The Climate Prediction CenterRainfall Estimation Algorithm .ppt

1、The Climate Prediction Center Rainfall Estimation Algorithm Version 2 Tim Love - RSIS/CPC,Presentation Outline,Overview Input data / methodology Satellite estimate combination process Merging steps Output data System requirements,CPC RFE 2.0,RFE 2.0 Overview,Run daily at CPC for Africa, southern Asi

2、a, Afghanistan area domains Final output is minimally biased and greatly improves spatial resolution of information Inputs include satellite IR temperature data, microwave precip estimates, gauge fields Computing resources required are relatively minimal Code highly portable,CPC RFE 2.0,Input Data,M

3、eteosat files Half hourly 0.05 infrared temperature data thru a McIDAS server Files are ftpd to host machine once daily and gridded based on current satellite position constants Code conducts QC via lag and cross-correlation methods Fractional coverage for 235K and 275K determined,CPC RFE 2.0,Meteos

4、at Data, cont.,Resultant field = cold count duration (CCD) 0.1 resolution CCD used for GOES Precipitation Index (GPI) calculation,CPC RFE 2.0,GPI tends to overestimate spatial distribution but underestimates convective precipitation,GPI Quality Control,Each pixel must have 4 half hour values, or pix

5、el is undefined 70% of all pixels must be defined after incorporating all half hour data sets,CPC RFE 2.0,GPI Estimate,CPC RFE 2.0,GTS Data,2534 stations available daily Only 400-800 report daily Few reports from Nigeria, none from Liberia, Sierra Leone Data ingested from GTS line, QCd, fed to opera

6、tional machine, then gridded to 0.1 resolution file Other station data may be readily used as input to algorithm via changing 2 tables in base code Requirements for RFE processing: GPI and GTS inputs,CPC RFE 2.0,GTS Quality Control,Must have 200 stations available daily Station undefined if GTS dail

7、y rainfall: 200 mm 1 mm and fc275 = 0 in all surrounding pixels2 mm 50 mm and all satellites 20 mm, and if sat-GTS 20 20 mm and all satellites 1 mm,CPC RFE 2.0,GTS Interpolation Technique,Shepard technique Using an initial search radius (rs0), a new radius is determined depending on number of statio

8、ns within rs0 If an adequate # of gauges is within new radius, interpolate rainfall to 0.1 grid using station-station vector Otherwise, interpolate using least squares regression If rainfall is undef or 0 within a 1.0 degree box, rainfall at center grid is zero,CPC RFE 2.0,Initial Search Radius,CPC

9、RFE 2.0,GTS Inputs,CPC RFE 2.0,GTS vs GPI,CPC RFE 2.0,SSM/I Inputs,2 instruments estimate precip twice daily6 hourly data frequency Fails to catch other rainfall in temporal gaps Data needs only small conversion in preparation for input to algorithm,CPC RFE 2.0,SSM/I Quality Control, 70% of pixels m

10、ust be defined after combining each input data set SSM/I daily rainfall is zero if: fc275 = 0 (no clouds) SSM/I rain 5 mm target grid is over the coast and 1 or less neighboring grids have SSM/I rain = 0,CPC RFE 2.0,SSM/I Estimate,CPC RFE 2.0,SSM/I vs GTS vs GPI,CPC RFE 2.0,As with SSM/I, data is av

11、ailable 4 times daily, staggered temporally Tends to overestimate most precip, but does well with highly convective systems Data sent in HDF format, thus needs to be deciphered before input to RFE algorithm Preprocessing straightforward,AMSU-B Data,CPC RFE 2.0,AMSU-B Quality Control, 60% of pixels m

12、ust be defined after incorporating all input data AMSU-B daily rainfall is zero if: fc275 = 0 (no clouds) AMSU rain 5 mm target grid is over the coast and 1 or less neighboring grids have AMSU rain = 0,CPC RFE 2.0,AMSU-B Estimate,CPC RFE 2.0,CPC RFE 2.0,Combining Satellite Estimates,Combines 3 satel

13、lite data sets linearly,where Wi = weighting coefficientsSi = precip estimatesi = random error,CPC RFE 2.0,Bias Removal,Satellite estimates are merged with station data to remove bias,where S = first step outputG = gauge observationsP = final output,CPC RFE 2.0,Combining Satellite Estimates,Combined

14、 analysis is a linear combination of each satellite estimate Satellite rainfall estimates are weighted by 1 / error variance Output dataset is then input to merging algorithm Estimates combined for all 6 resolutions, all satellite inputs Combined output =,CPC RFE 2.0,Calculating Error Variance,First

15、 guess at precipitation computed from mean of all inputs Satellite estimates compared to GTS data Areas without GTS data employ satellite estimate interpolation Proportional Constant calculated for every KSTP grids, to ease computation time Bi-linear interpolation used for remaining grids,CPC RFE 2.

16、0,Output Data,Operational: GTS+GPI+SSM/I+AMSUB Other: GTS+GPI GTS+GPI+SSM/I+AMSUB+GDAS With and without bias removal Archival: All inputs needed for reprocessing Some mid-processing outputs,CPC RFE 2.0,System Requirements,Linux or Unix operating system System has also been ported to Windows Minimum 2Gb hard drive space Minimum 500MHz processor Fortran 77/90 compiler C, Korn, or Bourne Shell GrADS software to display/create graphics,CPC RFE 2.0,System Outreach,Seek collaboration with external users to: Develop local capability Develop independent validation Improve algorithm,CPC RFE 2.0,