Cross-comparison of Landsat-8 and Landsat-9 data: a three-level approach based on underfly images

Figures & data

Figure 1. Images of five test sites. (a) Yantai, China (175 km²); (b) Xuancheng, China (1185 km²); (c): Rancho Viejo, Mexico (397 km²); (d) Ochiltree, USA (547 km²); and (e) Gibson, Australia (298 km²).

Table 1. Image pairs used for the study.

Image pair	Sensor	Path/Row	Acquisition date	Time of overpass (GMT)	Sun elevation angle (°)	Sun azimuth angle (°)	Area (km^2)
Yantai (China)	Landsat-8 OLI	119/34	14 November 2021	2:29:57	32.12	161.32	175
	Landsat-9 OLI2			2:32:13	32.09	161.19
Xuancheng (China)	Landsat-8 OLI	119/39	14 November 2021	2:31:56	38.41	157.88	1185
	Landsat-9 OLI2			2:34:13	38.37	157.75
Rancho Viejo (Mexico)	Landsat-8 OLI	30/45	14 November 2021	17:24:17	45.54	152.88	397
	Landsat-9 OLI2			17:24:34	45.48	152.74
Ochiltree (USA)	Landsat-8 OLI	30/35	14 November 2021	17:20:18	33.23	160.70	547
	Landsat-9 OLI2			17:20:36	33.20	160.57
Gibson (Australia)	Landsat-8 OLI	103/77	14 November 2021	01:07:56	64.43	80.92	298
	Landsat-9 OLI2			01:10:24	64.32	81.01

Table 2. Statistics of the first-level per-band comparison.

	Band 1	Band 2	Band 3	Band 4	Band 5	Band 6	Band 7
Yantai
Bias	−0.0006	−0.0019	−0.0012	−0.0017	−0.0034	−0.0027	−0.0016
PC (%)	−1.03	−2.68	−1.30	−1.67	−1.94	−1.59	−1.21
R^2	0.807	0.888	0.934	0.941	0.957	0.961	0.958
Xuancheng
Bias	−0.0008	−0.0013	−0.0017	−0.0018	−0.0053	−0.0040	−0.0029
PC (%)	−1.81	−2.70	−2.54	−2.65	−2.32	−2.21	−2.58
R^2	0.863	0.930	0.965	0.973	0.970	0.968	0.966
Rancho Viejo
Bias	0.0008	−0.0005	0.0002	−0.0005	−0.0018	−0.0016	−0.0008
PC (%)	5.55	−2.64	0.60	−1.88	−0.60	−1.10	−1.14
R^2	0.808	0.918	0.976	0.985	0.975	0.983	0.989
Ochiltree
Bias	0.0003	−0.0004	0.0003	−0.0002	−0.0004	0.0003	−0.0002
PC (%)	0.457	−0.463	0.24	−0.12	−0.14	0.09	−0.07
R^2	0.928	0.964	0.976	0.986	0.993	0.987	0.993
Gibson
Bias	−0.0022	−0.0028	−0.0056	−0.0072	−0.0087	−0.0093	−0.0093
PC (%)	−5.23	−5.10	−5.57	−3.62	−3.05	−2.34	−2.63
R^2	0.544	0.813	0.961	0.944	0.957	0.945	0.942
Average bias	−0.0005	−0.0014	−0.0016	−0.0023	−0.0039	−0.0035	−0.0030
Average |PC| (%)	2.81	2.72	2.05	1.99	1.61	1.47	1.53
Average R^2	0.790	0.903	0.962	0.966	0.970	0.969	0.970

Figure 2. Percentage stacked bar chart illustrating the trends of average PC and R² as wavelength increases and a mutually inverse relationship between the two metrics.

Table 3. Statistics of the second- and third-level comparisons.

	LST (^oC)	Wetness	NDVI	NDBSI	RSEI
Yantai
Bias	−0.5778	0.001	−0.0017	0.0013	0.0153
PC (%)	−3.34	1.01	−0.7	9.85	2.86
R^2	0.960	0.953	0.938	0.824	0.902
Xuancheng
Bias	−0.5394	0.0020	0.0005	0.0005	0.0252
PC (%)	−2.67	2.22	0.09	0.64	3.99
R^2	0.967	0.958	0.979	0.927	0.959
Rancho Viejo
Bias	−0.2696	0.0009	0.0020	−0.0028	−0.0104
PC (%)	−0.93	2.06	0.24	−1.04	−1.38
R^2	0.994	0.977	0.984	0.963	0.984
Ochiltree
Bias	−0.2232	−0.0002	−0.0002	0.0007	0.0007
PC (%)	−1.10	−0.09	−0.08	0.53	0.28
R^2	0.925	0.992	0.994	0.994	0.978
Gibson
Bias	0.4235	0.0058	0.0029	0.0035	0.0055
PC (%)	0.95	1.88	1.62	1.51	2.54
R^2	0.959	0.931	0.677	0.756	0.958
Average bias	−0.2373	0.0019	0.0007	0.0006	0.0073
Average |PC| (%)	1.80	1.45	0.55	2.71	2.21
Average R^2	0.961	0.962	0.914	0.893	0.956

Figure 3. Indicator-based comparisons between Landsat-8 (L8) and Landsat-9 (L9) over the five test sites, with scatter plots of L9 vs. L8 (first row of each site), the density scatter plots of bias (L9 – L8) as a function of Landsat-9 indicator values (second row of each site), and the histograms of the differencing images (L9 – L8) (third row of each site).

Figure 3. (Continued).

Figure 3. (Continued).

Figure 4. Composite index (RSEI)-based comparison between Landsat-8 and Landsat-9 over the five test sites, with original true-color images (left of the first row of each site), RSEI image pairs (center and right of the first row of each site), scatter plots of L9 vs. L8 (left of the second row of each site), scatter plots of bias (L9 – L8) as a function of Landsat-9 values (center of the second row of each site), and histograms of the differencing images (L9 – L8) (right of the second row of each site).

Figure 4. (Continued).

Table 4. Statistics of three-level comparisons.

	First level	Second level	Third level	Absolute average
Yantai
PC (%)	1.73	3.85	2.86	2.81
R^2	0.940	0.905	0.902	0.916
Xuancheng
PC (%)	2.50	0.98	3.99	2.49
R^2	0.962	0.955	0.959	0.959
Rancho Viejo
PC (%)	1.32	1.11	−1.38	1.27
R^2	0.971	0.975	0.984	0.977
Ochiltree
PC (%)	0.19	0.23	0.28	0.23
R^2	0.983	0.993	0.978	0.985
Gibson
PC (%)	3.72	1.65	2.54	2.64
R^2	0.927	0.788	0.958	0.891
Average |PC| (%)	1.88	1.63	2.21
Average R^2	0.957	0.933	0.956

Table 5. Statistics of the six land cover types averaged across the five sites.

	Average |PC| of mean (%)			Average R^2
	Band level	Indicator level	Composite index level	Band level	Indicator level	Composite index level
Forest	2.15	1.14	1.24	0.805	0.914	0.933
Crop	1.70	0.97	1.85	0.904	0.958	0.967
Soil	1.15	0.97	3.57	0.879	0.781	0.739
Barren	3.75	1.72	2.62	0.769	0.598	0.699
Built-up	2.48	3.82	4.87	0.730	0.614	0.601
Water	18.62	34.02	15.47	0.521	0.560	0.160

Figure 5. Dynamic data range of each band of Landsat-8 and −9 averaged from the five test sites.

Figure 6. Images of a lake in the Xuancheng site, highlighting the radiometric solution difference in identifying dark water areas between Landsat-8 and −9. (a) and (e): Landsat-8 and −9 false-color images (RGB: bands 5, 4, 3); (b) and (c): Pseudo-color images of Landsat-8’s bands 1 and 2; (f) and (g): Pseudo-color images of Landsat-9’s bands 1 and 2; (d) and (h): Pseudo-color images with a zoomed-in view of the white dot in Figure 6(a,e).

Data availability statement

The remote sensing image data utilized in this study can be obtained from the United States Geological Survey website (https://earthexplorer.usgs.gov). The other data are available from the corresponding author upon request.