presentation reynolds

Sea Surface Temperature (SST) In Situ Network for Climate: 

Sea Surface Temperature (SST) In Situ Network for Climate Richard W. Reynolds Huai-Min Zhang Thomas M. Smith National Climatic Data Center Asheville, NC

SST Analysis Errors: 

SST Analysis Errors There are three types of analysis errors: Sampling, Random, Bias If the random observational errors are known, analysis sampling and random errors can easily be computed Optimum Averaging results show that random plus sampling errors on monthly 5o scales are small: < 0.3oC This is due to the high density of satellite observations

SST Data and Analysis for 1 Week14-20 December 1997: El Niño: 

SST Data and Analysis for 1 Week 14-20 December 1997: El Niño oC

Bias Errors: 

Bias Errors Biases occur with all satellite data due to instrument and algorithm problems Typical bias: 0.2 to 0.5oC Worst case bias: 2 to 3oC In situ (ship & buoy) data are needed to correct any satellite bias There is no convenient algorithm to compute bias We don't know when biases will occur GOAL: Assume required SST analysis bias accuracy is 0.2-0.5oC monthly on 5o spatial grid Needler, et al. 1999, OceanObs’99

AVHRR Nighttime Biases: 

AVHRR Nighttime Biases Volcanoes El Chichón: Mar '82 - Sep '83 Mt. Pinatubo: Jun '91 - Mar '92 NOAA Satellite Periods NOAA-7: Nov '81 - Feb '85 NOAA-9: Feb '85 - Nov '88 NOAA-11: Nov '88 - Dec '94 NOAA-12: Dec '94 - Apr '95 NOAA-14:Apr '95 - Feb '01 NOAA-16: Feb '01- ……… Zonal Difference: OI.v2-Night Satellite NOAA-7 NOAA-9 NOAA-11 NOAA-16 NOAA-14 NOAA-11

Simulation of Bias Errors: 

Simulation of Bias Errors OI analysis used with bias correction For Jan 1990 to Dec 2002 Climatology is first guess (FG) Spatial satellite bias EOF patterns computed Bias is OI with minus OI without bias correction Define Gaussian Noise Functions, a(t) & b(t), with mean of 0 and variance of 1 Satellite SSTs are simulated at actual data locations Satellite SSTs = FG(t) + Bias(t) where Bias(t) = EOF(i) * a(t), i is the EOF (1-6) EOF absolute maximum scaled to 2oC

Simulation of Bias Errors-2: 

Simulation of Bias Errors-2 Buoy data are simulated on different grids Buoy SSTs = FG(t) + 0.5oC * b(t), where the buoy random observational error is 0.5oC Compute RMS Differences between the simulated OI and First Guess over time If there were no buoy data, the RMS residual would be equal to the absolute value of the EOF If there were complete buoy and/or ship sampling, the RMS would be 0

Results: 

Results A Potential Satellite SST Bias Error as a function of buoy density Potential is used because if satellite data have no biases, no buoy data are needed By definition the satellite biases are scaled so that the potential bias error without buoys is 2oC, a worst case bias error

Potential SST Satellite Bias Error: 

Number of Buoys on 10o grid Goal: 0.5oC Potential SST Satellite Bias Error Goal Exceeded by 2 Buoys on 10o Grid SST Error (Co) 2 0.2oC Satellite Bias Error Set to 2oC

Buoy Equivalent Density: 

Buoy Equivalent Density "Buoy Equivalent" defined by: Number of Ships/7 + Number of Buoys Because ships are nosier than buoys, 7 ships equals 1 buoy

Potential SST Satellite Bias Error: 

Potential SST Satellite Bias Error Goal: 0.5oC To reach density of 2 per 10o grid need additional 60oS - 60oN 189 buoys 60oS - 20oS 102 buoys 20oS - 20oN 65 buoys 20oN - 60oN 22 buoys No Buoys

Conclusions: 

Conclusions Satellite data greatly reduces SST analysis sampling and random errors over analyses using ship and buoy data alone Bias errors were simulated with biased satellite data & unbiased buoy data Results show that an equivalent buoy density of 2 buoys on a 10o grid is required to reduce the potential satellite bias from 2oC to below 0.5oC An equivalent buoy is 7 ships or 1 buoy Present ship and buoy data distribution is not adequate Roughly 200 additional buoys are needed