Errors and correction of Metop/AVHRR derived SST in

ERRORS AND CORRECTION OF METOP/AVHRR DERIVED SST IN
ARCTIC CONDITIONS
Pierre Le Borgne, Sonia Péré, Hervé Roquet
Centre de Météorologie Spatiale, Météo-France, Lannion
Abstract
A positive bias has been identified on METOP/Advanced Very High Resolution Radiometer (AVHRR)
derived daytime Sea Surface Temperature (SST) data in the Arctic. In a preliminary study an increase
of the positive bias was found surprisingly correlated with an increase of the water vapor in the low
layers of the atmosphere. A dedicated experiment has been conducted on the European Arctic
“WASPARC” database to test the ability of using Numerical Weather Prediction (NWP) model outputs
to analyze this paradox and correct METOP/AVHRR systematic errors in the Arctic. Simulated
brightness temperatures (BTs) at 10.8 and 12 microns using the Radiative Transfer for TIROS
Operational Vertical Sounder model (RTTOV) and the regional High Resolution Limited Area Model
(HIRLAM) profiles have shown small biases compared to observed BTs and a simple daily constant is
added to simulated BTs to adjust to the observations. A correction field has been derived each day by
applying the operational algorithm to the simulated BTs. This correction reduces the average
difference to the Met Office SST analysis (OSTIA) from 0.57 to –0.05K and the difference to drifter
measurements from 0.43 K to –0.19 K. The positive bias is due to relatively high humidity below 800
hPa, associated with a temperature inversion. In such a case, any increase of humidity in the low
layers increases the positive bias of the calculated SST. This brief experiment confirms that the
sporadic positive biases observed on METOP results in the Arctic are of atmospheric origin.
INTRODUCTION
The EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF) has produced
METOP/Advanced Very High Resolution Radiometer (AVHRR) derived Sea Surface Temperature
(SST) since 2007. As shown by figure 1, a positive bias, typical of the Arctic, has been observed
regularly from summer 2008 till August 2011 (Poulter and Eastwood, 2008).
.
Figure 1 : Daytime SST validation results against drifters since April 2008 for the full METOP zone (black line), the
North East Atlantic (dark blue line) and the Arctic (light blue line).
This study will focus on summer 2008. For this year the bias is more significant in July than in June,
with a peak in the first 10 day period of July. The characteristics of the bias suggested a possible
diurnal warming effect. A detailed study of this effect has been conducted (Eastwood et al 2011) but it
was finally not found responsible for the positive biases (at least at this scale). In a preliminary study,
the METOP/AVHRR derived SST errors have been analyzed at CMS as a function of water vapor
profiles, and an increase of the positive bias was found surprisingly correlated with an increase of the
water vapor in the low layers of the atmosphere (1000 to 800 hPa). The present study aims to identify
the reasons for this paradox, and find a correction method.
The data used for this study are extracted from the WArm SPots in the ARCtic database (WASPARC,
Eastwood et al 2011). The way they have been preprocessed for this study is presented in the ‘Data’
section. The method we used is basically the same as developed in LeBorgne et al, 2011 for the
geostationary case. This method uses simulated brightness temperatures (BTs) derived from the
Radiative Transfer for TIROS Operational Vertical Sounder model (RTTOV, Saunders, 2008) RTTOV
applied on HIRLAM profiles. They have been corrected by comparison with the observed
METOP/AVHRR BTs as described in section 3 (‘Simulations results’). A correction to the
METOPAVHRR daytime derived SST is calculated from the simulated BTs. The corrected SSTs are
compared to OSTIA, and validated against drifter measurements. These validation results are
presented in the ‘Validation’ section. The origin of the biases is discussed in section 5 through the
detailed analysis of a case study.
DATA
This study uses the METOP/AVHRR granule “workfiles” remapped onto the WASPARC grid covering
the European Arctic (Figure 3). The METOP/AVHRR workfiles are CMS internal METOP SST
products (undelivered but for specific projects); they contain the observed brightness temperatures
(BTs) and other intermediate calculation ingredients in addition to the calculated SST. This study used
also the 3 hourly High Resolution Limited Area Model (HIRLAM) model forecasts and the Met Office
OSTIA (Stark et al, 2007) SST analysis of the day as input to RTTOV.
Figure 3: The area covered by the WASPARC data set with bathymetry and basin names (from Eastwood et al, 2011)
The METOP/ AVHRR workfiles have been preprocessed as follows: the confidence level 4 and 5 SST
data have been collected in one file for each “night” corresponding to local solar time within 4 hours
from the local midnight. In case of conflict, priority is given to data showing the highest solar zenith
angles. Brightness temperatures are collected consistently. These daily “collated” METOP SST files
are the basis of the further processing steps.
BRIGHTNESS TEMPERATURE SIMULATION RESULTS
RTTOV has been applied to the HIRLAM profile corresponding to 00:00 UTC each day. Simulated BTs
at 3.7, 10.8 and 12.0 microns have been calculated for every cloud free pixels of the collated files,
using the OSTIA SST values as surface temperatures. The 3.7 micron temperature case however will
not be discussed here, the area being almost entirely under daytime conditions throughout the period.
(a)
(b)
(c)
(d)
th
Figure 4: 5 of July 2008 case:
(a) OSTIA; (b) observed 10.8 micron BTS; (c) simulated 10.8 micron BTS; (d) difference between simulated and
observed 10.8 micron BTS
As expected and apparent on Figure 4, simulations do not fit exactly observations. In the particular 5th
of July case, simulations are colder than observations off the Norwegian coast (Figure 4d), which
could correspond to the remaining of a DW event the preceding day, or any other true temperature
difference between OSTIA and observations. However, as shown by Figure 5, the difference between
simulations and observations is rather stable throughout the period, with mean differences of -0.22 K
and -0.03 K for T10.8 and T120, respectively. The differences in channel 10.8 and 12.0 microns are
quite correlated, as demonstrated by the stability of T10.8-T12.0 throughout the period.
(a)
(b)
(c)
Figure 5: Simulated –observed BTS in June-July 2008. (a) T10.8, (b) T12.0, (c) T10.8-T12.0
We did not find any convincing analytic expression of a BT simulation adjustment as a function of the
key parameters potentially affecting simulations: SST, integrated water vapor and satellite zenith
angle. In consequence we decided to adjust the simulations to the observations of the day by adding
simply the mean difference observed this day: one constant value per day as shown in Figure 5.
VALIDATION OF SST CORRECTION
As in the geostationary case, SST are calculated by applying the NL operational algorithm to the
adjusted simulated BTs (equation (1), see also METOP/AVHRR SST users manual at http://www.osisaf.org )
SST = a T11 + (b TCLI + c Sθ) (T11 - T12) + d + eSθ + corr
NL :
(1)
T11, T12 are the brightness temperatures at 10.8 and 12.0 microns, respectively; corr is the correction
term resulting from preliminary adjustment on the MDB; Sθ=
 sec(θ) –1, θ is the satellite zenith angle
and TCLI is the mean climatological value.
The difference between this calculated SST and the input surface SST (OSTIA) gives an estimate of
the inability of the algorithm to retrieve the surface temperature, given the atmospheric conditions.
(a)
(b)
(c)
th
Figure 7: 5 of July case: (a) Operational – OSTIA difference (; (b) simulated error; (c) corrected SST – OSTIA.
Figure 6 shows an example of results for the 5th of July: After correction, the corrected –OSTIA SST
difference, null on the average, reveals structure off Norway, likely resulting from the DW event the
preceding day and not resolved by OSTIA, as discussed in the preceding section.
(a)
(b)
Figure 7: Daily validation results; solid line: before correction; dashed line: after correction. (a): results of comparison
with OSTIA, (b): validation results against drifters.
Figure 7a shows the results of the comparison with OSTIA. After correction, the difference is
systematically null on the average (from 0.57 K before to -0.05 K after correction) but the standard
deviation is not systematically reduced, revealing that there are structures in this difference that are
not induced by atmospheric artifacts in the METOP/AVHRR SST. The validation results against buoys
(Figure 7b) confirm also that the positive bias has disappeared after applying the correction (from 0.43
K before to –0.19 K after correction). Figure 7b shows validation results against buoy corresponding to
METOP orbits closest to midnight. In this time of the year, it corresponds to twilight conditions, where
cloud mask errors are more frequent and may result in negative biases .
BIAS ORIGIN: A CASE STUDY
The 5th of July case has been selected because it shows a peak of the positive bias. Figure 6 shows
that the simulated error fits well with the operational – OSTIA SST difference. Figure 8 shows that the
T10.8-T12.0 temperature difference is well reproduced by the simulations. This figure shows also 2
features of the brightness temperatures that are unusual and somewhat counter intuitive: In the region
of the largest positive errors the brightness temperatures at 11 microns is larger than the SST itself
(Figure 8c) and, considering the sign of T10.8-T12.0 differences (Figure 8a), this is even more
pronounced at 12 microns. The second intriguing fact is an increase of T12.0 as the integrated water
vapour content W increases, as shown by the sign of dT12.0/dW (Figure 8d). Formally, the NL
algorithm could cope with the situation (equation 1), however its coefficients, determined for a global
processing, are not adapted to the situation..
To illustrate this effect in the 5th of July cases, two points have been selected :
A : 70.53 N, 14.23W and B : 65.46N, 23.20E.
A is situated in a high simulated error area (DT=1.90 K), whereas B is a more normal case (although
still showing a positive error: DT=0.90 K)
(a)
A
(b)
B
(c)
(d)
th
Figure8: 5 of July case: (a) observed and (b) simulated T108-T120 BT difference; (c) : simulated T10.8-OSTIA SST ;
(d) : DT12.0/DW
Figure 9 shows a strong temperature inversion in the low layers in A with relatively colder surface
temperatures than in B. Low layers are also more humid in A than in B, which is consistent with the
paradox observed earlier at CMS.
The largest difference between A and B is however given in Figure 10, showing the Jacobian profiles
in A and B: dT/dq where q is the specific humidity and dT/dt where t is the air temperature at pressure
p. T is either the Top Of Atmosphere (TOA) 10.8 micron or the T12.0 BTs, or the SST deduced from
them by applying (1). The dT/dq profiles in A are quite unusual, where an increase of humidity in lower
layers induces an increase of the TOA BTs, and this increase is larger for T12.0 than for T10.8. In B
profiles are similar to what is commonly observed over the oceans. Note that there is a neutral point at
about 700 hPa, in each case, but with opposite trends in A and B.
0
200
200
400
400
Pressure
Pressure
0
600
600
800
800
1000
1000
200
220
240
260
280
200
300
0
200
200
400
400
Pressure
Pressure
220
240
260
280
300
T in K
T in K
0
600
600
800
800
1000
1000
0.000
0.005
0.010
Spec. Humidity in g/g
0.015
0.000
0.005
0.010
Spec. Humidity in g/g
0.015
Figure 9: Profiles of temperature (top) and specific humidity (bottom) in A (left) and B (right)
0
0
200
200
11 micron
400
Pressure
Pressure
400
600
600
Calc. SST
800
800
1000
12 micron
1000
0
200
200
400
400
Pressure
Pressure
0
600
600
800
800
1000
1000
Figure 10: Profiles of Jacobians: dT/dq (top) and dT/dt (bottom) in A (left) and B (right)
CONCLUSIONS
This experiment was dedicated to test the ability of using NWP outputs to interpret and correct
METOP/AVHRR systematic errors in the Arctic. Simulated BTs at 10.8 and 12 microns using RTTOV
and HIRLAM profiles have shown small biases compared to observed BTS (-0.22 and –0.03K
respectively for 10.8 and 12.0 micron). A simple daily constant is applied to simulated BTs to adjust to
the observations. A correction field has been derived each day by applying the operational algorithm
to the simulated BTs. Applying this correction reduces the average difference to OSTIA from 0.57 to
–0.05K and the difference to drifter measurements from 0.43 K to –0.19 K. These last figures suggest
that other factors than algorithm limitations explain the difference with buoy measurements. The
positive bias is due to relatively high humidity below 800 hPa, associated with a temperature inversion.
In such a case, any increase of humidity in the low layers increases the positive bias of the calculated
SST, explaining the paradox found earlier at CMS. This brief experiment confirms that the sporadic
positive biases observed on METOP results in the Arctic are of atmospheric origin. Simulations using
RTTOV and HIRLAM outputs provide good quality simulated BTs. These are encouraging results to
use NWP derived simulated METOP AVHRR BTs on the global scale.
ACKNOWLEDGEMENTS
The authors would like to thank David Poulter for his contribution to building the WASPARC data base
under a EUMETSAT OSI SAF visiting scientist funding.
REFERENCES
Poulter, D and Eastwood, S., 2008, “Validation of the OSI SAF products in Polar regions”, EUMETSAT
VS report. http://www.osi-saf.org
Eastwood, S., LeBorgne, P., Péré, S. and Poulter, D., 2011, “Diurnal variability in Sea Surface
Temperature in the Arctic”, Remote Sensing of Environment, Volume 115, Issue 10, 17 October 2011,
Pages 2594-2602
LeBorgne, P. Roquet, H. and Merchant,C., 2011, “Estimation of sea surface temperature from the
Spinning Enhanced Visible and Infrared Imager, improved using numerical weather prediction”,
Remote Sensing of Environment,Volume 115, Issue 1, 17 January 2011, Pages 55-65.
Saunders, R. (2008).
http://www.nwpsaf.org
RTTOV-9
science
and
validation
report.
NWPSAF-MO-TV-020,
1,
Stark, J.D, Donlon, C.J, Martin, M. J. and McCulloch, M.E., 2007, “OSTIA : An operational, high
resolution, real time, global sea surface temperature analysis system”, Oceans '07 IEEE Aberdeen,
conference proceedings. Marine challenges: coastline to deep sea. Aberdeen, Scotland.IEEE.