Utilizing the Intersection Between Simulated and Observed Hyperspectral Solar Reflectance

Utilizing the Intersection Between Simulated and Observed Hyperspectral

Solar Reflectance

Y. Roberts, P. Pilewskie, B. KindelLaboratory for Atmospheric and Space Physics,

University of Colorado, Boulder, CO

Collaborators: D. Feldman and W. Collins Lawrence Berkeley National Laboratory

SDT Tasks1. Trend Detection in Spectral RadianceTask SummaryObjective: Extract trends in TOA outgoing shortwave spectral radiance.

Method: PCA, examining PC score time series, and SSA/MSSA for trend

extraction. Data: Observed SCIAMACHY and simulated radiative transfer

(MODTRAN) shortwave spectral radianceTools: PCA using IDL/ENVI; SSA; MODTRAN. Expected outcomes: Validation of trend detection methods with

measured shortwave radiance and modeled simulations with known forcings; improved quantification and refinement of CLARREO requirements.

SDT Tasks1. Trend Detection in Spectral Radiances

Roberts, Y., P. Pilewskie, B. C. Kindel. (2011), Evaluating the Observed Variability in Hyperspectral Earth-reflected Solar Radiance, J. Geophys. Res., 116, D24119, doi:10.1029/2011JD016448.

SDT Tasks2. Intersection of Spectrally Decomposed SubspacesTask SummaryObjective: Use intersection to evaluate modeled reflectances with SCIAMACHY

reflectance. Attempt to separate the underlying physical variables that explain the variance in the measurements.

Method: Numerical methods to determine the angles between complementary subspaces. Look-up tables to match model input to variance as depicted by measurement eigenvectors.

Data: Observed SCIAMACHY and simulated radiative transfer shortwave spectral reflectance from Langley and UC-Berkeley groups.

Tools: PCA using IDL/ENVI; MODTRAN; IDL and multivariate numerical methodsExpected outcome: Improved attribution techniques that identify physical

variables driving spectral variability; improved quantification and refinement of CLARREO requirements.

Outline

• Why Reflectance?• Quantitative comparison description• Reflectance PCA results • Reflectance subspace comparison• Method to link model inputs to observations • Examples of intersection attribution method using OSSE and

SCIA data

Why Reflectances for Quantitative Comparison?

• Unstandardized PCA needed in quantitative comparison method– Normalizing by the standard deviation removes important

information about the data sets and what makes them different.

• Without normalizing the data, the spectral shape of the downwelling solar irradiance is still removed through reflectance computation

• SCIAMACHY takes solar irradiance reference measurements and nadir Earth-reflected measurements with the same sensors – the division in calculating reflectance cancels out systematic instrument defects

Comparing SCIAMACHY and OSSE Reflectances

• SCIAMACHY nadir reflectances – Spatial grid: 5.625° (4x the

original OSSE output)– Monthly averaged, spatially

gridded, 10 nm FWHM• OSSEs all-sky reflectances

– Spatially averaged and spectrally resampled over the same spatial grid and with spectral resolution

– Limited to locations present in SCIAMACHY data

Roberts Y., P. Pilewskie, B. C. Kindel, D. R. Feldman, and W. D. Collins, [In preparation] Quantitative Comparison of the Variability in Observed and Simulated Reflected Shortwave Reflectance.

Quantitative Comparison of SubspacesSCIA Reflectances OSSE Reflectances

SCIA Eigenvectors Calculate Intersection

Spectrally Decompose Intersection

The relationship between each pair of transformed eigenvectors. Range = [0,Subspace Dimension]

OSSE Eigenvectors

PCA

SCIA Transformed Eigenvectors

OSSE Transformed Eigenvectors

1 2 3

SVD

Retain 7 PC dimensions for the comparison.

Using similarity significance method found six dimensions to be equivalent.

Intersection Look-up Table Method

SCIA PCA Scores

SCIA Shared Intersection Scores

LUT Shared Intersection Scores

1. For each PC, find the SCIA spectra corresponding to scores more than 3 standard deviations from the mean.

2. Using the spectra found in (1.), calculate the Euclidean distance between the corresponding Shared Intersection SCIA Scores and all LUT Intersection Scores.

4. Examine LUT inputs used to simulate reflectances to understand which model inputs drive measured variance.

3. Find the minimum Euclidean distance for each spectrum.

This finds LUT spectrum with closest spectral shape to SCIA spectrum of interest.

SCIA Reflectances LUT Reflectances

LUT Physical Inputs

PCA Space

Transformed Space

Measurement Space

To use the October 2004 OSSE Reflectances as a LUT, recalculated PCA using all OSSE spectra without re-gridding to align with SCIAMACHY 5° grid.

Four dimensions were used to find the matching spectra between OSSE and SCIA.

Using transformed dimensions with correlations greater than 0.95 work best.

Extreme Positive ScoresExtreme Negative Scores

Six Best Spectra Matches from Most Negative PC01 Scores

SCIAOSSE

Six Best Spectra Matches from Most Positive PC01 Scores

SCIAOSSE

Summary• Reflectance PCA– OSSE and SCIA share 6 dimensions that explain

over 99.5 % total variance– Some physical spectral signals not apparent in

standardized or unstandardized radiance PCA• Intersection Look-up Table Method– Use intersection to match the spectral shape of

observations to simulated spectra efficiently– Quickly matching the spectral shapes provides link

between model physical inputs to observed data variance drivers

Future Work• Applying intersection method to actual LUT for

improved variance driver attribution• Comparison of SCIA and OSSE decadal trends• Trend detection to study centennial time-scale

patterns in OSSEs for different emission scenarios• Quantifying data set differences in addition to

similarities