Introduction

The Earth Surface Mineral Dust Source Investigation (EMIT) sensor is a hyperspectral electro-optical instrument developed by the National Aeronautics and Space Administration. EMIT is installed on the International Space Station and has been operating since July 2022 (Thompson and others, 2024). This section provides a background of the EMIT sensor and the scope of its quality assessment.

System Description

This section describes the satellite and operational details and provides information about the EMIT sensor.

Procedures

ECCOE has established standard processes to identify Earth observing systems of interest and to assess the geometric, radiometric, and spatial qualities of data products from these systems (Cantrell and Christopherson, 2024).

The assessment steps are as follows:

The specific procedures required to preprocess hyperspectral data are as follows:

Data analysis and test results are maintained at the USGS Earth Resources Observation and Science Center by the ECCOE project.

Measurements

The observed USGS measurements are listed in table 3. Details about the methodologies used are outlined in the “Analysis” section.

Analysis

This section of the report describes the geometric and radiometric performance of the EMIT sensor.

Geometric Performance

The geometric performance for EMIT is characterized in terms of the band-to-band alignment and image-to-image relative geometric accuracy.

Band to Band

For this analysis, each band of the EMIT imagery was registered against one reference band. For the visible and near infrared (VNIR) bands, a band centered at 559 nanometers (nm; band 24) was used as a reference. For the shortwave infrared bands, a band centered at 1,610 nm (band 165) was used. The relative differences are small, which indicates high quality of band-to-band performance. The scene identifier used as an example image to compute band-to-band error is EMIT_L1B_RAD_001_20230406T181557_2309612_005 and is shown in figure 1.

The grid system and error vectors for band 24 are shown in figure 1 as an example using the Worldwide Reference System-2 (WRS-2) 39/37 scene. The red arrow shows the relative error vector for each yellow grid, where x and y vector components represent easting and northing errors, respectively. Several grids with missing arrows represent the outliers. The mean and root mean square error values are computed from the grids for band 24 and represent the dots for band 24 in figures 2 and 3. The band-to-band results for the WRS-2 147/44 scene are similar in figures 4 and 5. Similar band-to-band results for the WRS-2 233/82 scene are shown in figures 6 and 7. The mean band-to-band easting and northing errors with associated root mean square error are shown in tables 4 (band 24 reference) and 5 (band 165 reference).

Table 4.    

Band-to-band results for reference band 24 (in pixels).

[ID, identifier; RMSE, root mean square error; WRS-2, Worldwide Reference System-2]

Table 4.    Band-to-band results for reference band 24 (in pixels).

Scene ID	Mean error (easting)	Mean error (northing)	RMSE (easting)	RMSE (northing)
WRS-2 path/row 39/37 (EMIT_L1B_RAD_001_20230406T181557_2309612_005)	−0.143 to 0.008	−0.160 to 0.076	0.082 to 0.251	0.079 to 0.234
WRS-2 path/row 147/44 (EMIT_L1B_RAD_001_20230404T055003_2309404_003)	−0.088 to 0.050	−0.171 to 0.109	0.019 to 0.161	0.019 to 0.201
WRS-2 path/row 233/82 (EMIT_L1B_RAD_001_20240307T140814_2406709_006)	−0.240 to 0.032	−0.035 to 0.166	0.066 to 0.278	0.055 to 0.228

Table 5.    

Band-to-band results for reference band 165 (in pixels).

[ID, identifier; RMSE, root mean square error; WRS-2, Worldwide Reference System-2]

Table 5.    Band-to-band results for reference band 165 (in pixels).

Scene ID	Mean error (easting)	Mean error (northing)	RMSE (easting)	RMSE (northing)
WRS-2 path/row 39/37 (EMIT_L1B_RAD_001_20230406T181557_2309612_005)	−0.214 to 0.053	−0.131 to 0.206	0.206 to 0.546	0.203 to 0.502
WRS-2 path/row 147/44 (EMIT_L1B_RAD_001_20230404T055003_2309404_003)	−0.104 to 0.095	−0.136 to 0.210	0.094 to 0.401	0.083 to 0.408
WRS-2 path/row 233/82 (EMIT_L1B_RAD_001_20240307T140814_2406709_006)	−0.355 to 0.064	−0.291 to 0.021	0.173 to 0.453	0.159 to 0.490

Band-to-band full shortwave infrared spectral geometric error estimation was performed using band 165 as a reference. The results for the WRS-2 39/37 scene are shown in figure 8 for easting error and in figure 9 for northing error. Similar band-to-band results for the WRS-2 147/44 scene are shown in figures 10 and 11, and results for the WRS-2 233/82 scene are shown in figures 12 and 13. Compared to VNIR bands, two noisy spectral regions are centered around 1.4 and 1.9 micrometers, which is the strong absorption band from atmospheric water vapor.

Image to Image

For the image-to-image analysis, spectrally resampled EMIT data were used. The resampling is based on the spectral response of the Landsat 8 and 9 OLI. Three scene pairs between EMIT and OLI were used for image-to-image analysis. A normalized cross-correlation matrix was computed, and its local maxima were determined to estimate the mean error and root mean square error results listed in table 6 with results represented in pixels at a 60-m ground sample distance. For each of the three EMIT images, geometric error maps show the directional shift and relative magnitude of the shift, when compared with the Landsat 8 and 9 OLI, in figures 14 through 22.

Table 6.    

Geometric error of Earth Surface Mineral Dust Source Investigation relative to Landsat 8 and 9 Operational Land Imager (OLI) imagery.

[ID, identifier; RMSE, root mean square error; WRS-2, Worldwide Reference System-2; m, meter]

Table 6.    Geometric error of Earth Surface Mineral Dust Source Investigation relative to Landsat 8 and 9 Operational Land Imager (OLI) imagery.

Scene ID	Mean error (easting)	Mean error (northing)	RMSE (easting)	RMSE (northing)
WRS2 Path/Row 39/37 LC08_L1TP_039037_20230409_20230420_02_T1 (EMIT_L1B_RAD_001_20230406T181557_2309612_005)	−1.341 m (−0.022 pixel)	40.439 m (0.67 pixel)	21.431 m (0.357 pixel)	43.810 m (0.730 pixel)
WRS2 Path/Row 147/44 LC09_L1TP_147044_20230406_20230406_02_T1 (EMIT_L1B_RAD_001_20230404T055003_2309404_003)	−15.966 m (−0.266 pixel)	43.844 m (0.731 pixel)	23.529 m (0.392 pixel)	46.572 m (0.776 pixel)
WRS2 Path/Row 233/82 LC08_L1TP_233082_20240309_20240316_02_T1 (EMIT_L1B_RAD_001_20240307T140814_2406709_006)	−4.708 m (−0.078 pixel)	28.804 m (0.48 pixel)	18.243 m (0.304 pixel)	33.175 m (0.553 pixel)

Radiometric Performance

For the radiometric performance analysis, cloud-free regions of interest were selected within three near-coincident EMIT and Landsat 8 and 9 OLI scene pairs. Top of Atmosphere (TOA) reflectance comparison results are listed in table 7. Once the relative georeferencing error between the Landsat 8 and 9 OLI and EMIT is corrected, TOA reflectance values from the two sensors are extracted. The scatterplots, figures 23 through 25, are drawn in a way that the x-axis is the Landsat 8 or 9 OLI reflectance, and the y-axis is the EMIT reflectance. Thus, the linear regression represents TOA reflectance relative to that of the reference sensor. Ideally, the slope should be near unity and the offset should be near zero, and if the slope is greater than unity, that means the comparison sensor tends to overestimate TOA reflectance compared to the reference sensor.

Table 7.    

Top of Atmosphere reflectance comparison for Earth Surface Mineral Dust Source Investigation against Landsat 8 Operational Land Imager (OLI) scene pairs.

[ID, identifier; B, band; CA, coastal aerosol; NIR, near infrared; SW, shortwave infrared; WRS-2, Worldwide Reference System-2; %, percent; R², coefficient of determination]

Table 7.    Top of Atmosphere reflectance comparison for Earth Surface Mineral Dust Source Investigation against Landsat 8 Operational Land Imager (OLI) scene pairs.

Scene IDs	Statistics	B1 (CA)	B2 (blue)	B3 (green)	B4 (red)	B5 (NIR)	B6 (SW1)	B7 (SW2)
WRS2 Path/Row 39/37 LC08_L1TP_039037_20230409_20230420_02_T1 (EMIT_L1B_RAD_001_20230406T181557_2309612_005)	Uncertainty (%)	4.355	5.524	7.181	8.224	7.722	7.900	8.112
	R2	0.897	0.901	0.907	0.912	0.915	0.916	0.923
	Regression offset	0.016	0.016	0.018	0.020	0.024	0.023	0.017
	Regression slope	0.900	0.925	0.934	0.923	0.926	0.949	0.948
WRS2 Path/Row 147/44 LC09_L1TP_147044_20230406_20230406_02_T1 (EMIT_L1B_RAD_001_20230404T055003_2309404_003)	Uncertainty (%)	2.231	3.049	4.441	6.146	5.208	6.884	8.495
	R2	0.740	0.757	0.791	0.784	0.843	0.873	0.843
	Regression offset	−0.016	−0.008	−0.002	0.011	0.017	0.019	0.021
	Regression slope	0.985	0.961	0.961	0.884	0.916	0.942	0.873
WRS2 Path/Row 233/82 LC08_L1TP_233082_20240309_20240316_02_T1 (EMIT_L1B_RAD_001_20240307T140814_2406709_006)	Uncertainty (%)	7.170	8.791	10.971	11.995	11.956	13.572	13.744
	R2	0.827	0.829	0.838	0.849	0.863	0.871	0.864
	Regression offset	0.025	0.022	0.020	0.019	0.019	0.014	0.009
	Regression slope	0.837	0.869	0.890	0.893	0.912	0.943	0.908

A band-by-band graphical comparison of the WRS-2 39/37, WRS-2 147/44, and WRS-2 233/82 scene image pairs are shown in figures 23, 24, and 25. respectively.

Comparison to Radiometric Calibration Network (RadCalNet)

RadCalNet provides surface and atmosphere measurements using automated instruments calibrated traceably to SI and with known uncertainties (Bouvet and others, 2019). RadCalNet data are typically used for postlaunch radiometric evaluation of multispectral and hyperspectral optical satellite sensors (Jing and others, 2019; Shrestha and others, 2021). EMIT hyperspectral data quality is also evaluated using coincident measurements between EMIT and RadCalNet. Four coincident measurements between EMIT and RadCalNet data are available for Railroad Valley Playa located in east central Nevada (USA); however, the coincident measurements were not indicated on other RadCalNet sites. EMIT reflectance on Railroad Valley Playa represents a 0.5-km x 0.5-km surface centered at 38.497° N. and 11.690° W. as shown by a red solid square in figure 26 (Bouvet and others, 2019).

RadCalNet provides TOA with associated uncertainties over the wavelength range of 400 to 2,500 nm with 10-nm spectral sampling completed at 30-minute intervals (Bouvet and others, 2019). RadCalNet measurements bounding EMIT acquisition time are linearly interpolated to estimate RadCalNet measurement during EMIT overpass time. Normalized EMIT TOA reflectance at Railroad Valley Playa for four dates (given in YYYYMMDD format) is shown in figure 27. Normalized EMIT TOA reflectance is calculated by dividing EMIT TOA reflectance by the coincident RadCalNet measurement. The error bar of normalized TOA reflectance is calculated using the standard deviation of EMIT TOA reflectance and RadCalNet measurement uncertainty. Data from the absorption regions, 1,350 to 1,450 nm and 1,800 to 1,950 nm, are not displayed in figure 27. EMIT measurements from different dates show different level of agreement with RadCalNet measurements. EMIT acquisition on 20230804 shows the best agreement with RadCalNet measurement where it agrees with RadCalNet measurement within 3 percent across most of the channels. The highest disagreement between EMIT and RadCalNet measurements is observed in the 20230823 acquisition where EMIT agrees with RadCalNet measurement within 8 percent across most of the spectral channels.

The mean and standard deviation of normalized EMIT TOA reflectance in figure 27 are shown in figure 28. Mean normalized EMIT TOA reflectance agrees with RadCalNet measurement within 5 percent across most of the spectral channels. The error bar of the mean normalized EMIT TOA reflectance is the standard deviation of four normalized EMIT TOA reflectance shown in figure 27. Spectral channels at shorter wavelengths (less than 450 nm) and near atmospheric absorption regions show more disagreement with RadCalNet measurements than the rest of the spectral channels.

Summary and Conclusions

This report summarizes the performance of the Earth Surface Mineral Dust Source Investigation (EMIT) sensor based on the U.S. Geological Survey Earth Resources Observation and Science Cal/Val Center of Excellence (ECCOE) system characterization process. In summary, we have determined that the EMIT sensor has a band-to-band geometric performance in the range of −0.355 to 0.210 pixel with a few exceptions in the shortwave infrared channels. Geometric offset relative to Operational Land Imager ranges from −15.966 meters (−0.266 pixel) to 43.844 meters (0.731 pixel). Offset of a radiometric comparison ranges from −0.016 to 0.025, and slope of a radiometric comparison ranges from 0.837 to 0.985. EMIT agreed with Radiometric Calibration Network (RadCalNet) measurements within 5 percent across most of the spectral channels.

In conclusion, the team has completed an ECCOE standardized system characterization of the EMIT sensing system. Although the team followed characterization procedures that are standardized across the many sensors and sensing systems under evaluation, these procedures are customized to fit the individual sensor as was done with EMIT. The team has acquired the data, defined proper testing methodologies, carried out comparative tests against specific references, recorded measurements, completed data analyses, and quantified sensor performance accordingly. The team also endeavored to retain all data, measurements, and methods. This is key to ensure that all data and measurements are archived and accessible and that the performance results are reproducible.

The ECCOE project and associated Joint Agency Commercial Imagery Evaluation partners are always interested in reviewing sensor and remote sensing application assessments and would like to review and discuss information on similar data and product assessments. If you would like to discuss system characterization with the U.S. Geological Survey ECCOE and (or) the Joint Agency Commercial Imagery Evaluation team, please email us at eccoe@usgs.gov.