SLIDE 1 AIRS stratospheric temperature retrievals at full horizontal resolution
Lars Hoffmann1 and M. Joan Alexander2
1Forschungszentrum Jülich, ICG-1, Jülich, Germany 2NWRA/CoRA, Boulder, CO
AIRS Science Team Meeting, Washington, DC, October 2007
SLIDE 2 Outline
1
Motivation
Why do we need temperature data at full horizontal resolution?
2
Forward modelling for AIRS
Brief description of the JURASSIC forward model. Model optimization and validation.
3
Stratospheric temperature retrievals
Brief description of the optimal estimation approach. Retrieval parameter studies and characteristics.
4
First results and summary
Retrieved temperature data for selected AIRS granules.
SLIDE 3 Motivation
AIRS radiance measurements provide information about stratospheric gravity waves on small horizontal scales...
−6 −4 −2 2 4 radiance pert. (667.7 cm−1) [mW/(m2 sr cm−1)] 280° 285° 290° 295° 300° 305° 310° 315° 320° 325° −65° −60° −55° −50° −45° −40°
10−SEP−2003, 04:26 UTC, near Antarctic peninsula (Typical horizontal wavelength in this area: λx ∼ 100 km)
SLIDE 4 Motivation
Example of gravity waves produced by deep convection...
−0.010 −0.005 0.000 0.005 0.010 0.015 0.020 radiance pert. (2362 cm−1) [mW/(m2 sr cm−1)] 115° 120° 125° 130° 135° 140° 145° −30° −25° −20° −15° −10° −5°
12−JAN−2003, 16:44 UTC, near Darwin, Australia
450 km 450 km
⇒ Loss of horizontal resolution in operational temperature retrieval (20 km → 60 km; cloud-clearing) is a drawback for gravity wave studies...
SLIDE 5
Forward Modelling for AIRS
Juelich Rapid Spectral Simulation Code (JURASSIC) Fast radiative transfer model for the mid-infrared spectral region (4 . . . 15 micron, LTE, no scattering, no surface). Approximations for fast radiative transfer calculations:
Band Transmittance Approximation Emissivity Growth Approximation Independent Gas Approximation Look-up tables for spectral mean emissivity
Flexible handling of different types of observation geometry and atmospheric data:
Interpolation of 1D, 2D or 3D atmospheric data (single profiles, satellite track, model output) Observer within or outside atmosphere Nadir, sub-limb, limb or zenith viewing
SLIDE 6
Forward Modelling for AIRS
Modelling of instrument effects:
Spectral filter functions (ILS, SRF,...) Vertical field of view (FOV) Offset and gain calibration
Retrieval interface:
Definition of state and measurement vector (x, b, y) Jacobians by numerical perturbation (z, p, T, qi, kj, c0, c1)
Optimization studies and validation studies:
Optimized ray-tracing step length Optimized emissivity look-up tables Comparisons against MIPAS RFM Comparisons against AIRS SARTA
Documentation and download: https://jurassic.icg.kfa-juelich.de
SLIDE 7
Forward Modelling for AIRS
Optimization of ray-tracing step size...
0.01 0.1 0.1 1 CPU-time [sec] ray-tracing step length [km] model error 0.2% model error 0.5% tropics mid latitudes polar summer polar winter
⇒ CPU-time for forward calculation is about 20 msec on a normal PC. Reduction by a factor 1000 compared to line-by-line reference calculations.
SLIDE 8 Forward Modelling for AIRS
Comparison of JURASSIC and RFM...
0.0002 0.0004 0.0006 0.0008 0.001 0.0012 650 655 660 665 670 675 680 radiance difference (JURASSIC - RFM) [W/(m2 sr cm-1)] wavenumber [cm-1] 15 micron temperature channels AIRS noise tropics mid latitudes polar summer polar winter
⇒ Reference model output is reproduced within AIRS noise. Results for 4 micron channels are similar.
SLIDE 9 Forward Modelling for AIRS
Comparison of temperature kernel functions...
10 20 30 40 50 60 70 0.02 0.04 0.06 0.08 altitude [km] kernel function [mW/(m2 sr cm-1) / K] 15 micron (667.0...670.1 cm-1) JURASSIC RFM SARTA 10 20 30 40 50 60 70 0.0001 0.0002 0.0003 altitude [km] kernel function [mW/(m2 sr cm-1) / K] 4 micron (2360...2380 cm-1) JURASSIC RFM SARTA
⇒ Good agreement! 4 micron kernels are rather broad (due to broad SRFs),
- i. e. provide less information on vertical distribution, but help to reduce noise.
SLIDE 10 Stratospheric Temperature Retrievals
Optimal estimation approach: Find optimal estimate (i. e. MAP solution) of retrieval targets x for given measurements y by minimizing a cost function: J(x) = [y − F(x)]TS−1
ε [y − F(x)]
- measurements – forward calculation
+ (x − xa)TS−1
a (x − xa)
- atmospheric state – a priori
x = atmospheric state y = radiance measurements Sε = measurement error covariance F(x) = simulated observations (forward model) xa = a priori state Sa = a priori covariance
SLIDE 11
Stratospheric Temperature Retrievals
Retrieval grid:
1D case: homogeneously stratified atmosphere Fixed altitudes: 3 km below 60 km, 5 km up to 90 km Retrieve only T, get p from hydrostatic equilibrium.
Measurement error covariance:
Consider only noise (uncorrelated).
A priori data:
Use AIRS operational retrieval results as a priori state (inter/extrapolate data gaps). Use a priori uncertainty of σi = 20 K, correlations from first-order autoregressive model: Sij = σiσj exp(−∆z/cz) Correlation length cz is an important tuning parameter!
SLIDE 12 Stratospheric Temperature Retrievals
Selection of AIRS channels for the retrieval...
0.1 1 10 100 2300 2310 2320 2330 2340 2350 2360 2370 2380 2390 tropospheric fraction of kernel functions [%] wavenumber [cm-1] 4 micron temperature channels 22.5 km 20 km 17.5 km 15 km 12.5 km 10 km 7.5 km 5 km
⇒ Exclude all channels where tropospheric fraction of kernel functions (ztrop = 17.5 km) exceeds 1% to minimize influence of clouds...
SLIDE 13
Stratospheric Temperature Retrievals
Influence of a priori data...
10 20 30 40 50 60 70 80 180 200 220 240 260 280 300 altitude [km] temperature [K] mid latitudes dT = -20 K dT = -15 K dT = -10 K dT = -5 K dT = 0 K dT = 5 K dT = 10 K dT = 15 K dT = 20 K
⇒ Varying the a priori profile by ±20 K causes differences below ±1.5 K in the retrieved profile at 20 . . . 55 km altitude.
SLIDE 14
Stratospheric Temperature Retrievals
Retrieval error due to noise...
10 20 30 40 50 60 70 1 2 3 4 5 6 7 8 altitude [km] retrieval error (noise) [K] cz = 1 km cz = 2 km cz = 5 km cz = 10 km cz = 20 km cz = 50 km cz = 100 km
⇒ We use an a priori vertical correlation length of 50 km to reduce the retrieval error due to noise: The resulting error is 1 . . . 2 K at 20 . . . 55 km.
SLIDE 15 Stratospheric Temperature Retrievals
Vertical resolution...
10 20 30 40 50 60 70 4 6 8 10 12 14 16 18 20 altitude [km] vertical resolution [km] cz = 1 km cz = 2 km cz = 5 km cz = 10 km cz = 20 km cz = 50 km cz = 100 km
⇒ For 50 km a priori vertical correlation length the vertical resolution is 7 . . . 11 km at 20 . . . 55 km.
SLIDE 16 Full Resolution Temperature Data – First results
Gravity waves near Antarctic peninsula...
220 230 240 250 260 temperature (33 km) [K] 285° 290° 295° 300° 305° 310° 315° 320° −60° −55° −50° −45°
full resolution retrieval
220 230 240 250 260 temperature (33 km) [K] 285° 290° 295° 300° 305° 310° 315° 320° −60° −55° −50° −45°
⇒ Full resolution retrieval results resemble operational data, but gravity wave amplitudes are larger.
SLIDE 17 Full Resolution Temperature Data – First results
Gravity waves produced by deep convection...
240 245 250 255 260 265 temperature (42 km) [K] 120° 125° 130° 135° −25° −20° −15° −10°
full resolution retrieval
240 245 250 255 260 265 temperature (42 km) [K] 120° 125° 130° 135° −25° −20° −15° −10°
⇒ Retrieval at full horizontal resolution reveals small-scale structures! Warm bias (about 3 . . . 5 K) in full resolution retrievals at the stratopause.
SLIDE 18
Summary
We use the fast radiative transfer model JURASSIC to simulate AIRS measurements:
The fast model helps to reduce CPU-time by a factor 1000. Reference calculations are reproduced within AIRS noise.
We use the optimal estimation approach to retrieve temperature data for the stratosphere:
Altitude Range: 20 . . . 55 km Vertical resolution: 7 . . . 11 km (about 6 dfs) A priori information: less than 5% Retrieval error (due to noise): 1 . . . 2 K
First retrieval results for selected granules look promising: The full resolution data much better reveal the horizontal small-scale structures caused by gravity waves.
