Summary of the SGLI productsValidation results
(Ver. 2.00)
S - 1
Earth Observation Research CenterJapan Aerospace Exploration Agency
29 June 2020
History
23 Dec. 2017 GCOM-C (SHIKISAI) launched from Tanegashima Space Center1 Jan. 2018 Release of the SGLI First-light images28 Mar. 2018 Initial function verification completed20 Dec. 2018 Public release of first version Level- 1 and Level- 2 SGLI products
Aug. 2019 GCOM-C Science team decided version-up productsJun. 2020 Public release of second version Level- 1 and Level- 2 SGLI products
S - 2
Validation summary of the SGLI L1/L2 products
Level/Area[The number of products]
L1[1]
Land[9]
Atmosphere[8]
Ocean[7]
Cryosphere[4]
Total[29]
Release accuracy 1 9 8 7 4 29
Standard accuracy 1 (+1) 4 (+3) 5 (+1) 5 (+2) 2 (+1) 17 (+8)
Target accuracy* 0 0 2 1 (0) 0 3 (0)
SuccessLevel
Minimum Success [L + 1 yr]
Full Success[L + 5 yr]
Extra Success[L + 5 yr]
StandardProducts
Complete the Cal. & Val. phase and start data distribution of more than 20 products achieving the release accuracy thresholds
Achieve standard accuracy thresholds of all standard products
Achieve target accuracy thresholds of one or more standard products
GCOM-C Success criteria (data production aspect only)
*the number of products achieved standard and target accuracy thresholdThe numbers in parentheses are the differences of achieved product number from Ver.1.Confirmation of achievement of standards and target accuracy will take place five years after launch.
Results of the version-up validation
Eight products achieved newly standard accuracy through version-up validation.
S - 3
- Product Algorithm Validation Ver.2主な変更点
Atm
osp
here
Cloud flag/Classification Atmosphere PIsJAXA,
Atmosphere PIs NA
Classified cloud fraction
Atmosphere PIs Atmosphere PIs
Cloud top temp/height
NAWater cloud optical thickness/effective radius
Ice cloud optical thickness
Aerosol over the oceanJAXA,
Atmosphere PIsAtmosphere PIs
Revised inversion methodUsing data assimilation model result as initial valueLand aerosol by near ultra-violet
Aerosol by polarization Revised cloud and snow screening
- Product Algorithm Validation Ver.2 Major changes
Land
Precise geometric correction JAXA NA (bug fixed)
Atmospheric corrected reflectance (incl. cloud detection) JAXA Land PIs
Revise of BRDF estimation and QAUpdate of the calibration coefficient
Add the Solar radiation in 8 bits
Vegetation Index NA
Shadow Index Land PIBrush up the estimation coefficient and validation
method
Above-ground biomass Land PIs Land PIs Revise of LUT and coefficientAdd the past observation for QA and data complementVegetation roughness index Land PI
fAPARJAXA, Land PI Land PIs
Revise of LUT and coefficientRevise of Understory LAILeaf area Index
Land surface temperature Land PI Revise of cloud screening using CLFG
Standard products version up summary
S - 4
- Product Algorithm Validation Ver.2 Major changes
L1 Level-1 JAXA Correction for sensor sensitivity aging
Reduction of linear noise (VNR) and horizontal stripes (TIR)
- Product Algorithm Validation Ver.2 Major changes
Ocean
Normalized water leaving radiance (incl. cloud detection) Ocean PIs,
JAXA
Ocean PIs
Revised algorithm and QAsAtmospheric correction parameters
Photosynthetically available radiation JAXA, Ocean PINA
Chlorophyll-a concentration JAXA
Suspended solid concentration Ocean PI Revised coefficients and QAs
Colored dissolved organic matter Ocean PI Revised coefficients and QAs
Sea surface temperature (incl. cloud detection)
JAXA Revised cloud detection
- Product Algorithm Validation Ver.2 revision
Cry
osp
here
Snow and ice covered area (incl. cloud detection)
Cryosphere PI Cryosphere PI
Revised cloud detection algorithmOkhotsk sea-ice distribution
Snow and ice surface temperature Updated emissivity tableRevised neural-network algorithmSnow grain size of shallow layer
S - 5
Standard products version up summary
Products Ver.2 Accuracy Release Accuracy Standard Accuracy Target Accuracy
L1 Level-1VNR/SW:5%(abs.),1%(rel.)
TIR: 0.24 K (@300K)Geometric acc.< 0.43 pixel
VNR/SW:5%(abs.)Geometric acc.< 1 pixel
VNR/SW:5%(abs.),1%(rel.)TIR: 0.5 K (@300K)Geometric acc.< 0.5 pixel
VNR/SW:3%(abs.),0.5%(rel.)TIR: 0.5 K (@300K)Geometric acc.< 0.3 pixel
Lan
d
Precise geometric correction 0.15~0.43 pixel <1pixel <0.5pixel <0.25pixel
Atmospheric corrected reflectance (incl. cloud detection)
0.02 (<=443nm)0.04 (>443nm)
0.3 (<=443nm), 0.2 (>443nm)
0.1 (<=443nm), 0.05 (>443nm)
0.05 (<=443nm), 0.025 (>443nm)
Vegetation Index* NDVIEVI
Grass:12%, Forest:10%Grass:16%, Forest:20%
Grass: 25%, Forest: 20% Grass: 20%, Forest: 15% Grass: 10%, Forest: 10%
Shadow Index 26.3% 30% 20% 10%
Above-ground biomass Grass:30%, Forest:52% Grass: 50%, Forest: 100% Grass:30%, Forest:50% Grass:10%, Forest:20%
Vegetation roughness index 20.3% 40% 20% 10%
fAPAR Grass:16%, Forest:17% Grass: 50%, Forest: 50% Grass: 30%, Forest: 20% Grass: 20%, Forest: 10%
Leaf area Index Grass:27%, Forest:36% Grass: 50%, Forest: 50% Grass: 30%, Forest: 30% Grass: 20%, Forest: 20%
Land surface temperature 2.1 K 3.0 K 2.5 K 1.5 K
Atm
osp
here
Cloud flag/Classification - 10% (with whole-sky camera) Incl. below cloud amount Incl. below cloud amount
Classified cloud fraction - 20% (on solar irradiance) 15% (on solar irradiance) 10% (on solar irradiance)
Cloud top temp/height - 1K 3K/2km 1.5K/1km
Water cloud optical thickness/effective radius - 10%/30% (CloutOT/raduis) 100% (as cloud liquid water) 50%/20%
Ice cloud optical thickness - 30% 70% 20%
Aerosol over the ocean 0.09** 0.1 (monthly ta_670,865) 0.1 (Scene ta_670,865) 0.05 (Scene ta_670,865)
Land aerosol by near UV 0.13 0.15 (monthly ta_380) 0.15 (scene ta_380) 0.1 (scene ta_380)
Aerosol by polarization 0.14 0.15 (monthly ta_670,865) 0.15 (scene ta_670,865) 0.1 (scene ta_670,865)
*The vegetation index is evaluated with the accuracy of both NDVI and EVI. Therefore,
the vegetation index was not achieved standard accuracy.
**The number of validation samples for Ocean AOT was small. Therefore, Ocean AOT was
not judged to achieved standard accuracy.
Accuracy achieved on Ver.1
Version Up algorithms Accuracy achieved on Ver.2
Accuracy requirements of SGLI standard products and Current Evaluation Status
S - 6
Products Ver.2 Accuracy Release Accuracy Standard Accuracy Target Accuracy
Ocean
Normalized water leaving radiance (incl. cloud detection)
19~47% (<600nm)0.9W/m2/sr/um(>6
00nm)60% (443~565nm)
50% (<600nm) 30% (<600nm)
0.5W/m2/str/um (>600nm)
0.25W/m2/str/um (>600nm)
Atmospheric correction parameters
79% 80% (ta_865nm) 50% (ta_865nm) 30% (ta_865nm)
Photosynthetically available radiation
9.5% 20% (10km/month) 15% (10km/month)10% (10km/month)
Chlorophyll-a concentration
-58~+140% -60~+150% (offshore) -60~+150%-35~+50% (offshore) -50~+100% (coast)
Suspended solid concentration
-58~+138% -60~+150% (offshore) -60~+150% -50~+100%
Colored dissolved organic matter
-58~+139% 60% (443~565nm) -60~+150% -50~+100%
Sea surface temperature (incl. cloud detection)
day:0.4K,night:0.7K 0.8K (daytime) 0.8K (day/night) 0.6K (day/night)
Cry
osp
here
Snow and Ice covered area (incl. cloud detection)
8.5% 10% (vicarious val. with other sat. data) 7% 5%
Okhotsk sea-ice distribution 9.0% 10% (vicarious val. with other sat. data) 5% 3%
Snow and ice surface Temperature
2 K 5K (vicarious val. with other sat. data and climatology data)
2K 1K
Snow grain size of shallow layer
34% 100% (vicarious val. with climatology between temp-size)
50% 30%
*Version numbers of all products including version unchanged algorithms will be changed to “2xxx”.
The accuracy number is defined by the root mean square error (RMSE), which has the same unit as the physical quantities. The accuracy ratio
(%) is evaluated by the ratio of RMSE and the average value of the in-situ data. For flag type products, the false determination rate (%) is
statistically evaluated using in-situ data.
Version Up algorithms
Accuracy requirements of SGLI standard products and Current Evaluation Status
Accuracy achieved on Ver.1
Accuracy achieved on Ver.2
not achieved night-time
not achieved >670nm
S - 7
First Light
Val.
InitialCheck-out
GCOM-C launchReview for the
development Completion
Field campaigns for the Ver.1 data release
Ver.1 Ver.2 Ver.3
Sensor Operation
AlgorithmDevelopment & Improvement
Past data
re-processing
Proc.testGround System
Cal. for the Ver.1Cal.
In-situObservation
Cal. for the Ver.2
Initial Cal. Normal Operation Post-normal
Operation
Val. for the Ver.1
Ver.1 processing Ver.2 processing Ver.3 processing
2016 2023
10-11-12 1 - 2 - 3 4 - 5 - 6 7 - 8 - 9 10-11-12 1 - 2 - 3 4 - 5 - 6 7 - 8 - 9 10-11-12 1 - 2 - 3 4 - 5 - 6 7 - 8 - 9 10-11-12 1 - 2 - 3 4 - 5 - 6 7 - 8 - 9 10-11-12 1 - 2 - 3 4 - 5 - 6 7 - 8 - 9 10-11-12 1 - 2 - 3 4 - 5 - 6 7 - 8 - 9 10-11-12 1 - 2 - 3
H30 H31 H32 H33 H34
2018 2019 2020 2021 20222017
H28 H29
Sensor performance
Evaluation
Pre-launch algorithm development
Initial Cal./Val. Algorithm improvement and validationPost-
normal operation
Coef. fixing
Algo. Improvement for Ver.1
Review for the Ver. 1
data release and minimum success
Cal. coef. Determination
Vi- cal.obs. data
Cal for the Ver.3
Ground truth data
AccuracyEvaluation
Coef. Turing
FY2016 FY2017 FY2018 FY2019 FY2020 FY2021 FY2022
Current - 2020.06.29
Schedule for the GCOM-C mission
Review for the Ver. 2
data release
Review for the Ver. 3
data releaseFinal review for the
full/extra success
Proc.test Proc.test
Past data re-processing Past data re-processing
Algo. Improvement for Ver.2 Algo. Improvement for Ver.3
Val. for the Ver.2 Val. for the Ver.3 Val. for the Final review
Field campaigns for the Ver.2/Ver.3 data release Field campaigns for the Final Review
S - 8
0
Ver2. Land Products
Evaluation Summary
ProductRelease threshold
Standard accuracy
Target accuracy
Status*1 Evaluation Methods
Precise geometric correction
<1 pixel <0.5 pixel <0.25 pixel ◎Evaluation of geolocation accuracies with GCPs prepared using AVNIR-2 data.
Atmospheric corrected reflectance (incl. cloud detection)
0.3 (<=443nm)0.2 (>443nm)(scene)
0.1 (<=443nm)0.05 (>443nm)(scene)
0.05 (<=443nm), 0.025 (>443nm)(scene)
○⇒◎ Comparison with in-situ observed reflectance.
Vegetation index
Grass:25% (scene), forest:20% (scene)
Grass:20% (scene), forest:15% (scene)
Grass:10% (scene), forest:10% (scene)
○ Comparison with in-situ observation and other satellitedata.
Above-ground biomass
Grass:50%, forest: 100%
Grass:30%, forest:50%
Grass:10%, forest:20% ○
Comparison with in-situ observation (incl. the data from the literatures).
Vegetation roughness index
Grass & forest: 40% (scene)
Grass & forest:20% (scene)
Grass & forest:10% (scene)
○ Comparison with other satellite data.
Shadow indexGrass & forest: 30% (scene)
Grass & forest:20% (scene)
Grass & forest:10% (scene)
○ Comparison with in-situ observations.
fAPARGrass:50%, forest: 50%
Grass:30%, forest:20%
Grass:20%, forest:10% ○⇒◎ Comparison with in-situ observation and other satellitedata.
Leaf area indexGrass:50%, forest: 50%
Grass:30%, forest:30%
Grass:20%, forest:20% ○ Comparison with in-situ observation and other satellitedata.
Surface temperature
<3.0 K (scene) <2.5 K (scene) <1.5 K (scene) ○⇒◎ Comparison with in-situ observation and other satellitedata.
*1 Symbols denote as follows; 〇: the release threshold achieved, ◎: the standard accuracy achieved,☆: the target accuracy achieved.
1
2
VN11-c1 VN11-c2
Ver. 1 RSRFQ RGB
Ver. 2 RSRFQ RGB
✓ Bug fix about relative zenith angle (should be abs() in BRDF kernel c1
✓ PI02/01 and VN11P/08P samples are used for the BRDF regression of VN11P/08P
1) BRDF calculation
VN11-c1 VN11-c2
Cloud areas where BRDF is available are recovered by the BRDF (flagged by bit-14 of QA_flag)
BRDF kernel 1 BRDF kernel 2
BRDF kernel 1 BRDF kernel 2
Atmospherically corrected reflectance(RSRF)
3
Ver. 1 2018/08/01 daily data Ver. 2 2018/08/01 daily data
2) Cloud screening and aerosol correction of the slant-view telescopes
X-axis values are calculated by MODIS BRDF product with SGLI geometry
Ver.1 PI02 2019/08/04 T0911 Ver.2 PI02 2019/08/04 T0911
✓ Adding cloud tests by radiance intensity (I) of Pol data
✓ Aerosol is estimated by PI01 for PI01/02 correction (AE is from non-pol data)
✓ Improve agreement with MODIS BRDF(bellows)
Improvement of clouds screening (some are remaining)
Atmospherically corrected reflectance(RSRF)
4
Validation results Release accuracy (L+1yr) Standard accuracy (L+5yr) Target accuracy (L+5yr)
0.02 (<=443nm)0.07 (>443nm)
0.3 (<=443nm) →150%0.2 (>443nm) →100%
0.1 (<=443nm) →50%0.05 (>443nm) →25%
0.05 (<=443nm) →25%0.025 (>443nm) →13%
The accuracy targets are defined as error ratios of “solar
zenith angle<30 deg, smooth surface of reflectance ~0.2”
O2A absorption
Polarization telescopes
3) Increase of validation samples
✓ About 550 samples for Ver.2 (about120 samples for Ver.1)
✓ Takayama (PEN), Fuji-hokuroku (PEN), RadCalNet (Namibia, France, US), RRV (PI Moriyama), Mongolia (PI Nasahara), tottori(PI Susaki)
Targ
et
accura
cy
Sta
ndard
accura
cy
Rele
ase
accura
cy
Sta
ndard
accura
cy
Validation result at the Ver.1 release
Sample N~120
0.019(<=443nm)0.084 (>443nm)
Worst value (VN09)
Worst value (PI02)
Ver.1 resultsAtmospherically corrected reflectance(RSRF)
5
Validation results Release accuracy (L+1yr) Standard accuracy (L+5yr) Target accuracy (L+5yr)
0.02 (<=443nm)0.06 (>443nm)
0.3 (<=443nm) →150%0.2 (>443nm) →100%
0.1 (<=443nm) →50%0.05 (>443nm) →25%
0.05 (<=443nm) →25%0.025 (>443nm) →13%
Error ratio of reflectance~0.2
Polarization telescopes
✓ Channel number achieving target accuracy is increased by improvement of L1B calibration and QA (cloud screening)
Targ
et
accura
cy
Sta
ndard
accura
cy
Rele
ase
accura
cy
Targ
et
accura
cy
Worst value (PI02)
Ver.2 results
O2A absorption
3) Increase of validation samplesAtmospherically corrected reflectance(RSRF)
6
Validation results Release accuracy (L+1yr) Standard accuracy (L+5yr) Target accuracy (L+5yr)
0.02 (<=443nm)0.04 (>443nm)
0.3 (<=443nm) →150%0.2 (>443nm) →100%
0.1 (<=443nm) →50%0.05 (>443nm) →25%
0.05 (<=443nm) →25%0.025 (>443nm) →13%
Error ratio of reflectance~0.2
偏光鏡筒
✓ All CH achieve the standard accuracy and CH number achieving target accuracy is increased with conversion to the in-situ observation angles (nadir) by derived BRDF
✓ All CH except for CH1 can achieve target accuracy by following the validation definition, “solar zenith angle<30 deg, smooth surface of reflectance ~0.2”
0.02127%
0.01920%
0.01816%
0.02014%
0.02012%
0.02115%
0.02410%
0.02410%
0.03913%
0.03210%
0.03210%
0.0257%
0.0348%
0.0268%
0.0208%
0.0309%
Targ
et
accura
cy
Sta
ndard
accura
cy
Targ
et
accura
cy
Sta
ndard
accura
cy
Ver.2 results with BRDF correction3) Increase of validation samplesAtmospherically corrected reflectance(RSRF)
(4) Addition of shortwave radiation (SWR)
7
Monthly RMS/AVG=14%8-day RMS/AVG=18%
Yearly average
Daily RMS/AVG=27%
✓ Daily SWR is estimated by instantaneous transmittance of visible channels and stored by 8 bit (/Image_data/SWR)
✓ Comparison with in-situ measurements by BSRN
✓ 14% does not achieve the SWR target accuracy (monthly 10W/m2); the daily change should be considered in the future version
✓ Anomaly is useful for applications such as agriculture and long-term weather monitoring
Atmospherically corrected reflectance(RSRF)
Around the east part of Mongol
Key updates✓ No modification for the algorithm.
✓ Input data [RSRFQ] was changed.
✓ Validation data was increased.
T2A/VeGetation Index(VGI):NDVI
CGLS/GIOGL1_NDVI300_V1.0.1
[2019/09/01-10]
✓ NDVI became slightly higher than ver.1
by the improvement of QA.
8
0 1NDVI others
Ver.2/NDVI(MVC, input: Ver.2)[2019/8/29-9/13]
Ver.1/NDVI(MVC)[2019/8/29-9/13]
Around Japan
MOD13Q1 [2019/08/29-9/13]
Similar distribution
Ver.2
Ver.1
PROBA-V
MODIS
[for reference]
Results by the input of RSRFQ [ver.1]
Estimated errors Release threshold Standard accuracy Target accuracy
12.2→11.7%11.6→10.2 %
20% (Forest) Scene,25% (Grassland) Scene
15% (Forest) Scene,20% (Grassland) Scene
10% (Forest) Scene,10% (Grassland) Scene
Validation method- SGLI/NDVI is evaluated using reference NDVI at validation sites for forests and grasslands. The reference NDVI is calculated using spectral reflectance
data collected at validation sites.
- The SGLI data is not used when the quality flag indicates cloud shadow or probably cloud, or the pixel is adjacent to cloud pixels.
- The effect of observation tower is corrected for TKY, FHK and PFRR. It is simply corrected for TSE using the data of TKY and FHK.
- The in-situ spectral reflectance data for PFRR and MSE sites are upscaled to the SGLI scale using the Sentinel-2A/B Multispectral Imager(MSI) Level-
1C/2A data (Band4 and 8b). The reference NDVI is calculated using the upscaled data and used as the validation.
Validation data and period- RSRFQ [ver.2] is used as the input data.
- The data is used from April 1, 2018 to November 30, 2018 for TKY and FHK, from April 26, 2018 to October 31, 2018 for TOC, from May 2, 2018 to
September 2, 2018 for TSE, from August 30, 2018 to September 30, 2018 for Lambir and PFRR.
- The data at MSE site during summer season (from July 11, 2018 to August 10 and from August 1, 2019 to August 31, 2019) is used as the data for grassland.
- The data at observation date ± 10 days is used for MBA, MBU and WTR regarding the NDVI that stay consistent during this period.
T2A/ VGI:NDVI
9
✓ The number of Validation data was increased.
✓ The accuracy became higher than ver.1 inputs by the improvement of QC.
< validation sites >< Validation results >
T2A/VGI:EVI
10
0 1EVI others
Ver.2 EVI(ndvi_MVC, input: Ver.2)[2018/8/29-9/13]
Ver.1 EVI(ndvi_MVC)[2018/8/29-9/13]
Around the east part of Mongol
Around Japan
MOD13Q1 [2019/08/29-9/13]
Key updates✓ No modification for the algorithm.
✓ Input data [RSRFQ] was changed.
✓ Validation data was increased.
✓ EVI became slightly higher than ver.1 by
the improvement of QA.
Estimated errors Release threshold Standard accuracy Target accuracy
21.8→20.3%21.5→16.4 %
20% (Forest) Scene,25% (Grassland) Scene
15% (Forest) Scene,20% (Grassland) Scene
10% (Forest) Scene,10% (Grassland) Scene
T2A/ VGI:EVI
11
< All data >< Data at cloud-free days >
※All data was used for Lambir/WTR/MBU/MBA.
✓ The number of Validation data was increased. ✓ The accuracy became higher than ver.1 inputs by the improvement of QC.
< validation sites >
[for reference]
Results by the input of ver.1 RSRFQ.All dataData at cloud-free days
< Validation results >
Validation method- SGLI/EVI is evaluated using reference EVI at validation sites for forests and grasslands. The reference EVI is calculated using spectral
reflectance data collected at validation sites.
- Only the data collected at cloud-free days are used for the validation. Cloud-free days are identified using the images taken by digital
camera and incoming sky radiation taken by spectroradiometers.
- The SGLI data is not used when the quality flag indicates cloud shadow or probably cloud, or the pixel is adjacent to cloud pixels.
- The effect of observation tower is corrected for TKY, FHK and PFRR. It is simply corrected for TSE using the data of TKY and FHK.
- The in-situ spectral reflectance data for PFRR and MSE sites are upscaled to the SGLI scale using the Sentinel-2A/B Multispectral
Imager(MSI) Level-1C/2A data (Band4 and 8b). The reference EVI is calculated using the upscaled data and used as the validation.
Validation data and period- RSRFQ [ver.2] is used as the input data.
- The data is used from April 1, 2018 to November 30, 2018 for TKY and FHK, from April 26, 2018 to October 31, 2018 for TOC, from
May 2, 2018 to September 2, 2018 for TSE, from August 30, 2018 to September 30, 2018 for Lambir and PFRR.
- The data at MSE site during summer season (from July 11, 2018 to August 10 and from August 1, 2019 to August 31, 2019) is used as the
data for grassland.
- The data at observation date ± 10 days is used for MBA, MBU and WTR regarding the EVI that stay consistent during this period.
12
The list of validation sitesForests• AK06: AK06 Alaska Forest site, USA• FHK: Fuji Hokuroku Flux Observation site, Japan• Lambir: Lambir Hills site, Malaysia• Lambir Oil Palm: Lambir Oil Palm Plantation site, Malaysia• PFRR: Poker Flat Research Range site, USA• SSP: Spasskaya Pad site, Russia• TKY: Takayama Deciduous Broadleaf Forest site, Japan• TMM: Tokachi Mitsumata JAXA Super Site 500, Japan• TOC: Tomakomai Crane Site, Japan • TOS: Hokkaido University Tomakomai Experimental Forest site, Japan• TSE: Teshio CC-LaG Experiment site, Japan• URY: Hokkaido University Uryu Experimental Forest site, Japan• Yamashiro: Yamashiro site, Japan
Grasslands and paddy fields• DGT: Delgertsogt JAXA Super Site 500, Mongolia • KYM: Khar Yamaat JAXA Super Site 500, Mongolia• MBN: Baganuul JAXA Super Site 500, Mongolia• MBU: Bayan-Unjuul JAXA Super Site 500, Mongolia• MSE: Mase Paddy Flux site, japan• WTR: Watarase JAXA Super Site 500, Japan
13
Acknowledgements✓ The field data were mainly collected by GCOM-C PIs, AKITSU Tomoko, HONDA Yoshiaki, KAJIWARA Koji, KOBAYASHI Hideki,
MORIYAMA Masao, NAGAI Shin, NASAHARA Kenlo Nishida, and SUSAKI Junichi. ✓ The CGLS/GIOGL1_LAI300, FAPAR300 and NDVI300 products were used for the comparison with SGLI products in this
article. They were generated by the global component of the Land Service of Copernicus, the Earth Observation program of the European Commission. The research leading to the current version of the product has received funding from various European Commission Research and Technical Development programs. The products are based on PROBA-V 333m data ((c) Belgian Science Policy and distributed by VITO NV).
✓ Sentinel-2A/B Multispectral Imager(MSI) Level-1C/2A data were produced by European Space Agency (ESA). They were acquired from the Copernicus Open Access Hub of the ESA.
✓ MOD13Q1 MODIS/Terra Vegetation Indices 16-Day L3 Global 250m SIN Grid V006 were acquired from NASA EOSDIS Land Processes DAAC from https://lpdaac.usgs.gov, maintained by the NASA EOSDIS Land Processes Distributed Active Archive Center (LP DAAC) at the USGS Earth Resources Observation and Science (EROS) Center, Sioux Falls, South Dakota.
References➢ Akitsu, T. K., Nakaji, T., Yoshida, T., Sakai, R., Mamiya, W., Terigele, Takagi, K., Honda, Y., Kajiwara K., Nasahara, K. N., 2020.
Field data for satellite validation and forest structure modeling in a pure and sparse forest of Picea glehnii in northern Hokkaido. Ecological Research, 2020, pp.1–15. doi:10.1111/1440-1703.12114.
➢ Akitsu, T. K., Nakaji, T., Kobayashi, H., Okano, T., Honda, Y., Bayarsaikhan, U., Terigele, Hayashi, M., Hiura, T., Ide, R., Igarashi, S., Kajiwara, K., Kumikawa, S., Matsuoka, Y., Nakano, Ta., Nakano, To., Okuda, A., Sato, T., Tachiiri, K., Takahashi, Y., Uchida, J., Nasahara, K. N., (in press). Large-scale ecological field data for satellite validation in deciduous forests and grasslands. Ecological Research.
➢ Kobayashi, H., Nagai, S., Kim, Y., Yang, W., Ikeda, K., Ikawa, H., Nagano, H. and Suzuki, R., 2018. In situ observations reveal how spectral reflectance responds to growing season phenology of an open evergreen forest in Alaska. Remote Sensing, 10(7), p.1071. (Data are available from Arctic Data archive System (ADS), Japan, https://ads.nipr.ac.jp/dataset/A20181212-002)
Key updates✓ Minor revision of coefficients for estimating SDI.✓ Modification of validation method.
Validation method- SGLI/SDI is evaluated by the shadow areas calculated using the in-situ observation data. - The in-situ shadow areas are calculated from the DSM data of USGS 3 Dimensional Elevation Programme (3DEP) and Solar Zenith Angle
(SZA). SGLI/SDI is compared with the in-situ shadow areas correspond to each SGLI pixel (250[m]). The validation was conducted at Michigan, USA.
- The pixels with satellite zenith angle < 8 deg. and NDVI < 0.65 are validated. The pixels adjacent to cloud or probably cloud are excluded from validation.
Validation period- November 23, 2018.
T2A/ShaDow Index(SDI)
< Validation result >※ Validation result for ver.1 alg.
Estimated errors Release threshold Standard accuracy Target accuracy
30.0→ 24.0% 30% (Forest, Grassland) Scene 20% (Forest, Grassland) Scene 10% (Forest, Grassland) Scene
DSM Shadow areas
Shadow areas correspond
to each SGLI pixel
(truth)
SGLI_ver.2 SDI
comparison
14
※The accuracy was improved (the possibility increased to achieve standard accuracy).
Water areasTarget areas in Michigan
(DSM on GMTED)
#: 3242, Bias: -0.0172, RMSE: 0.102,
AVG. SDI: 0.426, Relative error: 24.0 %
Comparison between ver.1 and ver.2
Key updates✓ ver.1: LAI [Forest: Overstory_LAI, Non-forest: Total_LAI], Understory_NDVI [Forest: Understory_NDVI, Non-forest: 0]
⇒ ver.2: LAI [Total_LAI] 、Overstory_LAI [Overstory_LAI]
← The definition become independent to land cover on the BaseMap. LAI is estimated properly when BaseMap misclassified the land cover.
✓ Modification of LUTs. Use of the reflectance data for VN08, VN11, PI01 and PI02.
← The accuracy is improved and the subdivision of LUTs become possible for later versions.
✓ Use of multiple LUTs for each land cover. LAI estimation at sparse vegetation areas.
← LAI is retrieved properly when the Base Map misclassified the land cover.
✓ Increase of the number of validation data.
T2B/Leaf Area Index (LAI)
Ver.2 T2B/LAI (total LAI) [2019/09/01-10] (input: Ver.1) Ver.2 T2B/Overstory LAI [2019/09/01-10] (input: Ver.1)
Ver. 1 T2B/LAI (F: overstory, NF: total) [2019/09/1-10]CGLS/GIOGL1_LAI300_V1.0.1 (total LAI) [2019/09/01-10]
others0 8LAI [m2/m2]
Total_LAI is equivalent to other products.
LAI is estimated for sparse vegetation.
The overestimation of LAI at boreal forest was improved.
The accuracy of overstory LAI was improved
15
Validation method- SGLI/LAI is evaluated using reference LAI data at validation sites for forests and grasslands. The reference
LAI is calculated using in-situ observation data.
- The daily SGLI/LAI is averaged for 11 days and compared with reference LAI.
- The SGLI data is not used when the quality flag indicates cloud shadow, probably cloud, pol. cloud or
backup algorithm.
- RSRFQ [ver.2] is used as the input data for the sites in Japan and Mongolia.
- RSRFQ [ver.1] is used as the input data for the other sites.
Validation period- Lambir, Lambir Oil Palm, TMM: From August 1, 2019 to August 10, 2019
- Other sites: Observation date ± 5 days.
T2B/ LAI
Total LAI < Forest > Overstory LAI < Forest >
Estimated errors Release threshold Standard accuracy Target accuracy
Forest (Total LAI): 79.1→35.9%
Forest (Overstory LAI): 112.6%→44.6%
Grassland: 34.7→27.0%
50% (Forest),50% (Grassland)
30% (Forest),30% (Grassland)
20% (Forest),20% (Grassland)
16※ The accuracy was improved especially for Overstory LAI.
Total LAI < Forest > Total LAI < Grassland >Overstory LAI < Forest >
Total LAI < Grassland >
The accuracy is low in
sparse grasslands.
Cropland was assigned
for LC in BaseMap
< validation sites >
[for reference]
Validation results for ver.1 alg.
< Validation results >
Key updates✓ Modification of LUTs. Use of the reflectance data for VN08, VN11, PI01 and PI02.
← The accuracy is improved and the subdivision of LUTs become possible for later versions.
✓ Use of multiple LUTs for each land cover. FAPAR estimation at sparse vegetation areas.
← FAPAR is retrieved properly when the Base Map misclassified the land cover.
✓ Increase of the number of validation data.。
T2B/Fraction of Absorbed PhotosyntheticallyActive Radiation(FAPAR)
Ver.2 T2B/FAPAR [2019/09/01-10] (input: Ver.1)
Ver. 1 T2B/FAPAR [2019/09/01-10]
others0 1FAPAR
CGLS/GIOGL1_FAPAR300_V1.0.1 [2019/09/01-10]
SENTINEL-3A daily FAPAR
[2019/10/10, resol. 300m]Ver.2 T2B/FAPAR daily (input: Ver.2)
[2019/10/10]
➢ The spatial distribution of retrieved FAPAR is equivalent
to other products.
17
The overestimation of FAPAR at boreal
forest was improved.
FAPAR is retrieved at
sparse vegetation
Forest*1(Green FAPAR) Grassland
Estimated errors Release threshold Standard accuracy Target accuracy
Forest: 33.3→16.9%*1
Grassland: 6.3→7.1%
50% (Forest),50% (Grassland)
20% (Forest),30% (Grassland)
10% (Forest),20% (Grassland)
*1 Validation for Green FAPAR, though the total FAPAR was used as in-situ data.
※FAPAR by canopy was used as the definition in PFRR.
Forest*2(Total FAPAR)
18※ Standard accuracy is achieved.
Grassland
*2 Validation for Total FAPAR.
Forest*1 (Green FAPAR)
Over-estimation at
sparse forest.
< validation sites >
T2B/ FAPAR
[for reference]
Validation results for ver.1 alg.
< Validation results >
Validation method- SGLI/FAPAR (Green FAPAR) is evaluated using reference FAPAR (Total FAPAR) data at validation sites
for forests and grasslands. The reference FAPAR (Total FAPAR) is calculated from in-situ observation data.
- Total FAPAR in forests is re-estimated from SGLI data using the same algorithm with T2B, and also
compared with reference FAPAR (Total FAPAR) for the direct evaluation with the same definition.
- The daily SGLI/FAPAR is averaged for 11 days and compared with reference FAPAR.
- The SGLI data is not used when the quality flag indicates cloud shadow, probably cloud, pol. cloud or
backup algorithm.
Validation data and period- RSRFQ [ver.2] is used as input data for URY, TOS, WTR and FHK.
- RSRFQ [ver.1] is used as input data for PFRR.
- Validation period is observation date ± 5 days.
19
The list of validation sitesForests• AK06: AK06 Alaska Forest site, USA• FHK: Fuji Hokuroku Flux Observation site, Japan• Lambir: Lambir Hills site, Malaysia• Lambir Oil Palm: Lambir Oil Palm Plantation site, Malaysia• PFRR: Poker Flat Research Range site, USA• SSP: Spasskaya Pad site, Russia• TKY: Takayama Deciduous Broadleaf Forest site, Japan• TMM: Tokachi Mitsumata JAXA Super Site 500, Japan• TOC: Tomakomai Crane Site, Japan • TOS: Hokkaido University Tomakomai Experimental Forest site, Japan• TSE: Teshio CC-LaG Experiment site, Japan• URY: Hokkaido University Uryu Experimental Forest site, Japan• Yamashiro: Yamashiro site, Japan
Grasslands and paddy fields• DGT: Delgertsogt JAXA Super Site 500, Mongolia • KYM: Khar Yamaat JAXA Super Site 500, Mongolia• MBN: Baganuul JAXA Super Site 500, Mongolia• MBU: Bayan-Unjuul JAXA Super Site 500, Mongolia• MSE: Mase Paddy Flux site, japan• WTR: Watarase JAXA Super Site 500, Japan
20
Acknowledgements✓ The field data were mainly collected by GCOM-C PIs, AKITSU Tomoko, HONDA Yoshiaki, KAJIWARA Koji, KOBAYASHI Hideki,
MORIYAMA Masao, NAGAI Shin, NASAHARA Kenlo Nishida, and SUSAKI Junichi. ✓ The CGLS/GIOGL1_LAI300, FAPAR300 and NDVI300 products were used for the comparison with SGLI products in this
article. They were generated by the global component of the Land Service of Copernicus, the Earth Observation program of the European Commission. The research leading to the current version of the product has received funding from various European Commission Research and Technical Development programs. The products are based on PROBA-V 333m data ((c) Belgian Science Policy and distributed by VITO NV).
✓ Sentinel-3A daily FAPAR (OLCI Global Vegetal Index (faPAR)) were used for the comparison with SGLI/FAPAR. They were produced by European Space Agency (ESA) and acquired from the Copernicus Open Access Hub of the ESA.
References➢ Akitsu, T. K., Nakaji, T., Yoshida, T., Sakai, R., Mamiya, W., Terigele, Takagi, K., Honda, Y., Kajiwara K., Nasahara, K. N., 2020.
Field data for satellite validation and forest structure modeling in a pure and sparse forest of Picea glehnii in northern Hokkaido. Ecological Research, 2020, pp.1–15. doi:10.1111/1440-1703.12114.
➢ Akitsu, T. K., Nakaji, T., Kobayashi, H., Okano, T., Honda, Y., Bayarsaikhan, U., Terigele, Hayashi, M., Hiura, T., Ide, R., Igarashi, S., Kajiwara, K., Kumikawa, S., Matsuoka, Y., Nakano, Ta., Nakano, To., Okuda, A., Sato, T., Tachiiri, K., Takahashi, Y., Uchida, J., Nasahara, K. N., (in press). Large-scale ecological field data for satellite validation in deciduous forests and grasslands. Ecological Research.
➢ Kobayashi, H., Nagai, S., Kim, Y., Yang, W., Ikeda, K., Ikawa, H., Nagano, H. and Suzuki, R., 2018. In situ observations reveal how spectral reflectance responds to growing season phenology of an open evergreen forest in Alaska. Remote Sensing, 10(7), p.1071. (Data are available from Arctic Data archive System (ADS), Japan, https://ads.nipr.ac.jp/dataset/A20181212-002)
➢ Nagai, S., Kotani, A., Sato, T., Sugimoto, A., Maximov, T. C., Nogovitcyn, A., Miyamoto, Y., Kobayashi, H., and Tei, S. (in press). Direct measurement of leaf area index in a deciduous needle-leaf forest, eastern Siberia, Polar Science, (in press).
0 400AGB [t/ha] others
Key updates✓ Modification of coefficients for retrieving AGB.
← The accuracy was improved by retrieving AGB using coefficient maps.
✓ Use of RSRFQ and AGB for past 2 months, and NDVI for one year.
← QA flag was set for the possibility of land cover change.
✓ Revision of QA information.
T3A/Above Ground Biomass(AGB)
[2019/09/01-30]Ver.2 T3A/AGB (input: ver.1) Ver.1 T3A/AGB [2019/09/01-30]
21
Ver.2 AGB (input: Ver.1) Ver.1 AGB
< Detail map for the part of USA >
The overestimation of biomass in cropland
was improved.
Improved point
Artificial changes in biomass are seen at the specific latitude. These are caused by the
inadequate coefficient map, and will be improved at ver.3 algorithm.
The accuracy was improved
for boreal forest.
Improved point
※ The data with the QA information of low quality, probably cloud or bad geometry was not used in these maps.
Validation method- SGLI/AGB is evaluated using reference AGB data at validation sites for forests and grasslands. The reference AGB data is calculated using in-situ observation data.
- Forests: The daily SGLI/AGB is averaged for one month and compared with the reference AGB regarding the forest biomass as that stay consistent for one month.
- Grasslands: The daily SGLI/AGB is averaged for the date ± 10 days when the in-situ observation data was collected. The averaged SGLI data is compared with
reference AGB.
- The SGLI data is not used when the quality flag indicates low quality, probably cloud or bad geometry.
Validation data and period- Forests
✓ The data from September 1, 2019 to September 30, 2019 are used.
✓ RSRFQ [ver.2] is used as the input data for TOS, FHK and URY sites.
✓ RSRFQ [ver.1] is used as the input data for other sites.
- Grasslands
✓ The data for the observation date ± 10 days are used.
✓ RSRFQ [ver.2] is used as input data.
T3A/AGB
※The accuracy was improved (the possibility increased to achieve standard accuracy).
Estimated errors Release threshold Standard accuracy Target accuracy
Forest: 94.0→80.9%Grassland: 406.7→30.8%
100% (Forest),50% (Grassland)
50% (Forest),30% (Grassland)
20% (Forest),10% (Grassland)
※Grassland only
The accuracy was improved
for WTR site (grassland).
22
< validation sites >
< Validation result >
[for reference]
Validation result for ver.1 alg.
T3A/Vegetation Roughness Index(VRI)
Ver.2 T3A/VRI Ver.1 T3A/VRI
0 1VRI
Validation method- SGLI/ VRI is evaluated using reference VRI which is calculated from in-situ observation data (spectral directional reflectance factor). In-situ observation data
is collected for the same sun-target-satellite geometry with SGLI data.
- The SGLI data is not used when the quality flag indicates cloud, probably cloud or pol. Cloud.
・Validation data and period– SGLI/VRI: May 1, 2018 to June 10, 2018 (daily data). Only the data with ±10°of the view zenith angle for the in-situ data is used.
– Reference VRI: May 21, 2018 at WTR site.
: Standard accuracy
Estimated errors Release threshold Standard accuracy Target accuracy
34.8→20.3% 40% (Forest, Grassland) Scene 20% (Forest, Grassland) Scene 10% Forest, Grassland) Scene23
※The accuracy was improved.
※The number of validation data is not enough.
Key updates✓ No modification for the equation for retrieving VRI.
✓ Modification by the revision of AGB algorithm and input RSRFQ.
⇒ The accuracy was improved and the cloud was strictly identified.
2019/09/06 (input: ver.2) 2019/09/06Average for 2019/09/01-30 (input: ver.1) Average for 2019/09/01-30
[for reference]
Validation result for ver.1 alg.
< Validation result >
24
The list of validation sitesForests• AK06: AK06 Alaska Forest site, USA• FHK: Fuji Hokuroku Flux Observation site, Japan• Lambir: Lambir Hills site, Malaysia• Lambir Oil Palm: Lambir Oil Palm Plantation site, Malaysia• PFRR: Poker Flat Research Range site, USA• SSP: Spasskaya Pad site, Russia• TKY: Takayama Deciduous Broadleaf Forest site, Japan• TMM: Tokachi Mitsumata JAXA Super Site 500, Japan• TOC: Tomakomai Crane Site, Japan • TOS: Hokkaido University Tomakomai Experimental Forest site, Japan• TSE: Teshio CC-LaG Experiment site, Japan• URY: Hokkaido University Uryu Experimental Forest site, Japan• Yamashiro: Yamashiro site, Japan
Grasslands and paddy fields• DGT: Delgertsogt JAXA Super Site 500, Mongolia • KYM: Khar Yamaat JAXA Super Site 500, Mongolia• MBN: Baganuul JAXA Super Site 500, Mongolia• MBU: Bayan-Unjuul JAXA Super Site 500, Mongolia• MSE: Mase Paddy Flux site, japan• WTR: Watarase JAXA Super Site 500, Japan
25
References➢ Akitsu, T. K., Nakaji, T., Yoshida, T., Sakai, R., Mamiya, W., Terigele, Takagi, K., Honda, Y., Kajiwara K., Nasahara, K. N., 2020.
Field data for satellite validation and forest structure modeling in a pure and sparse forest of Picea glehnii in northern Hokkaido. Ecological Research, 2020, pp.1–15. doi:10.1111/1440-1703.12114.
➢ Akitsu, T. K., Nakaji, T., Kobayashi, H., Okano, T., Honda, Y., Bayarsaikhan, U., Terigele, Hayashi, M., Hiura, T., Ide, R., Igarashi, S., Kajiwara, K., Kumikawa, S., Matsuoka, Y., Nakano, Ta., Nakano, To., Okuda, A., Sato, T., Tachiiri, K., Takahashi, Y., Uchida, J., Nasahara, K. N., (in press). Large-scale ecological field data for satellite validation in deciduous forests and grasslands. Ecological Research.
➢ Suzuki, R., Kim, Y., & Ishii, R., 2013. Sensitivity of the backscatter intensity of ALOS/PALSAR to the above-ground biomass and other biophysical parameters of boreal forest in Alaska. Polar Science, 7(2), pp.100-112.
Validation results of Land Surface Temperature (1/2)
• Algorithm Overview• Algorithm PI: Dr. Moriyama (Nagasaki University) • Features :
• Simultaneous estimation of the LST and emissivity by inputting LTOA data, which considers the temporal fluctuation of emissivity.
• Development and implementation of radiation transmission code for high speed processing.
• Issues in V1a. Low accuracy of the cloud screening (Not use the cloud flag product)b. Discontinuity of meteorological data profile (Solid line on the right fig.)c. Sparse validation points (Only 3 points at V1 release)
• V2 update① Cloud screening using cloud flag product(← a)② Revision of the meteorological data loading function(← b)③ Adding Flux site data as in-situ data(←c)
• Validation data• In-situ:Fuji-hokuroku, Mase, AmeriFLUX• Match-up: time±15min.• QC: Convergence residual<1[K] • Period: Jan. 23 2018 to Oct. 14 2019
Meteorological profile discontinuityQuoted from PI-WS 2020
by Dr. Moriyama
Validation results of Land Surface Temperature (1/2)
V1 Day V2 DayV1 Night V2 Night
Improved accuracy due to cloud screening update.
V1 Day+Night V2 Day+Night
Results Release threshold Standard accuracy Target accuracy
RMSE=2.15 [K](scene) <3.0[K](scene) <2.5[K](scene) <1.5[K](scene)
Matchup score at night decreased.
Conclusion• V2 is more accurate than V1 and achieves standard accuracy.• In the future, plan to consider adding high-latitude in-situ sites.• Adjusting the threshold is needed after improving the accuracy of the nighttime cloud flag.
1
Ver2. Atmosphere Products
Evaluation Summary
*1 Symbols denote as follows; 〇: the release threshold achieved, ◎: the standard accuracy achieved,☆: the target accuracy achieved.
2
ProductRelease threshold
Standard accuracy
Target accuracy
Status*1 Evaluation Methods
Cloud flag/Classification
10% (withwhole-skycamera)
Incl. below cloud amount
Incl. below cloud amount ☆
Comparison with in-situ observation (sky-camera images) for release threshold. Evaluations for standard and target accuracies were performed as the Classified cloud fraction products.
Classified cloud fraction
20% (on solar irradiance)
15% (on solar irradiance)
10% (on solar irradiance) ☆
Comparison of SGLI-derived solar irradiance using cloud products including cloud flag, cloud fraction etc. with ground-measured solar irradiance.
Cloud top temp/height
1 K3 K/2 km (top temp/height)
1.5 K/1 km (temp/height) ◎
Evaluation was made as vi-cal. of SGLI TIR bands for the release threshold. In addition, comparison with other satellite data for evaluating the achievement of the standard accuracy.
Water cloudOT/effectiveradius
10%/30%(CloudOT/radius)
100% (as cloud liquid water)
50% / 20% ○ Comparison with other satellite (MODIS) data.
Ice cloud optical thickness
30% 70% 20% ○ Comparison with other satellite (MODIS) data.
Aerosol overthe ocean
0.1 (Monthlya_670,865)
0.1 (scenea_670,865)
0.05 (scenea_670,865) ○ Comparison with other satellite (MODIS) data.
Land aerosol by near ultra violet
0.15 (Monthlya_380)
0.15 (scenea_380)
0.1 (scenea_380 ) ○ → ◎ Comparison with in-situ observation and other satellite
(MODIS) data.
Aerosol by Polarization
0.15
(Monthlya_670, 865)
0.15 (scenea_670,865)
0.1 (scenea_670,865) ◎ Comparison with in-situ observation and other satellite
(MODIS) data.
3
■Changes from Ver. 1
・Aerosol retrieval by inputting model forecast of AHI as an initial value
・Simultaneous estimation of 3 variables with canonical correlation analysis
(Optical thickness, Angstrom exponent, Single scattering albedo)
■Using channels
・Land :VN01, VN02, VN03, VN05, VN08
・Ocean:VN10, SW01, SW03, SW04
■Ver. 2 algorithm
Atmosphere Product
- Aerosol over Land and Ocean with Near UV
Priori (𝒙𝒑)
AHIObserved brightness
AnalyzedValue( 𝒙𝒂)
Forecast
( 𝒙𝒇)Retrieval
Aerosolestimation
( 𝒙𝒐)
Adaptation
Forecast
Restec / Yoshida
SGLIObserved brightness
Retrieval
Aerosolestimation
( 𝒙𝒐’)
ARNP
Validation methods
4
■Validation Method:
・Aerosol over ocean:
Standard accuracy: RMS errors are evaluated comparing SGLI derived AOT with
those from other satellite sensors and shipborne in-situ observations
(AERONET/Maritime Aerosol Network) (scene by scene).
・Aerosol over land:
Standard accuracy: RMS errors are evaluated comparing SGLI derived AOT with
those from other satellite sensors and sky-radiometer at ground observation network
(Skynet and AERONET) (scene by scene).
■Validation data and condition etc.:
The SGLI data within 10 km from the in-situ sites were used and averaged for
comparison with the in-situ data.
■Validation period:
(1) Land: Sky-radiometer
Source – Irie PI@Chiba univ. (SKYNET), Aoki PI@Toyama univ., JMA,
AERONET
Period – March 2018, June 2018, Feb. 2018, Jan. 2020, Feb. 2020
(2) Ocean: Microtops on ships
Source –AERONET/Maritime Aerosol Network
Period – March 2018, June 2018, Feb. 2019
ARNP
5
Validation Results
Aerosol over land with near UV (ARNP_land)
AOT over land(Ver. 2)Averaged SGLI within 10km vs.Sky-radiometer at the same time
AOT over land(Ver. 1)Averaged SGLI within 10km vs.Sky-radiometer at the same time
Standard accuracy is archived.
R
=0.82
Bias
=0.08
RMSE
=0.16
Data num=37
ProductEstimated RMSE Standard accuracy
(L+5)Target accuracy
(L+5)Ver. 1 Ver. 2
AOT over land 0.16 0.13 (Scene) 0.15 (Scene) 0.10 (Scene)
ARNPARNP
6
Validation Results
Aerosol over ocean with near UV (ARNP_ocean)AOT over ocean(Ver. 2)Averaged SGLI within 10km vs.shipborne microtops at the same time
AOT over ocean(Ver. 1)Averaged SGLI within 10km vs.shipborne microtops at the same time
R
=0.91
Bias
=0.05
RMSE
=0.12
Data num=8
R
=0.94
Bias
=0.04
RMSE
=0.09
Data num=8
It has potential to archive standard accuracy.
ProductEstimated RMSE Standard accuracy
(L+5)Target accuracy
(L+5)Ver. 1 Ver. 2
AOT over ocean 0.12 0.09 (Scene) 0.10 (Scene) 0.05 (Scene)
AOT(670nm) AOT(670nm)
ARNPARNP
7
Validation Results
Aerosol over ocean with near UV (Monthly ARNP_ocean)
Aerosol over land with near UV (Monthly ARNP_land)
Ver2Ver1
Ver2Ver1
The cloud contamination reduced over both land and ocean.
ARNPARNP
Validation results of Aerosol by Polarization (1/2)
• Algorithm Overview• Algorithm PI: Dr. Sonoyo Mukai (The Kyoto College of Graduate Studies for Informatics)
Dr. Itaru Sano (Kindai Univ.)• Features:
• Retrieval methods of aerosol properties based on POLDER research. • Improvements of the precision of aerosol properties by using polarization and near
ultraviolet wavelength.
• Issues in V1a. Large deviation of validation resultsb. False detection of AOT in clear-sky areac. Low AOT in dirty sky area
• V2 update① Cloud screening using polarization bands(←a)② Masking snow cover area (see cloud-flag product)( ← b)*Issue c will continue to be considered for the next version
• Validation data• In-situ: SKYNET, AERONET• Match-up:
• distance±10km(ave.) / time±30min.• Period: Apr.1st to Dec. 26 2018
GC1SG1_20190501D01M_D0000_3MSG_AOTLF_1000
Conclusion• Decreased in deviation and false detection of AOT in snow area at V2.• Plan to consider the following for V3:
• Re-examination of cloud screening threshold in high AOT region.• Detailed investigation of high AOT event cases. (air pollution, forest fires, etc.)• Re-examination of radiative transfer. (including LUT)
Validation results of Aerosol by Polarization (2/2)
AOT@670nm(V2)
Results Release threshold Standard accuracy Target accuracy
RMSE=0.137(scene)
0.15 (monthly average) 0.15 (scene) 0.10 (scene)
AOT@670nm(V1)
However, match-up points are not
enough in higher AOT area (due to
cloud screening, etc.)
Decreased in the deviation and false detection of AOT in snow area.
2019年5月1日daily全球
Decreased false detection of AOT
False detection of AOT in clear-sky area
1
Ver2. Ocean Products
2
Evaluation Summary
*1 Symbols denote as follows; 〇: the release threshold achieved, ◎: the standard accuracy achieved,☆: the target accuracy achieved.
ProductRelease threshold
Standard accuracy
Target accuracy Status*1 Evaluation Methods
Normalized water leaving radiance (incl. cloud detection)
60% (443~565nm)50% (<600nm)0.5W/m2/str/um (>600nm)
30% (<600nm)0.25W/m2/str/um (>600nm) ◎ Comparison with in-situ observation data.
Atmospheric correction parameters
80% (AOT@865nm)
50% (AOT@865nm)
30% (AOT@865nm) ○ Comparison with in-situ observation data.
Photosynthetically available radiation
20% (10km/month) 15% (10km/month) 10% (10km/month) ◎ Comparison with in-situ observation data.
Chlorophyll-a concentration
−60~+150%(offshore)
−60~+150%
−35~+50%(offshore),−50~+100% (coast)
○⇒◎ Comparison with in-situ observation data.
Total suspended matter concentration
−60~+150%(offshore)
−60~+150% −50~+100% ○⇒◎Comparison with other satellite data (GOCI).
Colored dissolved organic matter
−60~+150%(offshore)
−60~+150% −50~+100% ○⇒◎Comparison with in-situ observation and other satellite data (MODIS).
Sea surface temperature
0.8 K (daytime)0.8 K (day & night time)
0.6 K (day & night time)
☆ Comparison with in-situ observation data.
Major Updates of Algorithm:• Improved aerosol estimation method, sunglint correction method, in-water model
• Applied the vicarious calibration of version 2.0
Validation Method:Validated the accuracies of predicted NWLR data from the SGLI algorithm comparing with in-
situ data: ship observation, buoy (MOBY and BOUSSOLE) and AERONET-OC
Quality Control:• In-situ data: time difference in ±3 hours to SGLI observation
• SGLI data: average of the data passed the following conditions within a 5 by 5 pixel
centered the in-situ point (for details refer to Bailey et al., 2006) :
1. 13 or more pixels which satisfies the following conditions: aerosol optical thickness <
0.5, solar zenith angle < 70 degrees, NWLR of all channels > 0, CLDAFFCTD flag
isn’t set.
2. Median CV (coefficient of variation) computed from NWLR_380-565nm and
Taua_865 nm less than 0.15
Period of Validation• January 1, 2018-February 29, 2020
Reference: Bailey, S.W., and Werdell, P.J. (2006). A multi-sensor approach for the on-
orbit validation of ocean color satellite data products. Rem. Sens. Environ. 102, 12-23.
Validation Results of Ocean NWLR Products:
Normalized Water-Leaving Radiance - NWLR
Validation Results of Ocean Products:
Normalized Water-Leaving Radiance - NWLR
Achieved the standard accuracies (<600 nm)
・Ver. 2: Validation Results
Accuracy:
28%
N=113
Accuracy:
47%
N=721
Accuracy:
32%
N=722
Accuracy:
19%
N=722
Accuracy:
22%
N=722
In-situ nLw [W/m2/sr/μm] In-situ nLw [W/m2/sr/μm] In-situ nLw [W/m2/sr/μm] In-situ nLw [W/m2/sr/μm]
SG
LI
nL
w[W
/m2
/sr/
μm
]
SG
LI
nL
w[W
/m2
/sr/
μm
]
SG
LI
nL
w[W
/m2
/sr/
μm
]
SG
LI
nL
w[W
/m2
/sr/
μm
]
SG
LI
nL
w[W
/m2
/sr/
μm
]
In-situ nLw [W/m2/sr/μm]
n=722
Accuracy:
0.9 W
/m2/sr/um
Accuracy:
22%
In-situ nLw [W/m2/sr/μm] In-situ nLw [W/m2/sr/μm]
SG
LI
nL
w[W
/m2
/sr/
μm
]
SG
LI
nL
w[W
/m2
/sr/
μm
]
NWLR_380 NWLR_412 NWLR_443 NWLR_490 NWLR_530
NWLR_565 NWLR_670
N=722 N=659
Achv. Target Achv. Target Achv. Target
Achv. Target
Achv. Standard
Not Achv. Stnd. • NWLR(380nm, 490-565nm) achieved the target accuracies,
NWLR(412-443nm) achieved the standard accuracies 。• NWLR(670 nm) didn't achieved the standard accuracy on coastal
regions. It is a issue to be solved by version 3. However, version 2
accuracy improved than version 1 (1.2→0.9W/m2/sr/um).
• Number of In-situ points increased than the validation result at
version 1 release
(ver.1 release: 14-102 points → ver.2 release: 113-722 points).
Validation Result Release Accuracy Standard Accuracy Target Accuracy
[Ver. 1]30-40 → [Ver. 2] 19-47% (<600nm) 60% (443-565 nm) 50% (<600 nm), 30% (<600 nm)
[Ver. 1] 1.2 → [Ver. 2] 0.9W/m2/sr/um (>600nm) N/A 0.5W/m2/sr/um (>600nm) 0.25W/m2/sr/um (>600nm)
Achv. Standard
Validation Results of Ocean NWLR Products :
Atmospheric Correction Parameters - ACP
Major Updates of Algorithm:• Improved aerosol estimation method, sunglint correction method, in-water model
• Applied the vicarious calibration of version 2.0
Validation Method:Validated the accuracy of predicted aerosol optical thickness at 865 nm (Taua_865) from the
SGLI algorithm comparing with in-situ data of AERONET-OC
Quality Control:• In-situ data: time difference in ±3 hours to SGLI observation
• SGLI data: average of the data passed the following conditions within a 5 by 5 pixel
centered the in-situ point (for details refer to Bailey et al., 2006) :
1. 13 or more pixels which satisfies the following conditions: aerosol optical thickness <
0.3, solar zenith angle < 70 degrees, NWLR of all channels > 0, CLDAFFCTD,
GAMMA-OUT and OVERITER flags aren’t set.
2. Coefficient of variation for Taua_865 nm less than 0.05
Period of Validation• January 1, 2018-February 29, 2020
Validation Results of Ocean Products:
- Atmospheric Correction Parameters: ACP
Achieved the release accuracy
Validation Result Release Accuracy Standard Accuracy Target Accuracy
[Ver. 1] 80%
→ 79% (AOT@865nm)80% (AOT@865nm) 50% (AOT@865nm) 30% (AOT@865nm)
Achieved
Release Accuracy
Accuracy:
80%Accuracy:
79%
N=98 N=142
TAUA_865: Ver. 1 [for reference] TAUA_865: Ver. 2
✓ Kept the release accuracy
✓ Increased the validation points than version 1
Achieved
Release Accuracy
Photosynthetically available radiationMonthly RMS/AVG=9.5%8-day RMS/AVG=15%
Yearly RMS/AVG=5.6%
Daily RMS/AVG=21%
✓ Daily PAR is estimated by instantaneous transmittance from visible channels.
✓ Reference PAR is made from daily SWR observed by buoy, PAR/SWR ratio estimated by Pstar-4 calculation and objective analysis water vapor data
✓ SGLI PAR will be able to achieve the target accuracy; improvement from Ver.1 is mainly from L1B improvement
Validation result Release accuracy Standard accuracy Target accuracy
[Ver. 1] 15% (10km/month)→[Ver.2] 9.5% (10km/month)
20% (10km/month) 15% (10km/month) 10% (10km/month)2.2 - 7
Colors indicate difference of the buoy
Bias for each buoy; dependency seems small
Validation Results of Ocean IWPR Products:
Chlorophyll-a Concentration - CHLA
Major Updates of Algorithm:• None of major updates
Validation Method:Validated the accuracy of predicted CHLA data from the SGLI algorithm comparing with in-situ
data (High-Performance Liquid Chromatography: HPLC or fluorescence method) of ship
observation
Quality Control:• In-situ data: time difference in ±3 hours to SGLI observation
• SGLI data: average of the data passed the following conditions within a 5 by 5 pixels
centered the in-situ point :
1. 13 or more pixels which satisfies the following conditions: aerosol optical thickness <
0.5, solar zenith angle < 70 degrees, NWLR of all channels > 0, CLDAFFCTD flag
isn’t set.
Period of Validation• January 1, 2018-February 29, 2020
Validation Results of Ocean IWPR Products:
Chlorophyll-a Concentration - CHLA
Achieved the standard accuracy
Validation Result Release Accuracy Standard Accuracy Target Accuracy
[Ver. 1] -59~142%
→[Ver. 2] -58~140%-60~+150% (offshore) -60~+150%
-35~+50% (offshore)
-50~+100% (coastal)
Accuracy:
-59~142%Accuracy:
-58~140%
N=108 N=135
CHLA: Ver. 1 [for reference] CHLA: Ver. 2
✓ Increased the validation points
(at time of Ver. 1 release: 9 points → 135 points)
✓ Achieved the standard accuracy (include the coastal area)
Achieved
Standard Accuracy
Achieved
Standard Accuracy
Validation Results of Ocean IWPR Products:
Total Suspended Matter - TSM
Major Updates of Algorithm:• Improved the empirical formula
Validation Method:Validated the accuracy of predicted TSM data from the SGLI algorithm comparing with in-situ
data of ship observation
Quality Control:• In-situ data: time difference in ±3 hours to SGLI observation
• SGLI data: average of the data passed the following conditions within a 5 by 5 pixels
centered the in-situ point:
1. 13 or more pixels which satisfies the following conditions: aerosol optical thickness <
0.5, solar zenith angle < 70 degrees, NWLR of all channels > 0, CLDAFFCTD flag
isn’t set.
Period of Validation• January 1, 2018-February 29, 2020
Validation Results of Ocean IWPR Products:
Total Suspended Matter - TSM
Achieved the standard accuracy
Validation Result Release Accuracy Standard Accuracy Target Accuracy
[Ver. 1] -65~182%
→[Ver.2] -58~138%-60~+150% (offshore) -60~+150% -50~+100%
Accuracy:
-65~182%Accuracy:
-58~138%
N=19 N=34
TSM: Ver. 1 [for reference] TSM: Ver. 2
✓ Validated for in-situ data (at time of version 1 release, in-situ data comparison
was not possible)
✓ Achieved the standard accuracy (issue: continue to validate with wider range
data)
Not Achieved
Standard Accuracy
Achieved
Standard Accuracy
Validation Results of Ocean IWPR Products:
Colored Dissolved Organic Matter - CDOM
Major Updates of Algorithm:• Improved the empirical formula
Validation Method:Validated the accuracy of predicted CDOM data from the SGLI algorithm comparing with in-situ
data of ship observation
Quality Control:• In-situ data: time difference in ±3 hours to SGLI observation
• SGLI data: average of the data passed the following conditions within a 5 by 5 pixel
centered the in-situ point (for details refer to Bailey et al., 2006) :
1. 13 or more pixels which satisfies the following conditions: aerosol optical thickness <
0.5, solar zenith angle < 70 degrees, NWLR of all channels > 0, CLDAFFCTD flag
isn’t set.
2. Median CV (coefficient of variation) computed from NWLR_380-565nm and
Taua_865 nm less than 0.05.
Period of Validation• January 1, 2018-February 29, 2020
Reference: Bailey, S.W., and Werdell, P.J. (2006). A multi-sensor approach for the on-
orbit validation of ocean color satellite data products. Rem. Sens. Environ. 102, 12-23.
Validation Results of Ocean IWPR Products:
Colored Dissolved Organic Matter - CDOM
Achieved the standard accuracy
Validation Result Release Accuracy Standard Accuracy Target Accuracy
[Ver. 1] -55~121%
→[Ver.2] -58~139%-60~+150% (offshore) -60~+150% -50~+100%
N=8 N=14
CDOM: Ver. 1 [for reference] CDOM: Ver. 2
✓ Increased the validation points
(at time of Ver. 1 release: 5points →14 points)
✓ Achieved the standard accuracy (include the coastal area)
Accuracy:
-55~121%
Accuracy:
-58~139%
Ver. 2 CDOM
inputted in-situ NWLR
Accuracy:
-47~88%
Achieved
Standard AccuracyAchieved
Standard Accuracy
The underestimation bias is cased by the
accuracies of the input NWLR. The bias
doesn’t occur when the in-situ NWLR is
inputted, as shown below.
Validation Results of Ocean SST Products:
Sea Surface Temperature- SST
Major Updates of Algorithm:• Improved the cloud mask algorithm (modified cloud mask on coast and inland waters)• Modified the probability density function of cloud mask
Validation Method:Validated the accuracy of predicted SST data from the SGLI algorithm comparing with in-situ
data of iQuam (Ship and buoy observations)
Quality Control:• In-situ data: bouy data from NOAA/ iQuam passed the following conditions:
1. quality control level on iQuam: iquam_flag=0, quality_level=5
2. time difference in ±30 minutes to SGLI observation
• SGLI data: average of the 1km data passed the following conditions within a 5 by 5 pixels
(3 km) centered the in-situ point:
1. Standard deviation of the 5 by 5 pixels < 1.0 degrees C
2. Minimum and maximum temperature differences of the 5 by 5 pixels < 3.0 degrees C
3. Temperature difference between SGLI and iQuam < 5.0 degrees C
4. Flag of quality assurance: Good or Acceptable
Period of Validation• January 1, 2018-Decenber 31, 2019
Validation Results of Ocean SST Products:
Sea Surface Temperature- SST
Achieved the target accuracy in the daytime andthe standard accuracy in the nighttime
Validation Result Release Accuracy Standard Accuracy Target Accuracy
[Ver.1] 0.4 → [Ver.2] 0.4 ℃ (D)
[Ver.1] 0.5 → [Ver.2] 0.7 ℃ (N)* 0.8℃ (Daytime) 0.8℃ 0.6℃
Accuracy:
0.4℃Accuracy:
0.7℃ Ver. 1 Ver. 2
*) In the night time, more pixels on coastal areas became available by
modified cloud detection algorithm, whereas the accuracy deteriorated by
the slight increase of cloud affected pixels.
On coastal area, pixels masked on version 1
as shown green of the left figure became
available as shown red of the right figure.
Achv. Target Achv. Standard
SST (Daytime) SST (Nighttime)
1
Ver2. Cryosphere Products
2
ProductRelease threshold
Standard accuracy
Target accuracy
Status*1 Evaluation Methods
Snow and Ice covered area (incl. cloud detection)
10% 7% 5% ○Comparison with other satellites data (e.g. MODIS, VIIRS, Sentinel-3…).
Okhotsk sea-ice distribution
10% 5% 3% ○ Comparison with other satellites data (e.g. MODIS, VIIRS, Sensinel-3…).
Snow and ice surface Temperature
5K 2K 1K ◎Comparison with in-situ observation (Automatic weather station thermal radiometer data) and other satellites data (e.g. MODIS, VIIRS Sentinel-3…).
Snow grain sizeof shallow layer
100% 50% 30% ○ → ◎
Comparison with climatology (relationship between snow surface temperature and snow grain size) for the release accuracy threshold. In addition, comparison with in-situ data for the standard and target accuracy thresholds.
Evaluation Summary
*1 Symbols denote as follows; 〇: the release threshold achieved, ◎: the standard accuracy achieved,☆: the target accuracy achieved.
Major Change for the C1AB - Snow and Ice cover area algorithm- Revised cloud detection/surface classification algorithm from ordinary threshold method to Neural
network machine learning method- All training data were simulated by DISORT radiative transfer model
Major change for the C1C - Okhotsk sea-ice distribution algorithm- Revised cloud detection/surface classification algorithm from ordinary threshold method to Neural
network machine learning method partly.- All training data using Neural network were simulated by DISORT radiative transfer model
Validation data for the C1AB/ SICE - Snow and Ice cover Extent product (snow/ice fraction > 15%*)- Snow area: MOD10C2 Snow Cover Extent Product- Sea ice area: MOD29E1D Sea Ice Product
Validation data for the C1C/OKID - Okhotsk sea-ice distribution product (sea ice fraction > 15%*)- Sea ice area :MOD29E1D Sea Ice Product
*NSIDC defines sea ice exists in case of the ice fraction/ice concentration more than 15%.
- Version 2 Major changes and validation data details
Validation results of Cryosphere products - SICE
Validation result Release accuracy Standard accuracy Target accuracy
Ver.1: 9.4 % (Mar. 2018 - Aug. 2018)
Ver.2: 8.5 % (Mar. 2018 - Dec. 2019)10 % 7 % 5 %
Ver.2 sample image2019.05.09 - 2019.05.24
Snow Fraction > 15 %2018.03.22 - 2019.12.3116 days composite
N = 40r = 0.99 (p < 0.01)
Rel. Err. = 8.5 %Bias = 462,283 km2
Accuracy improved and SICE product is likely to achieve the standard accuracy
Snow Fraction > 10 %2018.03.14 - 2018.08.208 days composite
N = 21r = 0.99 (p < 0.01)
Rel. Err. = 9.4 %Bias = 744,065 km2
Validation results of Cryosphere products - SICE- Snow and Ice cover extent product validation results using other satellite products
Validation result of Ver. 1 Validation result of Ver. 2
1. Validation period was expanded: 0.5 year to 1.5 year.2. Composite period and valid snow/ice fraction were revised: 8 day to 16 days, 10% to 15% .
Major Change for the C1AB - Snow and Ice cover area algorithm- Revised cloud detection/surface classification algorithm from ordinary threshold method to Neural
network machine learning method- All training data were simulated by DISORT radiative transfer model
Major change for the C1C - Okhotsk sea-ice distribution algorithm- Revised cloud detection/surface classification algorithm from ordinary threshold method to Neural
network machine learning method partly- All training data using Neural network were simulated by DISORT radiative transfer model
Validation data for the C1AB/ SICE - Snow and Ice cover Extent product (snow/ice fraction > 15%*)- Snow area: MOD10C2 Snow Cover Extent Product- Sea ice area: MOD29E1D Sea Ice Product
Validation data for the C1C/OKID - Okhotsk sea-ice distribution product (sea ice fraction > 15%*)- Sea ice area :MOD29E1D Sea Ice Product
*NSIDC defines sea ice exists in case of the ice fraction/ice concentration more than 15%.
- Version 2 Major changes and validation data details
Validation results of Cryosphere products - OKID
Ice Fraction > 15 %2018.02.10 - 2020.03.058 days composite
N = 30r = 0.99 (p < 0.01)
Rel. Err. = 9.0 %Bias = -5,986 km2
Ice Fraction > 10 %2018.03.14 - 2018.08.208 days composite
N = 21r = 0.99 (p < 0.01)
Rel. Err. = 9.1 %Bias = 91,081 km2
False Color Image OKID classification
sea ice
cloud
snow
ocean
land
Validation results of Cryosphere products - OKID- Okhotsk sea-ice distribution product validation results using other satellite products
Ver.2 sample image2020.02.27
Validation result of Ver. 1 Validation result of Ver. 2
1. Validation period was expanded: 0.5 year to 2 year.2. Surface classification was improved from visual evaluation compared with False color image.
Validation result Release accuracy Standard accuracy Target accuracy
Ver.1: 9.1 % (Mar. 2018 - Aug. 2018)
Ver.2: 9.0 % (Feb. 2018 - Mar. 2020)10 % 5 % 3 %
OKID product needs more detail evaluation using high resolution satellite images
- Version 2 Major changes and validation data details
Validation results of Cryosphere products - SIPR
Major Change for the C2AB - Snow grain size of shallow layer- Revised snow grain size estimation algorithm using Neural network machine learning method
- Add the layer numbers of Neural-net: improved the processing speed- Revised the training data set (BRDF data set) using Neural-net: improved inversion accuracy
Major change for the C2AB - Snow and Ice surface temperature- Revised the emissivity table
Validation data for the C2AB/ SGSL - Snow grain size of shallow layer- Field campaign carried out on the Greenland Ice Sheet East-GRIP site (Jul. 2018) and
Japan/Hokkaido Nakasatsunai site (Feb. 2020)- Surface Specific Area (SSA) measured by IceCube and HISSGraS and converted to optical equivalent
snow grain size- All data match-up conditions are in 10 minutes and 250 meters from nearest point of satellites
Validation data for the C2AB/ SIST - Snow and Ice surface temperature- Automatic Weather Station (AWS) installed by PROMICE (Fausto and van As, 2019) have measured
the Upward/Downward Longwave Radiation Flux- Ground surface temperature was converted from Longwave radiation Flux observation- All data match-up conditions are in 10 minutes and 250 meters from nearest point of satellites
Fausto, R.S. and van As, D., (2019). Programme for monitoring of the Greenland ice sheet (PROMICE): Automatic weather station data. Version: v03, Dataset published via Geological Survey of Denmark and Greenland. DOI: https://doi.org/10.22008/promice/data/aws
0
100
200
300
400
500
0 100 200 300 400 500
SG
LI S
no
w G
rain
Siz
e (µ
m)
In-situ Snow Grain Size (µm)
Ver.1 Ver.2
50%
100%
100%50%
★
EGRIP
N = 8r = 0.43 -> 0.31
Rel. Er. = 44.9% -> 28.5%bias = 19.5 -> 12.4 µm
Quality control was improved by cloud detection algorithm revision of SICE.Snow grain size estimation accuracy improved by revision of C2 algorithm.
Validation results of Cryosphere products - SIPR- Snow grain size of shallow layer product validation results using in-situ observation data (only E-GRIP)
Snow grain size estimation accuracy evaluation by comparison betweenVer.1 and Ver.2 results in same quality control condition using Ver.2 SICE result
Validation result Release accuracy Standard accuracy Target accuracy
Ver.1: 45 % (E-GRIP, QC: Ver.2)
Ver.2: 29 % (E-GRIP, QC: Ver.2)100 % 50 % 30 %
Ver.2 sample image2019.05.09 - 2019.05.24
EGRIP
0
100
200
300
400
500
0 100 200 300 400 500S
GL
I S
no
w G
rain
Siz
e (µ
m)
In-situ Snow Grain Size (µm)
Ver.2 EGRIP Ver.2 NSN
50%
100%
100%50%
N = 11r = 0.50
Rel. Er. = 34.1%bias = 19.1 µm
Validation results of Cryosphere products - SIPR- Snow grain size of shallow layer product validation results using in-situ observation data (wide-range)
★
Ver.2 sample image2019.05.09 - 2019.05.24
Adding the in-situ data measured in Nakasatsunai, Hokkaido, JapanSGSL was evaluated in a wide-range of arctic and mid-latitude region
Accuracy improved and SGSL product achieved the standard accuracy
Validation result Release accuracy Standard accuracy Target accuracy
Ver.1: 45 % (only arctic, QC: Ver.2)
Ver.2: 34 % (wide-range, QC: Ver.2)100 % 50 % 30 %
EGRIP
EGRIP
★
N = 33r = 0.97 (p < 0.01)
RMSE = 1.5 K -> 1.4 Kbias = 1.2 K -> 1.0 K
Validation results of Cryosphere products - SIPR- Snow and Ice surface temperature product validation results using AWS observation data (only E-GRIP)
Ver.2 sample image2019.05.09 - 2019.05.24
Snow and Ice surface temperature estimation accuracy evaluation bycomparison between Ver.1 and Ver.2 results using E-GRIP observation data
Validation result Release accuracy Standard accuracy Target accuracy
Ver.1: 1.5 K (EGRIP)
Ver.2: 1.4 K (EGRIP) & 2 K (GrIS, next page)5 K 2 K 1 K
Accuracy improved and SIST product achieved the standard accuracy
EGRIP
ALL
UPE_U
THU_U
KPC_U
SCO_U
KAN_U
N = 135r = 0.93 ( p < 0.01)
RMSE = 1.98 Kbias = 0.9 K
SIST: R/S/G = 5K/2K/1K
RMSE = 0.90 K
RMSE = 1.57 K
RMSE = 1.39 K
RMSE = 1.00 K
RMSE = 4.38 K
Validation results of Cryosphere products - SIPR- Snow and Ice surface temperature product validation results using AWS observation data (whole GrIS)
Snow and Ice surface temperature estimation accuracy evaluationusing all AWSs installed whole region of the Greenland Ice Sheet