image, with no pixels in the red frame; (c) comparison image, where the pixels in the red frame are eliminated.
Figure 9.
An example of the process of eliminating the erroneous pixels in a Sentinel-2 image. (a) Original image, where the pixels in the red frame are under the shadow of clouds; (b) experimental image, with no pixels in the red frame; (c) comparison image, where the pixels in the red frame are eliminated.
Figure 10.
Comparison of apparent reflectance scatter points for correction before and after eliminating erroneous pixels. (a) Correction before eliminating; (b) correction after eliminating. N is the number of image pixels, the black dashed line is the 1:1 line, and the red solid line is the regression line of the scatter plot.
Figure 10.
Comparison of apparent reflectance scatter points for correction before and after eliminating erroneous pixels. (a) Correction before eliminating; (b) correction after eliminating. N is the number of image pixels, the black dashed line is the 1:1 line, and the red solid line is the regression line of the scatter plot.
Figure 11.
Comparison of the apparent reflectance scatter points for land cover types correction and whole test area correction. N is the number of image pixels, the black dashed line is the 1:1 line, and the red solid line is the regression line of the scatter plot. From left to right: Whole test area, cropland, forest, grassland.
Figure 11.
Comparison of the apparent reflectance scatter points for land cover types correction and whole test area correction. N is the number of image pixels, the black dashed line is the 1:1 line, and the red solid line is the regression line of the scatter plot. From left to right: Whole test area, cropland, forest, grassland.
Figure 12.
Comparison of apparent reflectance scatter points for land cover types correction and classification pixel means correction. (a) Classification pixel means correction; (b) land cover types correction. N is the number of image pixels, the black dashed line is the 1:1 line, and the red solid line is the regression line of the scatter plot.
Figure 12.
Comparison of apparent reflectance scatter points for land cover types correction and classification pixel means correction. (a) Classification pixel means correction; (b) land cover types correction. N is the number of image pixels, the black dashed line is the 1:1 line, and the red solid line is the regression line of the scatter plot.
Figure 13.
Spectral response function of Sentinel-2 MSI and Landsat-8 OLI.
Figure 13.
Spectral response function of Sentinel-2 MSI and Landsat-8 OLI.
Figure 14.
(a) True difference image under the same sensor, i.e., the difference between the pixels of the benchmark image and the image to be corrected; (b) Corrected difference image under the same sensor, i.e., the difference between the pixels of the benchmark image and the corrected image. (c) True difference image under different sensors; (d) Corrected difference image under different sensors. Pixels with −0.02 ≤ difference ≤ 0.02 are defined as comparable; the rest of the pixels are incomparable.
Figure 14.
(a) True difference image under the same sensor, i.e., the difference between the pixels of the benchmark image and the image to be corrected; (b) Corrected difference image under the same sensor, i.e., the difference between the pixels of the benchmark image and the corrected image. (c) True difference image under different sensors; (d) Corrected difference image under different sensors. Pixels with −0.02 ≤ difference ≤ 0.02 are defined as comparable; the rest of the pixels are incomparable.
Figure 15.
Synthetic image comparison: (a) Shows a column of direct mosaic images and (b) shows a column of time-consistent images. A is the junction of images from different sensors; B is the junction of images from the same sensor.
Figure 15.
Synthetic image comparison: (a) Shows a column of direct mosaic images and (b) shows a column of time-consistent images. A is the junction of images from different sensors; B is the junction of images from the same sensor.
Figure 16.
Comparison of grassland growth classifications from different synthetic images. (a) Is a row of direct mosaic images; (b) is a row of time-consistent images; (c) is a row of high-scoring images, which is used to verify the accuracy of the wrong classification of grassland growth in the method in this paper. Grassland growth in A was misclassified as inferior growth rather than steady growth. Grassland growth in B was misclassified as steady growth rather than superior growth.
Figure 16.
Comparison of grassland growth classifications from different synthetic images. (a) Is a row of direct mosaic images; (b) is a row of time-consistent images; (c) is a row of high-scoring images, which is used to verify the accuracy of the wrong classification of grassland growth in the method in this paper. Grassland growth in A was misclassified as inferior growth rather than steady growth. Grassland growth in B was misclassified as steady growth rather than superior growth.
Figure 17.
Time-consistent monthly images of Zhaotong City in 2021.
Figure 17.
Time-consistent monthly images of Zhaotong City in 2021.
Figure 18.
Grassland growth process curve and NDVI distance level in 2021.
Figure 18.
Grassland growth process curve and NDVI distance level in 2021.
Figure 19.
Classification of grassland growth from July to October 2021. Classify the NDVI values into 5 classes.
Figure 19.
Classification of grassland growth from July to October 2021. Classify the NDVI values into 5 classes.
Table 1.
Details of the corresponding bands of Sentinel-2 MSI and Landsat-8 OLI used in this study.
Table 1.
Details of the corresponding bands of Sentinel-2 MSI and Landsat-8 OLI used in this study.
Sentinel-2 MSI | Landsat-8 OLI |
---|
Waveband | Spatial Resolution/m | Range/nm | Center/nm | Waveband | Spatial Resolution/m | Range/nm | Center/nm |
---|
B3 (Green) | 10 | 543–578 | 560 | B3 (Green) | 30 | 533–590 | 561 |
B4 (Red) | 10 | 650–680 | 664 | B4 (Red) | 30 | 636–673 | 655 |
B8 (NIR) | 10 | 785–900 | 843 | B5 (NIR) | 30 | 851–879 | 865 |
Table 2.
The dates of the experimental images used to construct time-consistent images in this study.
Table 2.
The dates of the experimental images used to construct time-consistent images in this study.
Month | Benchmark Image Sentinel-2 MSI | Image to be Corrected |
---|
Sentinel-2 MSI | Landsat-8 OLI |
---|
2021-1 | 01/01 | 01/14, 01/26 | - |
2021-2 | 02/10 | 02/20 | - |
2021-3 | 03/17 | 03/27 | |
2021-4 | 04/21 | 04/26 | 04/26 |
2021-5 | 05/01 | 05/21 | - |
2021-6 | 06/05 | - | - |
2021-7 | 07/20 | 07/25, 07/13, 07/30 | - |
2021-8 | 08/04 | 08/02 | - |
2021-9 | 09/23 | 09/21 | - |
2021-10 | 10/03 | 10/01, 10/18 | 10/01 |
2021-11 | 11/17 | 11/05 | 11/26 |
2021-12 | 12/22 | - | - |
Table 3.
Based on statistical comparisons between corrections before and after eliminating erroneous pixels.
Table 3.
Based on statistical comparisons between corrections before and after eliminating erroneous pixels.
Category | Correction Method | Equation | R2 | RMSE |
---|
Cropland | After eliminating | Y = 0.9769x − 0.0008 | 0.9361 | 0.0078 |
Before eliminating | Y = 0.8188x + 0.0486 | 0.8049 | 0.0224 |
Forest | After eliminating | Y = 1.0005x − 0.0117 | 0.9558 | 0.0095 |
Before eliminating | Y = 0.7790x + 0.0659 | 0.7493 | 0.0402 |
Grassland | After eliminating | Y = 0.9861x − 0.0052 | 0.9561 | 0.0061 |
Before eliminating | Y = 0.9285x + 0.0135 | 0.8943 | 0.0161 |
Table 4.
Statistical comparison between the land cover types correction and correction of the whole test area based on images from the same sensor.
Table 4.
Statistical comparison between the land cover types correction and correction of the whole test area based on images from the same sensor.
Different Correction Methods | Category | Equation | R2 | RMSE |
---|
Whole test area | - | Y = 0.9568x − 0.0004 | 0.9005 | 0.0130 |
Land cover types | Cropland | Y = 0.9769x − 0.0008 | 0.9361 | 0.0078 |
Forest | Y = 1.0005x − 0.0117 | 0.9558 | 0.0095 |
Grassland | Y = 0.9861x − 0.0052 | 0.9561 | 0.0061 |
Table 5.
Comparison statistics of pixel means between the corrected images based on different methods and the benchmark image.
Table 5.
Comparison statistics of pixel means between the corrected images based on different methods and the benchmark image.
Image | Correction Method | Cropland | Forest | Grassland |
---|
Benchmark image (04.21) | - | 0.2281 | 0.2272 | 0.2588 |
Correction image (04.26) | Whole test area | 0.2259 | 0.2304 | 0.2527 |
Land cover types | 0.2280 | 0.2296 | 0.2574 |
Table 6.
Statistical comparison between classification pixel means correction and land cover types correction based on the images from different sensors.
Table 6.
Statistical comparison between classification pixel means correction and land cover types correction based on the images from different sensors.
Category | Correction Method | Equation | R2 | RMSE |
---|
Cropland | Classification pixel means | Y = 0.8480x + 0.0253 | 0.8276 | 0.0097 |
Land cover types | Y = 0.8317x + 0.0299 | 0.7097 | 0.0165 |
Forest | Classification pixel means | Y = 0.8647x + 0.0189 | 0.9108 | 0.0117 |
Land cover types | Y = 0.8468x − 0.0232 | 0.8352 | 0.0184 |
Grassland | Classification pixel means | Y = 0.8381x + 0.0212 | 0.8717 | 0.0086 |
Land cover types | Y = 0.8232x + 0.0249 | 0.7808 | 0.0140 |
Table 7.
Statistics of the pixel’s mean values of the difference images from same sensor and from different sensors.
Table 7.
Statistics of the pixel’s mean values of the difference images from same sensor and from different sensors.
Type | True Difference Image | Corrected Difference Image |
---|
Same Sensor | −0.0106 | 0.0019 |
Different Sensors | −0.0221 | −0.0076 |
Table 8.
Grassland growth area statistics based on different image synthesis methods (km2).
Table 8.
Grassland growth area statistics based on different image synthesis methods (km2).
Image | Inferior Growth | Steady Growth | Superior Growth |
---|
Direct mosaic | 1332.03 | 1717.99 | 358.04 |
Time-consistent | 1082.26 | 1817.99 | 507.81 |
Table 9.
Statistics of grassland growth misclassification based on direct mosaic images.
Table 9.
Statistics of grassland growth misclassification based on direct mosaic images.
Misclassification Situation | Area/km2 | Proportion of the Total Grassland Area |
---|
Steady growth → Inferior growth | 249.77 | 7.33% |
Superior growth → Steady growth | 149.77 | 4.39% |