Snow Analysis for Numerical Weather prediction at ECMWF Patricia de Rosnay, Gianpaolo Balsamo, Lars Isaksen European Centre for Medium-Range Weather Forecasts IGARSS 2011, Vancouver, 25-29 July 2011 Slide 1 ECMWF
Land surface data assimilation 1999 2004 2010/2011 OI screen level analysis Douville et al. (2000) Mahfouf et al. (2000) Soil moisture 1D OI analysis based on Temperature and relative humidity analysis Revised snow analysis Drusch et al. (2004) Cressman snow depth analysis using SYNOP data improved by using NOAA / NSEDIS Snow cover extend data (24km) Structure Surface Analysis Optimum Interpolation (OI) snow analysis Pre-processing NESDIS data High resolution NESDIS data (4km) SEKF Soil Moisture analysis Simplified Extended Kalman Filter Drusch et al. GRL (2009) De Rosnay et al. ECMWF NewsLett. (2011) SYNOP Data NOAA/NESDIS IMS METOP-ASCAT SMOS IGARSS 2011, Vancouver, 25-29 July 2011 Slide 2 ECMWF
Snow Analysis Snow Quantities: - Snow depth SD (m) - Snow water equivalent SWE (m) ie mass per m 2 - Snow Density ρ s, between 100 and 400 kg/m3 SWE SD ρ S 1000 = [m] Background variable used in the snow analysis: - Snow depth S b computed from forecast SWE and density model parameterization revised in 2009 (Dutra et al., J Hydromet. 2009) Observation types: - Conventional data: SYNOP snow depth (S O ) - Satellite: Snow cover extent (NOAA/NESDIS) IGARSS 2011, Vancouver, 25-29 July 2011 Slide 3 ECMWF
NOAA/NESDIS Snow extent data Interactive Multisensor Snow and Ice Mapping System - Time sequenced imagery from geostationary satellites - AVHRR, - SSM/I - Station data Northern Hemisphere product - Daily - Polar stereographic projection Resolution - 24 km product (1024 1024) - 4 km product (6044 x 6044) Information content: Snow/Snow free Format: - 24km product in Grib - 4 km product in Ascii More information at: http://nsidc.org/data/g02156.html IGARSS 2011, Vancouver, 25-29 July 2011 Slide 4 ECMWF
NOAA/NESDIS Pre-Processing - Pre-processing revised in 2010. - NOAA/NESDIS data received daily at 23UTC. - Pre-processing: - Conversion to BUFR - BUFR content: LSM, NESDIS snow extent (snow or snow free), and orography, interpolated from the model orograpghy on the NESDIS data points. Orography (m) included in the BUFR used in the snow analysis IGARSS 2011, Vancouver, 25-29 July 2011 Slide 5 ECMWF
Snow Analysis SYNOP Pre-Processing: - SYNOP reports converted into BUFR files. - BUFR data is put into the ODB (Observation Data Base) Snow depth analysis in two steps: 1- NESDIS data (12UTC only): - First Guess snow depth compared to NESDIS snow cover NOAA snow & First Guess snow free put 0.1m snow in First Guess (First Guess snow free: < 0.01m of snow, ie SWE in [1; 4] mm; Update: SD 0.1m, snow density=100kg/m3, SWE=0.01m) - NESDIS snow free used as a SYNOP snow free data 2- Snow depth analysis (00, 06, 12, 18 UTC): - Cressman interpolation: 1987-2010 Still used in ERA-Interim - Optimum Interpolation: Used in Operations since November 2010 IGARSS 2011, Vancouver, 25-29 July 2011 Slide 6 ECMWF
Cressman Interpolation S a N w n b n= 1 = S + N ( ) O b S S - (Cressman 1959) - S O snow depth from synop reports, - S b background field estimated from the short-range forecast of snow water equivalent, - S b background field at observation location, and - w n weight function, which is a function of horizontal r and vertical difference h (model obs): w = H(r) v(h) with: n= 1 n w n ' H(r) 2 2 rmax r = max,0 2 2 rmax + r r max = 250 km (influence radius) v(h) = h 2 max h 2 max 1 if 0 < h Model above observing station + 2 h 2 h if h max < h < 0 0 if h < - h max h max = 300 m (model no more than 300m below obs) Obs point more than 300m higher than model IGARSS 2011, Vancouver, 25-29 July 2011 Slide 7 ECMWF
Some issues in the Cressman Snow analysis PacMan Snow Patterns where observations are scarce SNOW Depth and SYNOP data in cm (1) 20050220 at 12UTC 165 W 160 W 155 W 150 W 145 W 140 W 135 W 130 W 125 W 120 W 115 W 110 W 105 W 100 W 95 W 90 W 85 W 80 W 75 W 70 W 65 W 6 57 75 N 5 75 N 26 70 N 70 N 26 14 80 19 65 N 92 13 15 65 N 20 38 43 49 30 77 45 35 20 60 N 63 39 55 60 N 42 67 68 34 73 37 35 3 18 26 50 59 26 6 3 49 36 42 55 N 3 0 1 52 2 55 N 26 10 51 37 23 15 32 25 29 38 6 21 1910 2 39 6 3611 16 4 26 12 18 10 71 12 65 127 15 15 4 5 12 2 24 7 24 9 71 46 45 27 50 N 2 1 3627 8 15 40 23 71 3 9 15 51 72 60 34 50 N 6 4569 40 46 19 44 5 28 34 2 55 3 53 25 20 6 18 44 35 79 109 28 55 14 7 9 14 178 3 11 42 2521 14 9 17 17 3 6 45 N 10 18 7 512 2 1 2 18 15 27 45 N 5 3 5 2 7 4 0 2 10 10 3 3 3 165 W 160 W 155 W 150 W 145 W 140 W 135 W 130 W 125 W 120 W 115 W 110 W 105 W 100 W 95 W 90 W 85 W 80 W 75 W 70 W 65 W 10001 4000 150 100 50 20 15 10 5 2 1 North America 2005 SNOW Depth and SYNOP data in cm (1) 20070212 at 12UTC 85 E 90 E 95 E 100 E 105 E 110 E 115 E 120 E 125 E 130 E 135 E 140 E 145 E 150 E 155 E 75 N 75 N 70 N 70 N 65 N 65 N 60 N 60 N 55 N 85 E 90 E 95 E 100 E 105 E 110 E 115 E 120 E 125 E 130 E 135 E 140 E 145 E 150 E 155 E 55 N 10001 4000 150 100 50 20 15 10 5 2 1 Siberia 2007
Snow Patterns, PacMan issues example on 23 Feb 2010 85 E 75 N 75 N 70 N 70 N 65 N 65 N 60 N 60 N 55 N 55 N 85 E 90 E SNOW Depth and SYNOP data in cm (1) 20100223 at 12UTC 85 E 90 E 95 E 90 E 95 E 95 E 100 E Deterministic Analysis, T1279 (16km) (cm) SNOW Depth and SYNOP Integrated data Forecasting cm (1) 20100223 System IFS at 12UTC cycle 36r1 100 E 105 E 100 E 105 E 105 E 110 E 110 E 115 E 110 E 115 E 115 E 120 E 120 E 125 E 120 E 125 E 125 E 130 E 130 E 135 E 130 E 135 E 135 E 140 E 140 E 145 E 140 E 145 E 145 E 150 E 150 E 155 E ERA-Interim re-analysis, T255 (80km), IFS cycle 31r1 150 E 155 E 75 N 75 N 70 N 70 N 65 N 65 N 60 N 60 N 55 N 85 E 90 E 95 E 100 E 105 E 110 E 115 E 120 E 125 E 130 E 135 E 140 E 145 E 150 E 155 E 155 E 55 N 10001 4000 150 100 50 20 15 10 5 2 1 10001 4000 150 100 50 20 15 10 5 2 1
Revised snow analysis Winter 2009-2010 highlighted several shortcomings of the snow analysis related to the Cressman analysis scheme as well as to a lack of satellite data in coastal areas, as well as issues in the NESDIS product pre-processing at ECMWF, fixed in flight in operations on 24 Feb 2010. Revised snow analysis from Nov. 2010: (from Integrated Forecasting System, IFS cycle 36r4) OI: New Optimum Interpolation Snow analysis, using weighting functions of Brasnett, J. Appl. Meteo. (1999). The OI makes a better use of the model background than Cressman. NESDIS: NOAA/NESDIS 4km ASCII snow cover product (substituting the 24 km GRIB product) implemented with fixes in geometry calculation. The new NESDIS product is of better quality with better coverage in coastal areas. QC: Introduction of blacklist file and rejection statistics. Also allows easier identification of stations related to MS queries.
Snow depth Optimum Interpolation 1. Observed Increments from the interpolated background S i are estimated at each observation location i. 2. Analysis increments S ja at each model grid point j are calculated from: S a j = N i= 1 w i S i 3. The optimum weights w i are given for each grid point j by: (B + O) w = b b : background error vector between model grid point j and observation i (dimension of N observations) b(i) = σ 2 b. X µ(i,,j) B : error covariance matrix of the background field (N N observations) B(i 1,i 2 ) = σ 2 b µ(i 1,i 2 ) with the horizontal correlation coefficients µ(i 1,i 2 ) and σ b = 3cm the standard deviation of background errors. 2 2 r i i ri i z 1 2 1 2 i1i2 µ ( i ) = + 1,i2 (1 )exp.exp (Brasnett, 1999) Lx Lx Lz O : covariance matrix of the observation error (N N observations): O = σ 2 o I with σ o = 4cm the standard deviation of obs. Errors Lz; vertical length scale: 800m, Lx: horizontal length scale: 55km Quality Control: if S i > Tol (σb 2 + σo 2 ) 1/2 ; Tol = 5 IGARSS 2011, Vancouver, 25-29 July 2011 Slide 11 ECMWF
In both cases: S a j = N i= 1 OI vs Cressman w i S i Cressman (1959): weights are function of horizontal and vertical distances. OI: The correlation coefficients of B and b follow a second-order autoregressive horizontal structure and a Gaussian for the vertical elevation differences. OI has longer tails than Cressman and considers more observations. Model/observation information optimally weighted by an error statistics.
NESDIS 4km product - Data thinning to 24 km -> same data quantity, improved quality - Data Orography interpolated from high res (T3999, ie 5km) IFS
Snow Depth Analysis Old: Cressman NESDIS 24 km New: OI NESDIS 4km
Snow Water Equivalent Analysis increments (18UTC) Accumulated for January 2010, in mm of water Accumulated January 2010 In mm of water Old: Cressman NESDIS 24 km New: OI NESDIS 4km
Snow Depth Analysis Cressman, NESDIS 24 km OI, NESDIS 4km Case study: 22 december 2009 major snow fall in Europe - Removes snow free patches and overestimated snow patches - Better agreement with SYNOP data and NESDIS data
New snow analysis - Snow Optimum Interpolation using Brasnett 1999 structure functions - A new IMS 4km snow cover product to replace the 24km product - Improved QC and monitoring possibilities Number of SYNOP data used in the Analysis in January 2010 Identified a lack of SYNOP Snow depth data in Sweden
Use of Additional Snow depth data Since December 2010, Sweden has been providing additional snow depth Data, Near Real Time (06 UTC) through the GTS Snow depth analysis using SYNOP data Snow depth analysis using SYNOP data + additional snow data Implemented as a new report type, in flight from 29 March 2011
Use of Additional Snow depth data 0.08 control normalised fi28 minus fi29 Root mean square error forecast N.hem Lat 20.0 to 90.0 Lon -180.0 to 180.0 Date: 20101221 00UTC to 20110107 00UTC 500hPa Geopotential 00UTC Confidence: 90% Population: 18 Snow Depth difference (cm) Due to using additional non-synop data in Sweden 0.06 0.04 0.02 Impact on 500hPa Geopotential 0-0.02-0.04-0.06-0.08 0 1 2 3 4 5 6 7 8 9 10 11 Forecast Day
Comparison against SYNOP data Old analysis (Cressman and NESDIS 24km) New analysis (OI and NESDIS 4km) RMSD between analysis and observations
Independent validation Sodankyla, Finland (67.368N, 26.633E) Winter 2010-2011 ECMWF deterministic analysis SYNOP snow depths FMI-ARC snow pit and ultrasonic depth gauge Figures produced by R. Essery, Univ Edinburgh)
New snow analysis Impact RMSE Forecast 1000hPa Geopotentail in Europe Improved when above 0 Impact of OI vs Cressman (both use NESDIS 24km) Overall Impact of OI NESDIS 4km vs Cressman NESDIS 24 km
New snow Analysis in Operations Old: Cressman NESDIS 24km Cressman +24km NESDIS New: OI NESDIS 4km FC impact (East Asia): OI Brasnett 1999 +4km NESDIS - OI has longer tails than Cressman and considers more observations. -- Model/observation information optimally weighted by an error statistics.
Summary and Future Plans New snow analysis implemented at ECMWF Based on a 2D Optimum Interpolation Uses Brasnett 1999 structure functions Uses SYNOP and high resolution snow cover data from NOAA/NESDIS, Flexible to use non-synop reports (new report codetype) Improved QC (blacklisting and monitoring possibilities). OI has longer tails than Cressman and considers more observations. Model/observation information optimally weighted with error statistics. Positive impact on NWP. However extensive validation using independent data needs to be done In the future, use of combined EKF/OI system for NESDIS/conventional information data assimilation.