crn workshop, march 3-5, 2009 1 an attempt to evaluate satellite lst using surfrad data yunyue yu a,...

CRN Workshop, March 3-5, 2009 1

An Attempt to Evaluate An Attempt to Evaluate Satellite LST Using Satellite LST Using

SURFRAD DataSURFRAD Data

Yunyue YuYunyue Yuaa, Jeffrey L. Privette, Jeffrey L. Privettebb, Mitch Goldberg, Mitch Goldbergaa

a a NOAA/NESDIS/StARNOAA/NESDIS/StARb b NOAA/NESDIS/NCDCNOAA/NESDIS/NCDC


OutlinesOutlines

Motivation

Data Sources

Method

Results

Summary


MotivationMotivation

Satellite LST Validation Needs NeedsSatellite LST Validation Needs Needs» Over 30 years LST development at NOAA – Climate Over 30 years LST development at NOAA – Climate

Data RecordData Record

» Validation Needs for NPOESS/VIIRS LST ProductValidation Needs for NPOESS/VIIRS LST Product– Five years Cal/Val plan: 2009 – 2013Five years Cal/Val plan: 2009 – 2013– Core ground data source : 20 SURFRAD/CRN sitesCore ground data source : 20 SURFRAD/CRN sites

» Validation Needs for GOES-R/ABI LST ProductValidation Needs for GOES-R/ABI LST Product– Pre-launch validation plan: 2009 – 2013Pre-launch validation plan: 2009 – 2013– Core ground data source : 10-15 SURFRAD/CRN Core ground data source : 10-15 SURFRAD/CRN

sitessites

» Validation Needs for GIMPAP LST productValidation Needs for GIMPAP LST product

SURFRAD/CRN data plays critical role in NPOESS and GOES-R Programs !!SURFRAD/CRN data plays critical role in NPOESS and GOES-R Programs !!


Motivation (2)Motivation (2)

LST Validation DifficultiesLST Validation Difficulties» In Situ data limitationIn Situ data limitation

– measurement difficulty: emissivitymeasurement difficulty: emissivity

» Effect of cloud contaminationEffect of cloud contamination– Partial or thin cloudy pixels Partial or thin cloudy pixels

» Spatial and temporal variationsSpatial and temporal variations » Angle effectAngle effect

New Method ExplorationNew Method Exploration


Data SourcesData Sources

No.No. Site LocationSite Location Lat/LonLat/Lon Surface Type*Surface Type*

11 Pennsylvania State University, PAPennsylvania State University, PA 40.72/77.9340.72/77.93 Mixed ForestMixed Forest

22 Bondeville, ILBondeville, IL 40.05/88.3740.05/88.37 Crop LandCrop Land

33 Goodwin Creek, MSGoodwin Creek, MS 34.25/89.8734.25/89.87 Evergreen Needle Evergreen Needle Leaf ForestLeaf Forest

44 Fort Peck, MTFort Peck, MT 48.31/105.1048.31/105.10 Grass LandGrass Land

55 Boulder, COBoulder, CO 40.13/105.2440.13/105.24 Crop LandCrop Land

66 Desert Rock, NVDesert Rock, NV 36.63/116.0236.63/116.02 Open Shrub LandOpen Shrub Land

Down-looking PIR at 8 meter height from the ground

UP-looking PIR

Diffuse Radiometer

Down-looking PIR on the towerAt 8-m from ground

Thermometer

Anemometer

Duration of Data: Jan 1 – Dec 31, 2001

GOES-8 and GOES-10 Imager data were applied in validating the LST algorithm using ground data from SURFace RADiation (SURFRAD) budget network stations

Satellite and Ground DatasetsSatellite and Ground Datasets


Data Source (2)Data Source (2)

GeolocationMatch-up

SURFRADData

SatelliteData

TimeMatch-up(<15 mins)

Time SeriesSmoothness Check:Upwelling, Downwelling

Irradiances

SpatialDifference Test:T4 -- 3X3 pix STD,Visual -- 0.5 deg

Channel BTDifference Test:

(Ts, T4), (T4, T2)(T4, T5)

MatchedDataset

ManualTuning

Match-up Flow Chart


Data Sources (3)Data Sources (3)

» The difference between the top of the atmosphere channel 4 brightness temperature from GOES satellite for the spatially closest pixel and the land surface temperature derived from SURFRAD measurements should be generally 5 K or less for clear sky conditions.

» The standard deviation of the 3 by 3 pixel array GOES channel 4 brightness temperature should no exceed 1.5 K.

» The absolute difference between GOES channel 4 and channel 2 brightness temperatures should not exceed 5 K.

» The absolute difference between GOES channel 4 and channel 5 brightness temperatures should not exceed 1 K.

» The time series curves of solar irradiance should be smoothly varying without distortions.

» The time series curves of down-welling infrared irradiance also should be smoothly varying in time without any significant enhancement.

» The average reflectance for the spatially closest GOES-pixel should be generally less than 40% except for snow conditions which can be mostly identified from sequence of hourly GOES images. Snow is more static than clouds.

» Finally the 0.5 degree by 0.5 degree around the SURFRAD site must be visually clear of clouds to form coincident pairs of cloud-free SURFRAD and GOES data.

Match-up Data Processing



Atmosphericprofiles

TOAradiances

MODTRAN

Input (looping)parameters

Start

Surface Type Configuration

AlgorithmCoeffs.

Sensor SpectralResponse Funs

SensorBrightness

Temperatures

Sensor Brightness

TemperatureCalculation

Regression of LST

Algorithms

STD ErrorOf

Algorithms

Input parameters

Filter ofData

Distribution

AlgorithmSelected

AccuracyAnalysis

Sensitivities Analysis

tables

plots

Simulation Process

Regression Process

Analyzing Process

Developing for Satellite LST Algorithm



MonthSite 1 Site 2 Site 3 Site 4 Site 5 Site 6

Day Night Day Night Day Night Day Night Day Night Day Night

1 16 33 46 69 76 154 57 124 84 157 113 245

2 17 45 9 28 36 86 78 139 35 95 96 135

3 0 0 33 92 70 94 77 125 23 58 145 141

4 66 84 28 42 63 89 25 64 44 67 112 74

5 40 69 21 31 107 134 90 64 51 43 158 190

6 26 39 37 54 37 83 27 32 49 64 235 189

7 1 8 34 56 31 48 14 22 48 34 250 226

8 16 33 35 69 12 47 106 106 39 64 188 195

9 46 83 70 110 84 102 69 76 97 123 226 257

10 56 77 66 101 156 213 39 67 28 75 96 152

11 59 118 84 148 47 112 32 94 110 176 85 147

12 25 54 35 99 61 148 38 133 73 124 58 72

Number of Match-up Dataset: Number of Match-up Dataset: GOES-8 and SURFRAD and SURFRAD

Overall: Large number for statistical significance.Overall: Large number for statistical significance.


Data Source (6)Data Source (6)

LST estimation from SURFRAD measurementLST estimation from SURFRAD measurementThe LST values were derived through Stefan-Boltzmann (SB) Law from up-welling irradiance measured by the down-looking broad band PIR in the spectral range from 3 to 50 microns. Stefan-Boltzmann Law can be stated as

e = T4

e -- surface emitted irradiance

-- emissivity of the surface

-- the Stefan-Boltzmann constant

T -- temperature of the surface in degree Kelvin

To obtain the surface emitted irradiance e from the SURFRAD upwelling irradiance

measurements, we first subtract the reflected portion of the down-welling irradiance from the PIR

irradiance value. It is done as in equation (2) given below

pir = e – (1-)d pir -- the up-welling irradiance at the SURFRAD down-looking PIR

d -- the down-welling irradiance at the SURFRAD up-looking PIR


MethodMethod

Direct ComparisonDirect Comparison» Scatter PlotsScatter Plots

» Table of StatisticsTable of Statistics Correlation AnalysesCorrelation Analyses

» Two-measurement ComparisonsTwo-measurement Comparisons


Down-looking PIR at 8 meter height from the ground

UP-looking PIR

Diffuse Radiometer

Down-looking PIR on the towerAt 8-m from ground

Thermometer

Anemometer

Satellite LST: Algorithms applied to GOES-8/10 dataGround LST: Derived from SURFRAD site measurements

Duration: Jan 1 – Dec 31, 2001

SURFRADradiance

Temperature :

emitupdown

emit

Spectral Correction:

T=T+dTpir - dT

dTpir=(dpir/ )(T/4)

dT=(T/4)(d)

GOES 8/10data

Cloud filter

Match-up and Comparison

Statistics

Algorithm:LST Calculation

Plots & Tables

Method (2)Method (2)

Procedure of of Direct Comparison

CRN Workshop, March 3-5, 2009 13Comparison of SURFRAD Estimated LSTs and GOES Retrieved LSTs

Validation Results: FORT PECK, 2001

Results for GOES-8 Results for GOES-10

Results Results ---- Direct Comparison---- Direct Comparison

CRN Workshop, March 3-5, 2009 14Comparison of SURFRAD Estimated LSTs and GOES Retrieved LSTs

Validation Results: Boulder, 2001

Results for GOES-8 Results for GOES-10

Results Results ---- Direct Comparison (2)---- Direct Comparison (2)


Validation Results: direct comparison summary

Good:● Statistical significance● Two satellite validations● Accuracy satisfaction (average RMS=2.1 K)

Issues:● Point-pixel difference● Emissivity inaccuracy● Cloud screen effectiveness

ResultsResults

---- Direct Comparison (3)---- Direct Comparison (3)

SiteGOES-8 GOES-10

Bias (K) STD (K) Bias (K) STD (K)

1 1.47 1.87 N/A N/A

2 1.06 2.26 -0.41 2.27

3 0.15 1.31 -0.84 1.53

4 -0.27 2.10 -0.97 1.87

5 -0.59 1.67 -1.37 1.91

6 -1.63 1.69 -1.69 1.30

Total -0.37 1.76 -1.23 1.69

RMS 1.80/2.07* 2.09/2.17*

*weighted through each site*weighted through each site



Two-Measurement MethodLinear Approach Model

goesgoesgoes aLSTLST

surfradsurfradsurfrad bLSTLST

LST -- True LST

LSTgoes – The GOES LST measurement

LSTsurfrad – the SUFRAD LST estimation

goes – random noise of the GOES LST measurement

surfrad – random noise of the SURFRAD LST estimation

Perform Variation/Co-variation Computation

)(),()()( 2goesgoesgoesgoesgoes VARLSTCOVLSTVARLSTVAR

)(),()()( 2surfradsurfradsurfradsurfradsurfrad VARLSTCOVLSTVARLSTVAR

),(),(

),()(),(

surfradgoessurfradgoes

goessurfradsurfradgoessurfradgoes

COVLSTCOV

LSTCOVLSTVARLSTLSTCOV



Two-Measurement Method

Primary Assumptions

COV(goes, surfrad)COV(LST, goes) = 0,

COV(LST, surfrad) = 0,

Therefore

surfradgoes

surfradgoes LSTLSTCOVLSTVAR

),(

)(

),()(2surfradgoes

surfrad

goesgoesgoes LSTLSTCOVLSTVAR

),()(2surfradgoes

goes

surfradsurfradsurfrad LSTLSTCOVLSTVAR


Results Results ---- Correlation Analyses---- Correlation Analyses

Relative bias V(goes) V(surfrad) V (goes-surfrad) CV(goes, surfrad) ( goes, surfrad)

0.0017 39.971 34.645 2.096 36.260 0.974

No goessurfrad goes surfrad

1 1.05 1.378 0.334

2 1.06 1.239 0.661

3 1.07 1.083 0.870

4 1.08 0.900 1.035

5 1.09 0.669 1.174

6 1.10 0.293 1.230

Samples for GOES-8 LST vs SURFRAD: Bondeville Site Case StudyBondeville Site Case Study

In the table, the Relative bias is defined as

Relative bias = surfradgoes

surfradgoes

LSTLST

LSTLST

0.2 ;

Pearson Linear correction coefficient, defined as

)()(

),(

surfradgoes

surfradgoes

LSTVARLSTVAR

LSTLSTCOV . (

Using Flynn’s estimation for the ratio ,

BAAB 1/m ,m surfrad

goes

(

where mAB=COV(LSTsurfrad,,LSTgoes)/VAR(LSTsurfrad) and

mBA= COV(LSTgoes, LSTsurfrad )/VAR(LSTgoes),rspectively.

In our case, the range is calculated as [1.047 to 1.102].


SummarySummary

SURFRAD ground station data were used for GOES-R LST SURFRAD ground station data were used for GOES-R LST algorithm evaluation.algorithm evaluation.

GOES-8, -10 Imager data were used as proxies of GOES-R GOES-8, -10 Imager data were used as proxies of GOES-R ABI.ABI.

LST algorithm coefficients were derived from a radiative LST algorithm coefficients were derived from a radiative transfer simulation model (MODTRAN).transfer simulation model (MODTRAN).

Match-up dataset of satellite and ground data were created Match-up dataset of satellite and ground data were created carefully.carefully.

Direct comparisons indicate a promising algorithm accuracy.Direct comparisons indicate a promising algorithm accuracy. Correlation analyses showed good algorithm precision Correlation analyses showed good algorithm precision Further works will be performed using three-measurement Further works will be performed using three-measurement

comparison comparison


Backup slidesBackup slides


Two-directions from GOES SatellitesTwo-directions from GOES Satellites

135° W 75°W


Difference of LSTs observed by GOES-10 and GOES-8 imager at the same location of SURFRAD station Desert Rock, NV, 36.63ºN, 116.02ºW. The simultaneous observation pairs are about 2096.

View zenith of GOES-8: 60.140

View zenith of GOES-10: 46.810

LST Directional Effect in LST Directional Effect in GOES-8 and -10 ImagerGOES-8 and -10 Imager


Goodwin Creek, MS, observation pairs are about 510. View Zenith of GOES-8/-10: 42.680/61.890

LST Directional Effect in LST Directional Effect in GOES-8 and -10 Imager (2)GOES-8 and -10 Imager (2)


Boulder, CO, observation pairs are about 510. View Zenith of GOES-8/-10: 42.680/61.890




Bondville, IL. Data pairs: 710 Fort Peck, MT. Data pairs: 912

View Zenith of GOES-8: 48.120




Note the difference of the two sites

CRN Workshop, March 3-5, 2009 26Daytime Scatter plot comparison of the GOES LST and the SURFRAD

LST for all the match-up data in 2001, within 6 SURFRAD sites.

CRN Workshop, March 3-5, 2009 27Nighttime scatter plot comparison of the GOES LST and the SURFRAD LST for all the scatter plot comparison of the GOES LST and the SURFRAD LST for all the match-up data in 2001, within 6 SURFRAD sites.match-up data in 2001, within 6 SURFRAD sites.

CRN Workshop, March 3-5, 2009 28Dry atmos condition scatter plot comparison of the GOES LST and the SURFRAD LST scatter plot comparison of the GOES LST and the SURFRAD LST for all the match-up data in 2001, within 6 SURFRAD sites.for all the match-up data in 2001, within 6 SURFRAD sites.

CRN Workshop, March 3-5, 2009 29Moist atmos condition scatter plot comparison of the GOES LST and the SURFRAD LST scatter plot comparison of the GOES LST and the SURFRAD LST for all the match-up data in 2001, within 6 SURFRAD sites.for all the match-up data in 2001, within 6 SURFRAD sites.

crn workshop, march 3-5, 2009 1 an attempt to evaluate satellite lst using surfrad data yunyue yu a,...

Documents

imager data

crn workshop

core ground data source

lst validation difficulties

surfrad data yunyue

lst algorithm

npoessviirs lst product

years lst development