Algorithm for MERIS land surface BRDF/albedo retrieval and its validation using contemporaneous EO data products

Transcription

1 Algorithm for MERIS land surface BRDF/albedo retrieval and its validation using contemporaneous EO data products Jan-Peter Muller* (UCL) Carsten Brockmann, Marco Zühlke, Norman Fomferra (BC) Jürgen Fischer, Réné Preusker, Thomas Schröder (FUB) Peter Regner (ESA/ESRIN) *Professor of Image Understanding and Remote Sensing MODIS & MISR Science Team Member (NASA EOS Project) HRSC Science Team Member (ESA Mars Express 2003) Chair, CEOS-WGCV Terrain mapping sub-group

2 Overview Objectives BRDF/Albedo retrieval approach BRDF/Albedo algorithm details Initial Results Validation approach Preliminary Validation results Future Prospects

3 Objectives Derivation of a one-year (2003) land surface albedo from MERIS for 13 of the 15 MERIS wavelengths (excluding 2 inside O2 absorption bands) 4 broadband albedos ( µm, µm, 0.7-3µm, 0.4-3µm) 16-day and MONTHLY time step for 2003 Input Level 2 Rayleigh+O3 corrected 0.05º and 10km sinusoidal spatial resolutions» And a desire to publish MERIS albedo browse images (as Web Map Services layers) within the CEOS-WGISS EO Data Portal ( with the associated albedo files downloadable through a cascaded Web Coverage Server Main driver is to improve the retrieval of atmospheric parameters from MERIS. Hence, we need spectral albedos at the MERIS wavelengths Validation by inter-comparison with other EO sensors and possibly against BSRN data

4 BRDF/albedo approach(1) Inputs will be orthorectified, cloud-cleared, atmospherically-corrected Spectral/Surface Directional Reflectances (SDRs) from level 2 data at 1.2km spatial resolution and a typical sampling of every 2-3 days BRDF retrieval will NOT be directly performed on these SDRs as sampling of the bi-directional plane is insufficient for most land surfaces given the narrower swath (1130km) and lower temporal sampling (every 3 days at the equator) of MERIS cf. instruments such as MODIS (2550 km and daily sampling) Instead the BRDF shape and BRDF models will be taken for the 4 common spectral bands from the MOD43C2 (0.05º) product (see below for an intercomparison). N.B. also common with MISR/POLDER MERIS MODIS Spectral Bands 15 (13 for lev el 2 36 (7 for land produ cts) (seetable 2) products) nm nm (for land) IFoV (nadir) 300m (Ful l Resolution) 1200m (Reduced Resolution) 250m (2 lan d bands) 500m (5land bands) 1000m (29 non -land bands) FoV (swath width) ±34.25¼ (1150km) ±55¼ (2330km) Global coverage 3 days 1-2 times/day(both Terra and Aqua) Repeat cycle 35 days 16 days Orbit Altitude 800km 705km Equatorcrossing 10am (de scending) 10:30am escending-tera) (d 1:30pm scending-aqua) (a Launch date 1 March2002 (ENVISAT) 18 December 1999(Terra) MERIS MODIS MISR POLDER 490± ±10 560± ±10 665± ±10 865± ±

5 BRDF/albedo approach(2) Using a technique called magnitude inversion (see later), MERIS BRDFs are calculated for each set of SDRs which are co-located with the MODIS 0.05º pixel where MODIS returns a value Currently, linear spectral interpolation is performed for the remaining 9 MERIS spectral bands. In future, it is planned to use spectral databases such as ASTER or SDRs from CHRIS/PROBA to refine this approach Currently spectral interpolation for the 4 sets of broadband albedos ( µm, µm, 0.7-3µm, 0.4-3µm) is performed using the MISRequivalent bands. Work is in progress to refine this approach QC information will be provided for 4 common spectral albedos and Nadir BRDF Adjusted Reflectances (NBAR) through statistical summaries of intercomparison with MOD43C1 (albedo)/mod43c3 (NBAR) When cloud cover is too high either in MODIS albedo products or in MERIS SDRs, a gap-filled MODIS product (Moody et al., 2004) will be employed to fill in gaps

6 BRDF retrieval: vegetation Kernel-Driven Semiempirical BRDF Model BRDF Model Linear combination of two BRDF shapes and a constant BRDF shapes described by kernels, which are Trigonometric functions of incidence and view angles Derived from physical models for surface scattering (Ross-Thick Li- Sparse Model Reciprocal (RTLSMR) for leaf cloud and shadows) Analytical Form: R = f iso + f geo k geo + f vol k vol where f iso ; k geo, k vol is a constant for isotropic scattering are trigonometric functions f geo providing, f vol shapes for geometric-optical and volume-scattering BRDFs; and MOD43C2 Product supplies values of f for each are 0.05º constants pixel and separate that C weight code to calculate k the two BRDFs

7 BRDF retrieval: vegetation Magnitude inversion We determine a on a per-band basis by a least squares minimisation of the difference between directional reflectances (SDRs) predicted by the MOD43C2 BRDF parameters and those actually measured by the MERIS sensor. The predicted measurements are found by running the RTLSMR model in the forward mode using the MOD43C2 BRDF parameters under the same view and illumination angles as MERIS measurements available R MERIS (θ,υ,φ,λ) = a R MODIS (θ,υ,φ,λ) Performed on 4 common spectral bands between MODIS (469,555,645,859) and MERIS (490,560,665,865)

8 Albedo retrieval: vegetation Black-sky, White-sky and solar zenith dependence Direct Hemispherical Reflectance, h k (θ ) is given by h k (θ ) = 1 π 2π π /2 K k Black-sky (NO diffuse) albedo, 0 0 (θ,υ,φ )sin(υ)cos(υ)dυdφ a bs is given by a bs (θ,λ) = f k (λ)h k (θ ) Diffuse bi-hemispherical reflectance, H k = White-sky (diffuse ONLY) albedo, a ws (λ) = k π /2 h k 0 k (θ )sin(θ )cos(θ )dθ f k (λ)h k H k a ws is given by is given by

9 Albedo retrieval: vegetation Black-sky, Blue-sky and solar zenith dependence The solar angle dependence can be approximated by, a bs (θ,λ) = f iso (λ)(g oiso + g 1iso θ 2 + g 2iso θ 3 )+ f vol (λ)(g ovol + g 1vol θ 2 + g 2vol θ 3 )+ f geo (λ)(g ogeo + g 1geo θ 2 + g 2geo θ 3 ) g jk for kernel, k k=isotropic k=rossthick k=li-sparse g ok (term 1) g ok (term θ 2 ) g ok (term θ 3 ) White-sky Under actual atmospheric conditions given the aerosol optical depth, τ the blue-sky albedo is given by a(θ,λ) = { 1 S(θ,τ(λ))}a bs (θ,λ)+ S(θ,τ(λ))a ws (θ,λ) Where S(θ,τ(λ)) is the fraction of diffuse skylight

10 Albedo retrieval: vegetation Narrow-to-broadband conversion Gao et al. (2003) derived a first approximation to broadband albedo conversion factors based on those from MISR which are taken from his paper with VIS ( µm), NIR (0.7-3µm) and Shortwave (0.4-3µm) VIS-Broadband NIR-Broadband SW-Broadband Blue Green Red NIR Intercept

11 Meris L2 SDRs Albedo retrieval scheme MOD43C2 BRDF (0.05º) + QA#1 flags BIN MERIS SDRs (0.05º x 0.05º) over 16-day MOD43C2 QA#2 Nsamps, ave± stddev, min, max MOD43C3 NBAR (0.05º) MAGNITUDE INVERSION with MOD43C2 CALCULATE MERIS NBAR 0.05º DAILY CALCULATE <MERIS> NBAR OVER MODIS 16 DAY PERIOD MERIS 0.05º 16- DAY NBAR INTERCOMPARE WITH MOD43C3 INTEGRATE MERIS ALBEDO FOR 16- DAY PERIOD MERIS 0.05º 16- DAY ALBEDOS INTERCOMPARE WITH MOD43C1 DIFF STATS MONTHLY/ SEASONAL AVERAGE RE-PROJECT TO 10Km/0.05º QA3 Nsamps, ± std.dev. MOD43C1 ALBEDO (0.05º) INTERPOLATE ALBEDO VALUES TO 9 OTHER BANDS + INTEGRATE TO VIS, NIR, SW Broadbands MERIS 10KM/005º 13- SPECTRAL + 4 BROADBAND MONTHLY+ SEASONAL ALBEDOS N.B. Status: Initial products undergoing testing production due to start in October, to be completed by December 2005?

12 First MERIS albedo product: DoY 257 (16- day time period : 29/8/03-13/9/03): all bands

13 First MERIS albedo product: DoY 257 (16-day time period : 29/8/03-13/9/03): Band 5 (green)

14 Validation approach(1) Difference statistics between MERIS-Albedo and MOD43C1 for common bands are being analysed for the same 16-day time periods Overlapping MERIS swath NBARs (Nadir-equivalent BRDF Adjsuted Reflectance) will be used to assess how accurate the BRDF correction has performed as well as detect poorly corrected aerosol correction and poorly masked clouds Inter-comparisons will be performed with MISR 0.5º true monthly level-3 product (2003) POLDER2 0.05º resampled from 6km sinusoidal gridded 30-day products reported on the 15th of each month (Apr03-to-Oct03) MOD43C1 0.05º sampled for best albedo value of two 16-day time periods within the months of Jan, Feb, Sep, Oct, Nov-03 Initial inter-comparisons follow with POLDER2 and MISR

15 Validation issue: finding temporal coincidences (MOD43) Date Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

16 MERIS (16-day,DoY= ) cf. POLDER2 (30-day, DoY= ) at 0.05º resolution MERIS: 665,560,442.5nm POLDER2: 670,565,443nm N.B. Poor Atmospheric correction for POLDER-2

17 MERIS (16-day,DoY= ) cf. MISR (30-day, DoY= ) at 0.5º resolution 672,558,446nm 665,560,490nm 867,672,558nm 865,665,560nm N.B. MERIS higher Albedo cf. MISR MISR MERIS

18 Future Prospects Systematic production of global MERIS land surface albedo will commence in October 2005 Monthly and 16-day intercomparisons will be performed against MODIS, MISR and POLDERII Browse images in WMS format will be published through ICEDS DIMAP format WCS-accessible data will hopefully be made available through a cascaded service

19 ICEDS portal (