SMOS NRT wind products Validation Plan
SMOS Wind Speed Product Quality Monitoring
The SMOS NRT surface wind product validation relies on 4 types of validation datasets including :
- The Step Frequency Microwave Radiometer (SFMR) flight data in Tropical cyclones,
- The anemometer measurements installed onboard buoy networks,
- The Satellite radiometer and scatterometer winds, and,
- The Numerical Weather Predictions model data.
SMOS NRT Wind product validation in Tropical and Extra-Tropical Cyclones using SFMR data
An important validation analysis of the SMOS NRT wind will be performed post season using aircraft-based in-situ wind speed data in Tropical Cyclones coming from the Step Frequency Microwave Radiometer (SFMR, RD.12).
After each hurricane season in the Atlantic and East Pacific (from July to October), we plan to collect all SFMR flight data and validate the SMOS NRT (as well as the reprocessed SMOS winds) using those datasets that will be properly co-localized in space and time between SMOS swath intercepts of TCs and the SFMR flights. The SMOS–SFMR paired database will be first constructed by selecting co-located data with time differences (Δt) between both acquisitions of less than±12 h to provide the largest number of pairs. A subset only including pairs with |Δt| < 6 h will be further used to establish a more robust relationship between SMOS and SFMR wind speeds if enough data are available. If the central time lag, Δt, between SMOS and SFMR data as the aircraft will fly over the eye region exceed ±0.5 h, the storm center displacement between the aircraft and satellite acquisitions can be significant. The storm center position and translation speed will be derived from the Automated Tropical Cyclone Forecast TC best track 6-hourly data, interpolated at the SMOS local acquisition time (in the swath location around the storm center). To adjust for storm motion when |Δt| >0.5 h, SFMR tracks will be spatially translated (without rotation) from the original storm center location detected in SFMR data to the storm center location evaluated at the SMOS time. The SMOS NRT winds at-25 km resolution will then be bi-linearly interpolated along the location of the translated SMFR track. The SMFR data will then be spatially averaged along the flight track with a running Gaussian spatial filtering window with half-width, ~22 km (which is about half the SMOS sensor average resolution of ~43 kms).
The SMOS NRT wind/SFMR colocated database shall be updated yearly by the new forthcomming acquisitions, starting the first year of NRT wind production.
For the reprocessed SMOS wind dataset, we will perform a full period re-analyses using the revised calibration SFMR database provided to us by the SFMR reprocessing project from NOAA (pers contact with Joe Sapp in charge of the project). Note that this database also includes winter storm flights in the North Atlantic and Alaska, which will extend the validation domain of the reprocessed winds with SFMR to include Extra-Tropical Cyclones, which aren’t made available rapidly for each season, contrairly to the Tropical Cyclone data.
At the end of each hurricane season (November each year) but also when exploiting the full history of the past and yearly updated colocalized database, the following validation metrics will be generated and will be included in the service review’s validation report.
For each co-localized SMOS/SFMR flight legs we will provide :
1) A Geographical plot showing the SFMR flight track (in black in the figure below), superimposed with the SMOS SWS swath intercept, following :
Figure 1 : Map showing an SFMR flight track in 2012 Atlantic hurricane LESLIE (black thick lines) co-localized and superimposed on SMOS retrieved SWS swath data (color in knots). The ATCF storm track is shown in light blue.
2) A plot of the time/space 1D series of the 1) SFMR Wind Speed and rain rate as measured during/along the flight track and at the original 3 km high resolution, 2) the SFMR wind speed averaged at the SMOS (~43 km) lower spatial resolution and 3) the SMOS NRT wind interpolated along the aircraft flight track :
Figure 2 : Corresponding time series of the SFMR retrieved wind speed in m/s (black curve) and rain rate in mm/h (gray curve) at nominal resolution along aircraft track (~3 km). The SFMR retrieved wind speed that has been spatially averaged with a running window of 43 km width along track (corresponding to the mean resolution of SMOS interferometer pixels) is shown in blue. The retrieved wind speed from SMOS is shown in red. The x-axis shows the time lag between SFMR acquisitions and SMOS ones.
3) When feasible, wind radii along track at 34, 50 and 64 kts derived from SMOS will be compared to the ones estimated from SFMR low resolution data and stored in a co-localized database of SMOS/SFMR wind radii for further statistics,
4) Histograms of the differences between the co-localized winds, i.e. and scatterplot of SWSSMOS versus
5) Evaluation of the Mean, Median, standard deviation, rms, interquartile range, kurtosis and skewness of the probability distribution functions of the
6) For all accumulated flight data in the SFMR/SMOS database, a cummulated analysis similar to the previous metrics for individual flights will then be generated for all flights and updated yearly to provide an overall metric of the quality of SMOS wind versus the SFMR ones. A specific classification of the SMOS wind quality expressed in terms of will be provided as follows :
- For all data of all year and all basins (Atlantic and Pacific) and all wind speed range,
- Year per Year for all basins, and separatly for each basin,
- As function of the central time lag, Δt, between SMOS and SFMR co-localized data,
- As function of the distance to coast,
- As function of the SMOS SWS across-track position
- As function of the co-localized low resolution Rain Rate from SMFR,
- And separatly, in the following wind speed ranges :
- Full wind speed range
- Low to intermediate winds (SWS<12 m/s),
- Below Tropical storm force (12<SWS < 17.( m/s),
- Above Tropical Storm Force (17.5 <SWS<32.5 m/s),
- Above Hurricane strength (SWS>32.5 m/s)
- For the validation of the SMOS SWS reprocessed data, statistics will be produced separatly for Tropical Cylones and Extra-Tropical cyclones.
- Statistics will be evaluated separatly for data with quality levels 0 to 2.
SMOS NRT wind validation using Buoy wind data
We will also use anemometers data obtained from the NOAA National Data Buoy Center (NDBC), the NOAA Pacific Marine Environmental Lab (PMEL), Météo-France and the Canadian Marine Environmental Data Service (MEDS). These data are collected in the frame of IFREMER past and present involvments into Globwave, CMEMS and CCI sea state project. They will be available for the project on IFREMER IT disks.
We will quality check, clean and process the data into a single format for use in radiometer validations. Anemometer heights for each buoy vary so winds will be converted to 10-meters height neutral wind using a logarithmic wind profile. We will use a collocation window of 22 km-half width and 60 minutes to reduce the spatial-temporal mismatch that exists between the time averaged single point buoy observations and the instantaneous spatially averaged SMOS satellite wind product.
The overall SMOS minus BUOY wind speed difference and standard deviation and geographical distribution maps will be estimated/generated to check that SMOS winds meets the quality standards expected for ocean winds. The scatter of collocations and Mean errors for individual buoys will be plotted.
Cross-talk plots will show the SMOS minus BUOY wind speed differences with respect to other geophysical or observational parameters (SSS variability, buoy wind speed, SST, scan position, asc/desc). Statistics will be evaluated separatly for SWS data with quality levels 0 to 2, distance to coasts, etc...
SMOS NRT wind validation plan using satellite wind data
A systematic monthly comparison of SMOS NRT winds and the following satellite radiometer and scatterometer wind speed products will be conducted.
Table 1 : : List of satellite wind products that wil be used to validate SMOS NRT wind
Type | Format | Provider | Data record name | Spatial Resolution Of Products | Period & status | Equator Crossing Time (Local time zone) |
Binary | REMSS | v8 | 0.25 deg grid | May 2012- present | 1:30 PM Ascending 1:30 AM Descending | |
Binary | REMSS | V7 | 0.25 deg grid | Dec2006-present | 6:30 PM Ascending 6 :30 AM Descending | |
Binary | REMSS |
| 0.25 deg grid | Jan 2003-present | 6:11 PM Ascending 6 :11 AM Descending | |
Binary | REMSS | V1 | 0.25 deg grid | Apr 2015-present | 6pm ascending /6am descending |
Only those particular sensors were selected because other wind measuring instrument such as ASCAT on board Metop-A/B/C do provide very little co-localized data with SMOS within a 1h temporal window and within distances less than 25 km. This is attributed to their differing orbits and local time of ascending nodes. Co-located pairs between SMOS NRT winds and each of the satellite wind products from individual satellite sensors as listed in Table 1 shall however be collected using a 25 km radius and ±60 min collocation window. The main characteristics of the match-up database between SMOS NRT wind and each sensor will be provided in the form of histograms describring the distributions at the match-up pairs of:
- the time difference ∆t
- the theoretical wind error,
- the SMOS SWS across-track distance,
- the Percentages of the SMOS QC-levels
- the distance to coasts, and,
- the latitudes
The geographical density of co-localized points in 2°x2° boxes determined between SMOS NRT and each satellite wind speeds will be mapped .
Statistics of the will be provided as function of
- the central time lag, Δt, between SMOS and the other satellite co-localized wind data,
- the distance to nearest coasts,
- the across-track position at which the SMOS SWS is retrieved,
- and separatly, in the following wind speed ranges :
- Full wind speed range
- Low to intermediate winds (SWS<12 m/s),
- Below Tropical storm force (12<SWS < 17 m/s),
- Above Tropical Storm Force (17.5 <SWS<32.5 m/s),
- Above Hurricane strength (SWS>32.5 m/s)
- In addition, to monitor the efficiency of quality flags we will characterize the SMOS observation minus the co-localized product departure statistics partitioned by each of the SWS quality levels.
The percentage of match-up co-localized points in 2°x2° boxes between SMOS NRT and all satellite winds (SMAP, SSM/I-F16,17,18, AMSR-2 and WindSat) speeds for which the wind speed difference between the colocalized data |∆SWS| exceed 2.5 m/s will be mapped to reveal specifc areas where the SMOS NRT wind quality, evaluated with respect all other satellite winds, is degraded.
In addition we shall analyze the statistics of the differences between SMOS NRT and all satellite match-ups within the following 11 Validation regions :
- Reg 1 : Global ocean
- Reg 2 : North Atlantic
- Reg 3 : South Atlantic
- Reg 4 : North-East Pacific
- Reg 5 : North-West Pacific
- Reg 6 : South Pacific
- Reg 7 : North Indian Ocean
- Reg 8 : South Indian Ocean
- Reg 9 : Arctic Ocean
- Reg 10 : Roaring forties & furies fifties
- Reg 11 : Near coasts global region
On the following table of plots, the mask of the region are provided.
|
Masks of the SMOS WIND NRT products validation region shown in magenta color
For each region, histogram plots and table as well as monthly time series of the statistical moment of the differences between SMOS and each (and all, i.e combined) surface wind products will be provided.
SMOS NRT wind validation using ECMWF
The ECMWF surface wind forecast products colocalized at SMOS acquisition time and location can finally be used as a rough glocal metric of validation. While ECMWF forecasts may be erroneous and particularly in the high wind speed regions of the TCs, SMOS point by point and monthly averaged comparisons with ECMWF model winds shall help characterizing large-scale biases and systematic issues in SMOS NRT winds. We shall provide scatterplots of and similar statistics and analyses than the one proposed in previous sections but replacing satellite winds by ECMWF ones.
Global mean bias and standard deviations will be provided in a Table for each month of the year and for the entire period of the products and for all validation regions. Time series of daily-averaged wind speed differences and Monthly averaged wind speed difference maps will be generated to potentially detect regions of larger biases (expected in regions such as river plumes, warm and cold currents, the cold-tongue of the equatorial Pacific, …).