High Resolution Daily Sea Surface Temperature Analysis Errors - PowerPoint PPT Presentation

Loading...

PPT – High Resolution Daily Sea Surface Temperature Analysis Errors PowerPoint presentation | free to download - id: 682be0-ZTE4Y



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

High Resolution Daily Sea Surface Temperature Analysis Errors

Description:

High Resolution Daily Sea Surface Temperature Analysis Errors Richard W. Reynolds (NOAA, CICS) Dudley B. Chelton (Oregon State University) – PowerPoint PPT presentation

Number of Views:4
Avg rating:3.0/5.0
Date added: 4 December 2019
Slides: 33
Provided by: RichardW76
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: High Resolution Daily Sea Surface Temperature Analysis Errors


1
High Resolution Daily Sea Surface Temperature
Analysis Errors
Richard W. Reynolds (NOAA, CICS) Dudley B.
Chelton (Oregon State University)
2
GHRSST High Resolution
Reynolds Chelton
GHRSST
  • SST Analysis Must be a balance between Signal
    and Noise

3
Background
  • GHRSST (The Group for High Resolution SST)
    includes many high resolution SST analyses
  • There are differences in input data, grid
    resolution, analysis procedures
  • There are important differences in analyzed SSTs
    and analysis resolution
  • Reynolds and Chelton compared 6 SST analyses for
    2006-08 to try to identify analysis problems and
    determine whether any of the analyses are superior

4
SST Analyses, 1 January 2007
  • RSS OI
  • (1/11) grid
  • NCEP RTG-HR
  • (1/12) grid
  • UK OSTIA
  • (1/20) grid
  • NCDC Daily OI (AMSR AVHRR)
  • (1/4) grid
  • Spatial scales differ
  • Differences can exceed 5C off coast

5
Random Sampling Error
  • OI Derived
  • Top panel AVHRR-only
  • Bottom panel AMSRAVHRR
  • Sampling errors are larger than random errors
  • Sampling and random error reduced by observations
  • Lower for AMSRAVHRR

6
SST Analyses, 1 January 2007
  • RSS OI
  • (1/11) grid
  • NCEP RTG-HR
  • (1/12) grid
  • UK OSTIA
  • (1/20) grid
  • NCDC Daily OI (AMSR AVHRR)
  • (1/4) grid
  • This is a daily average
  • What spatial scales are justified?

7
Results
  • There is no clear correlation between resolution
    and spatial grid size
  • If the analysis resolution is pushed beyond the
    spatial and temporal resolution of the data the
    apparent SST signal is simply just noise
  • Is there a better way to do analyses?

8
Two Stage Analysis
  1. Low Resolution (25 km) analysis using microwave
    and infrared satellite data plus in situ data
  2. High Resolution (4.4 km) analysis using infrared
    satellite data only

9
Why Two Stages?
  • Both microwave (MW) and infrared (IR) satellite
    data are now available
  • Microwave has better coverage than infrared
  • Infrared has higher resolution than microwave
  • 2-Stage Analysis allows processing to take
    advantage of both types of data

10
Analysis Differences High Low
  • Top panel No filter of Pathfinder AVHRR
  • Note bulls eyes especially along 145E
  • Bottom panel Median filter of Pathfinder AVHRR
  • Data extremes tossed by eliminating data where
    the median observation gt 0.8C
  • Fewer bulls eyes

11
RMS monthly differences High Low
  • Top panel January 2003
  • Bottom panel July 2003
  • Note regions with little difference (no AVHRR
    hi-res signal)
  • Gulf Stream in January
  • Off Peru/Columbia coast in July

12
SST Gradient Magnitudes Castelao, et al. (2006),
hourly averages 2001-4
130W 125W
120W 115W 130W
125W 120W 115W
13
11 Sep. 03 SST Gradient (C/100 km)
  • Top panel Hi-Res
  • Middle panel Low-Res
  • Bottom panel OI Hi-Res Normalized Error
  • Norm. Error 1.0 if no Hi-Res Data
  • Norm. Error lt 0.8 with Hi-Res Data

14
11 Sep. 03 SST Gradient (C/100 km)
  • Top panel Hi-Res
  • Middle panel Low-Res
  • Bottom panel Difference (Hi-Res Low-Res)
  • With Hi-Res Normalized Error 0.8 Contour
  • Indicator of data edge e.g., cloud cover, swath
    edge
  • Note correlation between contour and high
    gradients

15
Input Data
  • Pathfinder AVHRR Data with respect to AMSRAVHRR
  • Data for 10 Sept. 2003 analysis shown
  • Data for 11 Sept. 2003 Analysis uses 10-12 Sept.
    2003 data
  • no data for 12 Sept.
  • The only useful field for 10 and 11 Sept.
    analysis is AVHRR day data for 10 Sept. 2003

16
Day Pathfinder 10 Sept. 2003
  • Top Median filtered observations
  • Bottom Additional gradient filtering using
    climatology if missing box in 3x3
  • Large 1C data to AMSRAVHRR difference near data
    edge
  • Gradient filtering reduces but cant eliminate
    it
  • Gradient filtering does reduce other gradients at
    date edge

17
Analysis 11 Sept. 2003
  • Top Median filtered analysis
  • Bottom Median gradient filtering near data
    edge 3x3 box with at least 1 missing ob
  • Some improvement
  • Gradients are unrealistic showing where analysis
    resolution changes
  • May need to reduce analysis resolution at data
    edge

3x3 edge with large gradient
18
(No Transcript)
19
Why are there Data Edge Biases?
  • Retrievals are from NOAA-17, a morning satellite
    bias is unlikely to be a diurnal signal also
    same positive bias seen at night on another
    several other days
  • Median and gradient filtering not adequate
  • Are the biases due to cloud detection problems?
  • How should these biases be corrected?
  • Can the Pathfinder team help?

20
Low-Res Analysis 11 Sept. 2003
  • Top AVHRR-only
  • Middle AMSRAVHRR
  • Bottom Difference AVHRR-only warmer than
    AMSRAVHRR by more than 1C
  • AMSR dominates AVHRR due to better sampling

21
Filtering for High-Res Data
  • Large-scale bias correction (15 day, 1000 km)
    to correct satellite data with respect to in situ
    data
  • Median Filtering To eliminate 1 pixel extremes
  • Gradient Filtering To eliminate errors near data
    (or cloud) edges
  • ADD ?? Small-scale bias correction (3 day, 100
    km) to correct AVHRR data with respect to AMSR
    data

22
Is it Real or is it Memorex?
  • Advertisement from the 1990s implying that a live
    and a Memorex taped performance were the same

23
Is it SST signal or SST noise?
24
Suggestions for Improving SST Analyses - 1
  • Intercompare the input data
  • Look for and reduce isolated extremes
  • Consider median filtering
  • Be careful at boundaries between regions with and
    without data
  • Computing an analysis is a bit like making
    sausage
  • The input impacts the output

25
Suggestions for Improving SST Analyses - 2
  • Compute several versions
  • Do intercomparsions among these versions and with
    other analysis products to uncover problems
  • Compute gradients
  • Look for large gradients at the boundaries
    between regions with and without data
  • Share results with others

26
Goldilocks the 3 Bears
  • Can users find the SST analysis that is just
    right for them??

27
One page summary-1 Truth in Advertising
  • SST Stats Grid resolution, data used, temporal
    resolution, and period of analysis, provider, ..
  • This is easily available
  • We need to also provide more difficult
    information

28
One page summary-2 Truth in Advertising
  • SST Analysis Resolution Explain that the
    resolution changes in time and space with data
    changes
  • Example For 2 stage analysis
  • With AVHRR data 10 km
  • Without AVHRR 100 km
  • Flag regions with/without AVHRR data

29
One page summary-3 Truth in Advertising
  • SST Bias Correction Explain why corrections
    needed
  • For 2 stage analysis
  • Ship SSTs reduced by 0.14C to better match buoys
  • Satellite data adjusted with respect to buoy and
    corrected ship data 15 day, 1000 km
  • AVHRR data adjusted with respect to AMSR data 3
    day, 100 km ??

30
One page summary-4 Truth in Advertising
  • Error Estimates Verification needed by analysis
    and by data intercomparions
  • For 2 stage analysis
  • Bias, random sampling error estimated
  • Relative values useful
  • Absolute values may be too small especially near
    the coast

31
One page summary-5 Truth in Advertising
  • Additional Data Processing
  • For 2 stage analysis
  • Satellite data median filtered to eliminate grid
    scale bulls eyes
  • Satellite data near data edges eliminated when
    gradients large
  • Other information
  • For 2 stage analysis
  • Two stages allow useful separation of IR and MW
    resolution and sampling

32
One page summary-6 Truth in Advertising
  • Three most important things about documentation
    and users
  • Users dont know that there is any
  • If they know that, they dont read it
  • If they read it, they dont understand it
  • That is why a one page summary would be useful
About PowerShow.com