Joint Agency Commercial Imagery Evaluation (JACIE) Workshop-2019, Reston, VA, USA, 9/24 – 9/26, 2019 Michele Kuester and Tina Ochoa Improvements in Calibration, and Validation of the Absolute Radiometric Response of MAXAR Earth-Observing Sensors
Joint Agency Commercial Imagery Evaluation (JACIE) Workshop-2019, Reston, VA, USA, 9/24 – 9/26, 2019
Michele Kuester and Tina Ochoa
Improvements in Calibration, and Validation of the Absolute Radiometric Response of MAXAR Earth-Observing Sensors
© 2019 Maxar Technologies 2
New facility in Fort Lupton, CO, USA
2018v0 Calibration – updates to methodology & validation
Stability of MAXAR sensors over Libya-4 in Saharan Desert
Comparisons to Landsat & Sentinel
OVERVIEW
© 2019 Maxar Technologies 3
3
Located in Fort Lupton, CO, USA
3 acres – plenty of room for tarps
Very rural location (less pollution, less adjacency from nearby neighborhoods)
A laboratory space for instrument maintenance and on-site work
Web-enabled weather stations alert us to non-ideal deploy conditions (clouds, wet soil, winds, temperatures)
MAXAR builds new calibration facility in 2019
© 2019 Maxar Technologies 4
4
Thuillier in 2003 as recommended by the Committee on Earth Observation Satellites (CEOS) used for the top-of-atmosphere solar irradiance (ChKur and WRC are available upon request)
Surface reflectance measured at sensor view angles
NASA GODDARD AERONET linked Cimel Sun Photometer
− Aerosol optical depth
− Column water vapor, ozone and carbon dioxide
− Single scattering albedo
− Asymmetry parameter
Online weather station
− Temperature
− Pressure
− UV and VNIR solar irradiance
− Wind velocity
− Rain bucket and soil moisture probes
− All-sky camera
MAXAR collects in-situ measurements on-site to best model surface and atmospheric behavior during sensor overpass
© 2019 Maxar Technologies 5
When using imagery with other datasets or performing atmospheric correction it is advised to match solar curves across project
Five years of ground truth on calibration tarps
A more robust surface reflectance measurement that uses angle dependent calibration (derived from measurements at UofA) of the reference Spectralon panel
What solar curve is being used makes a big difference (particularly in SWIR)
Automation software employed to decrease human error
Employ statistics to remove outlier data (versus by hand)
*.IMD abscalFactors checked for inconsistencies
2018v0 Calibration Improvements
© 2019 Maxar Technologies 6
* Committee on Earth Observation Satellites Radiometric Calibration Network portal: radcalnet.org
We now have enough coincident data over Railroad Valley, NV, USA (RVUS) to create meaningful validation plots
Only imagery with < 20-deg off-nadir angle are used
Next slides show pre-launch, 2016v0 and 2018v0 absolute radiometric calibrations on MAXAR sensor imagery compared to RadCalNet Truth
The aim of Absolute radiometric calibration is to assign absolute scientific units to pixels based on radiometric parameters such as scan direction, scan rate, TDI level, and DN value. For World-View class sensors, the absolute radiometric calibration product specification for ninety-five percent of products is:• < 10% of dynamic range there shall be no specification• Between 10% and 85% of the dynamic range the absolute radiometric response specifications shall not exceed +/- 10% uncertainty for VNIR sensors and +/- 15% for SWIR sensors• > 85% of dynamic range there shall be no specification
MAXAR is within specification as shown over RadCalNet*
© 2019 Maxar Technologies 75 Images
RVUS uncertainty reported as 5.3% @400nm and 3.8% @850nm
Pre-Launch 2016v0 2018v0
No On-orbit Cal%Dif to
RadCalNetOn-orbit Cal Adjustment
%Dif to RadCalNet
On-orbit Cal Adjustment
%Dif to RadCalNet
Gain Offset RVUS Gain Offset RVUS GAIN OFFSET RVUSPAN 1.000 0.000 0.942 -2.704 1.001 0.000BLUE 1.000 0.000 -1.246 0.988 -5.736 -6.288 1.041 0.000 2.480GREEN 1.000 0.000 2.869 0.936 -3.546 -6.012 0.972 0.000 0.343RED 1.000 0.000 4.252 0.952 -2.512 -2.484 0.979 0.000 2.868NIR 1.000 0.000 2.321 0.961 -3.300 -4.930 0.951 0.000 -1.978
*Note that pre-launch calibration could also be considered good for GE01 –compares < 5% to RadCalNet
Thuillier Solar Curve
2018v0 AbsRadCal GE01 within 5% difference of RVUS RadCalNet*
© 2019 Maxar Technologies 812 Images
RVUS uncertainty reported as 5.3% @400nm and 3.8% @850nm
Pre-Launch 2016v0 2018v0
No On-orbit Cal%Dif to
RadCalNetOn-orbit Cal Adjustment
%Dif to RadCalNet
On-orbit Cal Adjustment
%Dif to RadCalNet
Gain Offset RVUS Gain Offset RVUS Gain Offset RVUSPAN 1.000 0.000 0.942 -2.704 0.949 -5.523COASTAL 1.000 0.000 -4.351 1.151 -7.478 3.908 1.203 -11.839 2.788BLUE 1.000 0.000 11.141 0.988 -5.736 5.927 1.002 -9.835 4.108GREEN 1.000 0.000 14.995 0.936 -3.546 5.316 0.953 -7.218 4.401YELLOW 1.000 0.000 15.403 0.949 -3.564 7.175 0.946 -5.675 3.876RED 1.000 0.000 14.308 0.952 -2.512 7.094 0.955 -5.046 6.056REDEDGE 1.000 0.000 10.974 0.974 -4.120 4.854 0.980 -6.114 2.592NIR1 1.000 0.000 10.906 0.961 -3.300 3.474 0.966 -5.096 2.675NIR2 1.000 0.000 8.796 1.002 -2.891 4.886 1.010 -4.059 2.168
Thuillier Solar Curve
2018v0 AbsRadCal WV02 VNIR within 5% difference of RVUS RadCalNet (excepting Red Band at 6%)
© 2019 Maxar Technologies 910 Images
RVUS uncertainty reported as 5.3% @400nm and 3.8% @850nm
Pre-Launch 2016v0 2018v0
No On-orbit Cal%Dif to
RadCalNetOn-orbit Cal Adjustment
%Dif to RadCalNet
On-orbit Cal Adjustment
%Dif to RadCalNet
Gain Offset RVUS Gain Offset RVUS Gain Offset RVUSPAN 1.000 0.000 0.950 -3.629 0.955 -5.505COASTAL 1.000 0.000 20.544 0.905 -8.604 2.517 0.938 -13.099 0.471BLUE 1.000 0.000 12.189 0.940 -5.809 1.872 0.946 -9.409 1.264GREEN 1.000 0.000 11.723 0.938 -4.996 1.768 0.958 -7.771 1.276YELLOW 1.000 0.000 9.806 0.962 -3.649 3.409 0.979 -5.489 2.064RED 1.000 0.000 9.821 0.964 -3.021 3.947 0.969 -4.579 3.446REDEDGE 1.000 0.000 5.807 1.000 -4.521 2.530 1.027 -5.552 2.575NIR1 1.000 0.000 11.701 0.961 -5.522 2.536 0.977 -6.508 2.328NIR2 1.000 0.000 10.009 0.978 -2.992 3.679 1.007 -3.699 3.289
Thuillier Solar Curve
2018v0 AbsRadCal WV03 VNIR within 3.5% difference of RVUS RadCalNet
© 2019 Maxar Technologies 10
10 imagesRVUS Uncertainty 3.7% @1000nm and 3.6% @1550 nm
Pre-Launch 2016v0 2019v0Thuillier*
No On-orbit Cal
%Dif to RadCal
NetOn-orbit Cal Adjustment
%Dif to RadCal
NetOn-orbit Cal Adjustment
%Dif to RadCal
NetBAND Gain Offset RVUS Gain Offset RVUS GAIN OFFSET RVUSSWIR1 1.000 0.000 0.360 1.200 -5.546 8.5413 1.030 0.000 3.371SWIR2 1.000 0.000 0.276 1.227 -2.600 14.334 1.052 0.000 5.491SWIR3 1.000 0.000 -0.687 1.199 -2.309 10.269 0.992 0.000 -1.482SWIR4 1.000 0.000 -0.906 1.196 -1.676 10.462 1.014 0.000 0.481SWIR5 1.000 0.000 -3.001 1.262 -0.705 14.255 1.012 0.000 -1.837SWIR6 1.000 0.000 -5.813 1.314 -0.669 14.510 1.082 0.000 1.910SWIR7 1.000 0.000 -6.348 1.346 -0.512 17.909 1.056 0.000 -1.103SWIR8 1.000 0.000 -7.103 1.376 -0.372 19.328 1.101 0.000 2.279
2019v0 AbsRadCal WV03 SWIR within 5.5% difference of RVUS RadCalNet
© 2019 Maxar Technologies 11
A smooth transition from WorldView-3 VNIR to SWIR
© 2019 Maxar Technologies 12Best is to match the calibration with the solar irradiance curve that is used in your atmospheric correction method
Recall that RVUS data uses Thuillier
Important differences in TOA Radiance exist with calibration derived from different solar irradiance curves
© 2019 Maxar Technologies 13
*L. Helder, Dennis & Basnet, Bikash & L. Morstad, Daniel. (2010). Optimized identification of worldwide radiometric pseudo-invariant calibration sites. Canadian Journal of Remote Sensing. 36. 527-539. 10.5589/m10-085.
TOA Reflectance of the green band for GE01, WV02, WV03 and L8 shown over Pseudo-invariant site Libya4, Saharan Desert Libya4 can achieve variabilities as low as 2-
3% in the visible and near infrared (VNIR)* Data from 2013 – 2019 has no discernable
slope Variability up to 4% with WorldView-2 if view
angle is not restricted. Restricting sensor off-nadir angle to < 20 deg shows all sensors within uncertainty of the site reflectance variability
ONA < 20 AVERAGE STDEV PERCENT UNC # IMAGERY USEDGeoEye-1 0.332 0.008 2.5% 37
WorldView-2 0.338 0.009 2.6% 61WorldView-3 0.329 0.007 2.2% 40
Landsat-8 0.338 0.003 0.7% 8
ONA < 40 AVERAGE STDEV PERCENT UNC # IMAGERY USEDGeoEye-1 0.331 0.009 2.7% 60
WorldView-2 0.342 0.013 3.9% 96WorldView-3 0.332 0.011 3.4% 83
Landsat-8 0.338 0.003 0.7% 8
MAXAR sensors are radiometrically stable
© 2019 Maxar Technologies 14
Band Slope Intercept Average StdDevCoastal -8.61E-06 0.232 0.218 0.011Blue -3.86E-06 0.280 0.274 0.010Green -2.25E-06 0.371 0.368 0.009Yellow -2.13E-06 0.473 0.470 0.010Red -3.61E-06 0.525 0.520 0.015RedEdge 5.95E-06 0.507 0.516 0.020NIR1 -2.74E-06 0.586 0.582 0.028NIR2 1.07E-05 0.457 0.474 0.029
WorldView-2 (2014 -2019)Band Slope Intercept Average StdDevCoastal -1.04E-05 0.286748 0.27368 0.047889Blue -3.69E-06 0.290769 0.28615 0.012965Green -2.96E-06 0.366388 0.36268 0.010547Yellow -1.58E-06 0.449422 0.44745 0.009573Red 7.76E-07 0.506627 0.5076 0.009948RedEdge 1.21E-05 0.478172 0.49336 0.019277NIR1 1.40E-05 0.565498 0.58307 0.026085NIR2 7.74E-06 0.455639 0.46534 0.083162
WorldView-3 (2014 -2019)
WV02 2018v0WV03 2018v0
TOA
Ref
lect
ance
WorldView-2 & WorldView-3 radiometrically stable as shown over Libya-4
© 2019 Maxar Technologies 15
Landsat and Sentinel plot on this slide from Xin Jing, Larry Leigh, Cibele Pinto, and Dennis Helder, “Evaluation of RadCalNetOutput Data Using Landsat 7, Landsat 8, Sentinel 2A, and Sentinel 2B Sensors,” Remote Sens. 2019, 11(5), 541; https://doi.org/10.3390/rs11050541
RadCalNet data are given at Nadir. MAXAR sensors will typically have an ONA of 5 - 30 degrees. Variation in ONA will increase the variability in MAXAR data due to surface BRDF and longer atmospheric path and < 10% is considered good.
WV02 is showing a relative difference of < 6%, and WV03 and GE01 are < 5% (Right). Comparable Landsat bands are within 5% (Left).
MAXAR sensors show good agreement with Landsat and Sentinel over RadCalNet site Railroad Valley, NV
© 2019 Maxar Technologies 16
New absolute radiometric vicarious calibration employs improvements that tighten uncertainties and falls within specifications
MAXAR compares to RadCalNet RVUS at +/- 5%
MAXAR sensors are shown to be radiometrically stable throughout their lifetime
MAXAR radiometry is comparable to Landsat 8 and Sentinel at +/- 5%
We continue to investigate our atmospheric model and include more measured data to improve derived TOA radiance for truth
Future Plans:
− Update Libya-4 BRDF model and transfer 2018 calibration to CAVIS
− Validate data collected at new calibration range in Fort Lupton, CO, USA
− Step back and work on further automation
− Preparation for next generation of sensors
− Perform sensitivity analysis across system and look for more places to improve
MAXAR sensors are radiometrically accurate, reliable and stable
© 2019 Maxar Technologies 17 17
MAXAR.COM