1 / 22

SST ALGORITHMS IN ACSPO REANALYSIS OF AVHRR GAC DATA FROM 2002-2013

SST ALGORITHMS IN ACSPO REANALYSIS OF AVHRR GAC DATA FROM 2002-2013. B. Petrenko 1,2 , A . Ignatov 1 , Y. Kihai 1,2 , X. Zhou 1,3 , J . Stroup 1,4 1 NOAA/STAR , 2 GST, Inc ., 3 CIRA, 4 STG , Inc. Objectives.

neylan
Download Presentation

SST ALGORITHMS IN ACSPO REANALYSIS OF AVHRR GAC DATA FROM 2002-2013

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. SST ALGORITHMS IN ACSPO REANALYSISOF AVHRR GAC DATA FROM 2002-2013 B. Petrenko1,2, A. Ignatov1 , Y. Kihai1,2, X. Zhou1,3, J. Stroup1,4 1NOAA/STAR, 2GST, Inc., 3CIRA, 4STG, Inc. Reprocessing AVHRR GAC

  2. Objectives • Following a request from the NOAA Coral Reef Watch Program, 2002-2013 AVHRR GAC data from NOAA-16, -17, -18, -19, MetOp-A and –B are being reprocessed at NESDIS with the Advanced Clear-Sky Processor for Oceans (ACSPO-RAN) • The objective is to create multiyear L2P SST data sets which would meet the following requirements: • Minimum regional SST biases • Close reproduction of true SST variations • Temporal stability • Cross-platform consistency • This presentation discusses the selection of SST algorithms for ACSPO-RAN Reprocessing AVHRR GAC

  3. Versions of ACSPO-RAN • Version 1 (RAN1): • AVHRR data from 2002-2013 reprocessed with ACSPO v. 2.20 (heritage regression SST algorithm + ACSPO Clear-Sky Mask, ACSM, Petrenko et al., 2010) • Clear-sky AVHRR BTs were matched-up with QCed drifting/moored buoys’ from the in situ Quality monitor (iQuam; www.star.nesdis.noaa.gov/sod/sst/iquam/; Xu and Ignatov, 2014) • SST algorithms were evaluated using these multiyear time series of matchups. • Version 2 (RAN2): • Several selected SST algorithms will be implemented in ACSPO (v.2.31) RAN2 • The corresponding L2 SST products will be regenerated with all algorithms • Results evaluated in MICROS (www.star.nesdis.noaa.gov/sod/sst/micros/; Liang and Ignatov, 2011) and SQUAM (www.star.nesdis.noaa.gov/sod/sst/squam/; Dash et al., 2010) Reprocessing AVHRR GAC

  4. Explored SST algorithms • Regression • NOAA regressions were used in ACSPO (v2.20) RAN1 • In the class of “conventional regressions”, OSI-SAF equations minimize regional SST biases and balance precision and sensitivity (Petrenko et al., JGR, 2014) • In ACSPO v2.30 (NOAA operations) and 2.31 (RAN2), NOAA heritage SST equations were replaced with those developed at EUMETSAT OSI-SAF • Incremental Regression (IncR) • RTM-based algorithm, developed for GOES-R ABI (Petrenko et al., RSE, 2011) and shown to perform at least comparably with other RTM-based algorithms (Merchant et al., RSE, 2008, 2009; Le Borgne et al., RSE, 2011) • In IncR, SST increments (Retrieved minus L4 SST) are correlated against BT increments (observed minus simulated BTs), rather than absolute SSTs vs. BTs Reprocessing AVHRR GAC

  5. Regression Equations (OSI-SAF) Day: TS = a0+ (a 1 + a 2 S θ) T11 + [a 3 + a 4 TS0 + a 5 S θ] (T11- T 12) + a6Sθ Night: TS = b0 +( b 1 + b 2 S θ) T3.7 + (b 3 + b 4S θ) (T11- T 12) + b5Sθ T11,T 12, T3.7 observed BTs S θ=1/cos(θ)θis satellite view zenith angle TS0 first guess SST (in °C) a’s and b’s regression coefficients • OSI-SAF equations explicitly include angular dependencies in all regression coefficients Reprocessing AVHRR GAC

  6. Incremental Regression (IncR) • Day: TS = TS0 + c0 + α[c1ΔT11 + c2(ΔT11 - ΔT12)TS0]+ + c3Sθ+ c4θ + c5W + c6W2 + c7φ + c8φ2 • Night: TS = TS0+ d0+ α[d1 ΔT3.7 + d2ΔT12]+ + d3Sθ+ d4θ + d5W + d6φ + d7φ2 ΔT11,ΔT12, ΔT3.7 BT increments (OBS – CRTM) WTotal water vapor content in the atmosphere φLatitude c’s and d’s“Least squares” regression coefficients α>1Scaling factor • IncR coefficients are derived from matchups of in situ SST increments with BT increments • BT increments are small (~0.5 – 1 K), and the least-squares method underestimates coefficients. Scaling is needed • The (red) terms independent of BTs are introduced to correct for SST biases caused by BT biases Reprocessing AVHRR GAC

  7. Scaling Coefficients in IncR • Sensitivity μ of IncR SST to true SST is: • Day: μ=α[c1D11 + c2(D11 - D12)TS0], • Night: μ=α[d1D3.7 + d2D12] • D3.7 , D11, D12 are derivatives of BT wrt SST (from CRTM) • The histograms of μare shown for 2005 NOAA-17 matchups • The IncR with the least-squares coefficients (with α=1)has unacceptably low sensitivity • Scaling brings the mean IncR sensitivity closer to 1 (and on average, higher than for regression) DAY NIGHT Reprocessing AVHRR GAC

  8. IncR coefficients with and without bias correction (Day) TS = TS0 + c0 + c1ΔT11 + c2(ΔT11 - ΔT12)TS0+ c3S θ+ c4θ +c5W+c6W2+ c7φ + c8φ2 • S θ, θ, W and φ arestatistically independent of BT increments • Inclusion of the bias correction terms only insignificantly changes coefficients in front of the BT increments Reprocessing AVHRR GAC

  9. The effect of bias correction terms in IncR (Day) TS = TS0 + c0 + c1ΔT11 + c2(ΔT11 - ΔT12)TS0+ c3S θ+ c4θ +c5W+c6W2+ c7φ + c8φ2 BIAS STD Sensitivity VZA • Statistics are shown for 2005 NOAA-17 matchups • IncR without bias correction produces unacceptably large biases • IncR with bias correction minimizes SST biases and SDs, but does not affect the sensitivity TPW LAT 5/6/2014 5/6/2014 Reprocessing AVHRR GAC Reprocessing AVHRR GAC Reprocessing AVHRR GAC 9 9

  10. Using variable coefficients • In the NOAA operations, regression coefficients are calculated once and used for the life of the mission (or long period of time) • Global biases in operational SSTs varydue to calibration trends in AVHRR BTs, and seasonal pattern of SST domain • In the reprocessing, recalculation of regression and IncR coefficients was tested on a daily basis using matchups within 3 months running window (current day ±45 days). • Here, we compare three SST algorithms: • Conventional regression with constant coefficients (RCC) - benchmark • Regression with variable coefficients (RVC) – improvement over the RCC • Incremental regression with variable coefficients (IncR) – improvement over the RVC Reprocessing AVHRR GAC

  11. NOAA-17 (DAY): Time series of global mean bias, SD and mean sensitivity Bias wrt in situ SST Bias wrt CMC Mean sensitivity CMC L4 SST (produced by the Canadian Met Centre) SD wrt in situ SST SD wrt CMC • RVC and IncR biases wrt in situ SST reduced from ~0.2 K to < 0.05 K • SD wrt in situ SST is smaller for IncR SST than for RCC and RVC • SD wrt CMC is greater for IncR than for regression algorithms • Mean sensitivity for IncR is ~.95, and <0.9 for regression algorithms Reprocessing AVHRR GAC

  12. NOAA-17 (NIGHT): Time series of global mean bias, SD and mean sensitivity Bias wrt in situ SST Bias wrt CMC Mean sensitivity SD wrt CMC SD wrt in situ SST • Same features as for day, but with lesser difference between RVC and IncR • For night, RVC and IncR statistics are very close • Mean sensitivity is ~ 1 for IncR and ~0.98 for RCC and RVC Reprocessing AVHRR GAC

  13. NOAA-17 (DAY): Annual mean SST biases as functions of VZA, TPW and LAT RCC RVC IncR VZA • Dependencies are shown for 7 full years of NOAA-17 operation • RCC biases are most variable and non-uniform • RVC biases are less variable but still non-uniform • IncR biases are most uniform and least variable TPW LAT Reprocessing AVHRR GAC

  14. NOAA-17 (NIGHT): Annual mean SST biases as functions of VZA, TPW and LAT RCC RVC IncR VZA • RCC biases are non-uniform and most variable • RVC biases are less variable but still non-uniform • IncR biases are most uniform and least variable • The nighttime difference between RVC and IncR is smaller than for day TPW LAT Reprocessing AVHRR GAC

  15. NOAA-17, -18 and MetOp-A: Multiyear mean biases as functions of VZA, TPW and LAT (Day) RCC RVC IncR • The biases are averaged over the following years: 2003-2009 for NOAA-17 2006-2013 for NOAA-18 2007-2013 for MetOp-A • RCC biases are most variable between the satellites • RVC biases are less variable between the satellites but relatively non-uniform • IncR biases are most uniform, for all three satellites VZA TPW LAT Reprocessing AVHRR GAC

  16. NOAA-17, -18 and MetOp-A: Multiyear mean biases as functions of VZA, TPW and LAT (Night) RCC RVC IncR • The biases are averaged over the following years: 2003-2009 for NOAA-17 2006-2013 for NOAA-18 2007-2013 for MetOp-A • RCC biases are most variable between the satellites • RVC biases are less variable between the satellites but relatively non-uniform • IncR biases are most uniform, for all three satellites VZA TPW LAT Reprocessing AVHRR GAC

  17. NOAA-17: Median annual regional biases for 2005 BA=median(| TS - Tsinsitu|), within 10⁰× 10⁰ lat/lon box RCC, DAY RCC, NIGHT RVC, DAY RVC, NIGHT IncR, DAY IncR, NIGHT • The IncR biases appear to be most uniform, both for day and night • Saharan dust is the problem for all algorithms Reprocessing AVHRR GAC

  18. NOAA-17, NOAA-18, MetOp-A: Multiyear median regional biases (Day) BM=median(BA), within each 10⁰× 10⁰ lat/lon box • Regional biases are least uniform for RCC and most uniform for IncR Reprocessing AVHRR GAC

  19. NOAA-17, NOAA-18, MetOp-A: Multiyear median regional biases (Night) BIA=median(BIA), within each 10⁰× 10⁰ lat/lon box • Regional biases are least uniform for RCC and most uniform for IncR • The difference between RVC and IncR biases is smaller than for day Reprocessing AVHRR GAC

  20. Spatial variability of interannual biases V1=median(|BIA-median(BIA)|), BIAis a set of interannual biases within all 10⁰× 10⁰ lat/lon boxes • For day, variability of regional biases is highest for RCC and the lowest for IncR • For night, variability of regional biases for RVC and IncR is close. Reprocessing AVHRR GAC

  21. Multiyear median variability of annual biases V2=median(VIA), VIA=median(|BA-BIA|) VIAis multiyear variability of annual biases in each 10⁰× 10⁰ lat/lon box • Regional biases are most variable for RCC and least variable for IncR, both for day and night Reprocessing AVHRR GAC

  22. Summary and Future Work • The Regression and Incremental Regression SST algorithms were explored using NOAA-17, -18 and MetOp-A matchups collected during ACSPO-RAN1 • Recalculation of coefficients on a daily basis reduces variations in global and regional SST biases for both types of algorithms • The IncR further improves dependencies of SST biases as functions of VZA, TPW and LAT, and reduces regional SST biases • At the second stage of reprocessing (RAN2), the multiyear L2 datasets will be produced with RCC, RC and IncR and evaluated in SQUAM and MICROS • The set of metrics used to evaluate SST products will be further elaborated. In particular, the assessment methodology proposed in GHRSST Climate Data Assessment Framework (Merchant et al., 2013) will be explored • We also plan to explore possible improvements in the AVHRR calibration algorithms. This will further improve temporal and regional stability of SST biases, especially, for the unstable AVHRR sensors onboard NOAA-15, -16 (since mid-2004) and NOAA-18 (since mid-2011). Reprocessing AVHRR GAC

More Related