1

Saharan Dust Corrections for the ENVISAT AATSR SST Product

Xin Kong, Gary Corlett, Lizzie Noyes,

John Remedios and David Llewellyn-Jones

Chris Merchant and Owen Embury

ENVISAT Symposium 2007

2

Contents

• Background

– (A)ATSR Re-analysis for Climate - (A)RC project

– Saharan dust corrections on SST retrieval

• Datasets & Methodology

- AATSR data & SST retrieval

- SEVIRI Saharan Dust Index (SDI)

• Results

– Empirical Comparisons

– Radiative Transfer (RT) Modelling

• Conclusions & Future studies

3

Background

• This work is a part of (A)ATSR Re-analysis for Climate - (A)RC project (re-analysis 16 years (A)ATSR SST records).

• The propose of (A)RC project is to reduce regional biases in retrieved (A)ATSR SSTs to less than 0.1K for all global oceans and to create a very homogeneous SST record

• The re-analysis dataset will be extremely useful for climate change and NWP (* Dr Chris Merchant will give a talk on Section 5C1)

• Saharan dust can cause biases up to 3K on SST retrieval

• The aim of this work is to reduce AATSR SST biases caused by Saharan dust

4

(A)ATSR Sea Surface Temperature (SST) Retrieval

• SST RetrievalN2: n11, n12D2: n11, n12, f11, f12 N3: n3.7, n11, n12D3: n3.7, n11, n12, f3.7, f11, f12(*N-Nadir and D-dual view)

• AATSR design target accuracy ~0.3K

Error sources are cloud screening, water vapour and aerosols etc.

• Our previous results show that Saharan dust caused a warm bias in the AATSR dual-view SSTs (D2&D3) and a cold bias in the nadir-view SSTs (N2&N3).

Day/Night

Night Only

n

iiiTa a0 SST

5

SEVIRI Saharan Dust Index (SDI)

• SEVIRI SDI is a dust indicator, derived from four SEVIRI thermal channels at 3.9, 8.7, 10.8 and 12μm at nighttime (Merchant et al., 2006).

3.9 μm channel is contaminated during the day. A composite technique using the most recent value over the last 24hr to estimate SDI during the day.

6

Datasets and Methodology

Datasets: AATSR and SEVIRI SDI matchups (20 at nighttime and 18 at daytime)

Image Processing: extracted to AOI: Lats[0,30], longs[-50,0], cloud screened (by 1 extra pixel) and spatial averaged at 0.1 degree.

Empirical Comparisons:-AATSR Brightness Temperature (BTs) vs. SEVI SDI (∆BT/∆SDI)-AATSR BTs Difference (BTDs; e.g. n11-n37) vs. SEVI SDI (∆BTD/∆SDI)-AATSR dual minus nadir (D-N) SST vs. SEVI SDI ((∆D-N/∆SDI)

Radiative Transfer Modelling (Rttov-Disort wrapper developed at UE):

Input data are aerosol properties (OPAC & Haywood) and ECMWF atmospheric profile, model can predict TOA BTs for both clear-sky and aerosol conditions.

Dust detection and corrections

Methodology also repeated at some small AOIs to reduce the atmospheric effects on the BTs.

7

Model Simulation Results – BTs vs. SDI(Haywood Aerosol Properties)

Simulated AATSR ΔBT alsovs. water vapour, atmospherictemperature and found that these two parameters are alsocontribute to ΔBT

AATSR BTs depression increase with SDI. Effects are most significant in the 11 and 12μm channels than 3.7 μm channel and more in the forward views

8

Model Simulation Results – SSTs vs. SDI (Haywood Aerosol Properties)

Aerosol effects are most significant N2 retrievals and least significant in D3 retrievals. With increasing SDI: N2 & N3 decrease, D3 increase, D2 decrease using Haywood and not unclear using OPAC, which is not consistent with empirical results

9

Model Simulation Results – indices vs. SDI (Haywood Aerosol Properties)

The presence of dust can be detected in AATSR data, where certain BT relationships and dual-nadir SST differences correlate well with SDI

10

Model Simulation Results – indices for dust corrections

ATSR channels

OPAC HAYWOOD

c m R2 c m R2

N2 Biases vs. indicesD2 – N2 0.4 -0.9 0.81 -0.2 -1.4 0.93

D3 – N3 * 0.8 -1.4 0.85 1.7 -2.4 0.86(n11-n12)-(f11-f12)

-2.0 -5.2 0.80 -7.2 -13.0 0.57

D2 Biases vs. indicesD3 – N3 * - - 0.00 1.0 -0.8 0.74

N3 Biases vs. indicesD2 – N2 0.41 0.11 0.71 -0.4 -0.4 0.76

D3 – N3 * 0.87 0.02 0.93 0.6 -0.8 0.96D3 Biases vs. índices

D2 – N2 0.5 0.2 0.80 0.9 0.1 0.30D3 – N3 * 0.5 0.3 0.74 0.6 0.2 0.56* D3 – N3 applied for nighttime only

11

Empirical Results (Nighttime) over Large Areas: AATSR D-N SST vs. SEVI SDI

Dust Events

Sevi_time N_clear N_sdi>0.2 r

20050618_0100 7155 2031 0.81

20050619_0000 4667 1016 0.59

20050721_0000 4073 1397 0.86

20050802_0000 2036 1425 0.88

20050803_0000 2909 624 0.58

20050803_2300 3746 1455 0.91

20050804_0100 3436 1417 0.86

20050804_2300 2064 1103 0.86

20050904_2300 2004 526 0.8

20050905_2300 3612 2769 0.54

20051005_0000 2967 524 0.8

Non-Dust Events

Sev_time N_clear N_sdi>0.2 r

20050720_0100 5243 703 0.27

20050905_0100 5160 251 0.38

20051004_0100 5578 8 -0.04

20051004_2300 912 10 -0.08

20051212_0000 3505 0 -0.12

20051212_0100 837 1 0.32

20051212_2300 2354 2 -0.02

20051213_0100 1778 1 -0.04

20051213_2300 856 18 0.42

AATSR D-N SST is correlated well with SEVI SDI for the dust events, which consistent with Noyes et al., 2006

12

Empirical Results (Nighttime) over Large Areas - example images on 200507210000

DUST EVENTS!

(a) AATSR D-N SST vs. SEVI SDI (b) AATSR BTDs vs. SEVI SDI

di

The perpendicular distance to the clear-sky line – di is highly correlated with SEVI SDI, r=0.94

positive effect ↑

nega

tive

eff

ect ↓

(c) AATSR BT vs. SEVI SDI is not clear over large areas due to other effect factors (eg., water vapour and temperature), thus a further studies over some small AOIs.

r2 =0.73

13

Empirical and RT modelling Results at nighttime over a small area (AATSR vs. SEVIRI SDI)

Channel/view

Empirical results

(e.g. 200508042300)

RT modelling

(whole 2005 - OPAC)RT modelling (whole 2005 - Haywood)

slope1 r12 slope2 r2

2 slope3 r32

n37 -0.50 0.11 -1.84 0.66 -0.58 0.72

f37 -1.14 0.40 -3.42 0.65 -1.16 0.70

n11 -1.41 0.57 -3.90 0.69 -1.55 0.93

f11 -2.76 0.75 -5.74 0.55 -2.31 0.89

n12 -1.33 0.58 -3.84 0.59 -1.02 0.89

f12 -2.43 0.72 -5.13 0.47 -1.65 0.81

(n11-n12)-(f11-f12) 0.25 0.62 0.58 0.49 0.12 0.62

(n11-n12)-(f37-f11) -1.71 0.83 -3.69 0.80 -1.81 0.86

(f11-f12)-(n37-n12) -1.16 0.70 -4.18 0.78 -1.24 0.59

(f11-f12)-(n37-f11) -2.59 0.73 -4.27 0.86 -1.93 0.86

D2 – N2 - - 3.89 0.62 1.94 0.94

D3 – N3 0.96 0.52 2.39 0.58 1.03 0.73

Empirical and RT modelling results are generally consistent and Haywood simulation is more close to the empirical results which is consistent with Merchant et al., 2006

BTs

BTDs

D-N

14

Empirical Results (daytime) over large areas –example images on 200506181300

Daytime results are similar as nighttime results

r2 =0.77

15

Preliminary Dust corrections – Theoretical- An example on 200507210000 over a small area

SSTs Before

Correction

Delta

SSTs

After

Correction

N2

Cold bias

N3

Cold bias

D3

Warm bias

D2

Warm bias

SEVI SDI

ATSR D-N

SST (K) ∆SST (K)

AATSR D3-N3 indices for dust corrections derived from RT simulations (Haywood) are used in here

Empirical instead!

Threshold:

SDI > 0.2

16

Conclusions

• Saharan dust has significant effects on the AATSR SST retrievals

• Saharan dust has significant effects on the AATSR BTs and aerosol extinctions can be observed over some small areas. However, it is difficult to use single channel BTs for dust detection due to water vapour or temperature effects

• Dust can be detected using AATSR D-N SSTs or AATSR BTDs (e.g., n11-n12 vs. n37-n11)

• Our preliminary dust correction results suggest that Saharan dust could be corrected using the AATSR dual view or multi-channels capabilities. The regional biases on the SSTs caused by Saharan dust can be reduced

• Further work will focus on improve the dust correction technique and validations. The correction technique can be transferred to ATSR-1 and ATSR-2