In particular, all scatterometer orbit files have been scanned in order to extract storm features which are being cross correlated with storm tracks as derived from JRA 25 reanalyses of the 850 Mb vorticity. This results in a unique database of all storm observations since 1991, each individual feature being associated with a single and well identifed storm event when possible (i.e.when seen also by the model). These storm observations are now being associated with available roughness and wind field high resolution images by SAR onboard RADARSAT-1 and ENVISAT, as well as swell observation by SAR wave mode (onboard ERS-1,ERS-2 and ENVISAT) when it can be established they were generated by the same storm event constituting a new and promising source to estimate the intensity of these events and the total energy transmitted to the ocean or resulting exchange at air/sea interface. We plan to include in this scope new sources of data such as altimeter and SMOS in order to build the most exhaustive catalog of storm observations.
All the extracted features, together with their respective descriptive properties, are indexed and registered into an advanced and user-friendly data and knowledge storage and extraction system, NAIAD, developped by Ifremer. A dedicated user interface will be built to allow users to query quickly data with respect to content oriented search criteria (and not only space and time location like in most geospatial information systems), based on the registered knowledge (from the offline data mining mentioned above). In addition, it will make possible to easily cross-reference and intercompare observations with other available sources of data : starting for instance from a single observation of a feature or event, all connected data (at least through space and time proximity, but possibly also through causality, similarity or propagation relationships) can be collected.
Storm detection from scatterometer (StormWatch)
||Scatterometers are satellite embedded microwave radar specially designed to measure the sea surface wind speed and direction under all weather and cloud conditions. Since the launch of ERS-1 in 1991, sea surface winds have been continuously measured at global scale thanks to an uninterrupted series of missions such as ERS-2; ADEOS-1, QuikSCAT, ADEOS-2 and now METOP-A or OCEANSAT-2. We have scanned the complete archive of some of these missions (currently ERS-1, ERS-2, QuikSCAT and METOP-A) in order to identify and register a complete index of all storm observations. Users with a focus on extreme wind events can now access this extensive catalogue which spans over more than 17 years. This work was supported by ESA, as part of the enhancement of the legacy of ERS missions, and achieved in collaboration with CLS Radar division.|
The StormWatch index consists in an identification of all storm events (including hurricanes, typhoons but also high latitude storms) in the observations collected by the satellite embedded scatterometers since 1991. Here is the list of products and related time coverage parsed to build this index.
ERS-1 25 km-resolution wind vectors (WNF)
1996-03-19 / 2001-01-17
ERS-2 25 km-resolution wind vectors (WNF)
1991-08-04 / 1996-06-02
QuikSCAT 25 km-resolution wind vectors (L2B)
1999-07-19 / 2009
ASCAT 25 km-resolution wind vectors
2007 / ongoing
The identification of an extreme event on scatterometer data is primarily based on the high wind velocity detection. However care must be taken since high wind velocities retrieved from scatterometer measurement can come from contamination by rain or the presence of sea ice. Therefore, it is of primary importance to check the quality of the scatterometer measurement and apply the required corrections prior to any detection.
Once the scatterometer winds can be trusted, the first step of the identification of a storm event can be based on a threshold wind speed. However, since we know that scatterometer winds are significantly underestimated in the high wind range, the threshold wind speed cannot be based on the actual Hurricane force wind threshold, for instance, that would lead to missing most of the storm events on scatterometers datasets. Therefore, the wind threshold for the identification of storm events on scatterometer datasets can be adjusted to a smaller value determined for instance by the minimum wind speed of the 1% highest quality checked wind speed recorded by a given scatterometer over a period of 1 year. By doing such, the storm event criterium can be considered largely independent on the scatterometer model used in the wind vector retrieval.
Properties characterizing the observed storm feature are then extracted from the swath section, such as :
- the storm position, set to be the position of the highest wind speed associated with the identified storm event.
- the extension of the storm event, set as the location where the wind speed decreases continuously from the maximum recorded wind but still remains higher than a minimum threshold wind speed. This threshold is configurable and set by default to 15m/s based on experience.
- the storm center, estimated as the location where the wind speed is maximum. This convention is in line with the possible use of StormWatch results to initiate tracking of storm generated waves whose main source is the higher wind area of the Storm.
- the storm intensity, estimated by the total wind power over the detected storm area. The wind power is the square root of the wind speed times the individual wind cell size.
- the maximum wind speed together with the area where the wind speed is detected above the scatterometer extreme wind threshold, considered as the dominant extreme parameters to be extracted together with the maximum wind vorticity
The storm observations are also colocated with numerical model outputs for which similar properties are extracted, and with hurricane tracks and properties delivered by various hurricane centers.
The methodology is now being extended in order to build similar storm catalogs from other sources of data such as various multi-sensor or blended weather/satellite wind fields, in order to assess and intercompare the sensibility and the response of different sources of data to this algorithm.
|The method described above provides scattered observations from various events that need to be related to each other. For the same single event, there can be several days between two consecutive observations by the same sensor and therefore it is not possible from observation only to assess if two observations belongs to the same event or not, and subsequently relate each observation to unique and correctly identified events, which is required if one wants to establish some classification of these events based on observation.|
One way to achieve this is to run a storm detection and tracking algorithm on weather model outputs, the high temporal resolution of these models allowing to efficiently track events along time. We choosed for that purpose the JRA25 reanalysis (http://www.jreap.org) from the Japan Meteorological Agency which :
- covers all the scatterometer era (1991-today) and beyond, and therefore can be used consistently over our complete period of focus
- is continuously updated and therefore allows to update our catalog, including the SMOS era, using the same methodology and with the same source of data
- assimilates all scatterometer data and therefore ensures that the retrieved storm tracks should be consistent with the scatterometer storm observations, easing the matching of the two sources
- is arguably performing better wind retrieval in the tropical areas compared to other analyses such as ECMWF.
The applied methodology (Hodges, 94) relies on image segmentation and feature extraction over a sequences of 850 Mb vorticity fields from the JRA25 model, applied to the unit sphere. Further filtering and combination with other parameters (such as surface wind speed) is then applied to select only the most significant events in terms of intensity, duration and extent.
19 years (1991-2009) of JRA25 model were processed, at global scale, providing an equivalent time series of tropical and extra-tropical storm tracks over the whole globe to be matched with the available sources of observation.
The identification and lifetime history restitution of storm events can also be retrieved from indirect and remote observation several thousands of km away of the swells generated by these storms. Directional spectra of the swell is available from SAR wave mode observations (ERS from 1991 to 2011, Envisat from 2002-ongoing) based on an inversion algorithm developed by Ifremer. Retro-propagation of the observed swells can then be applied using a simple backward propagation model and the storm generation area identified from the focus point (Ifremer and CLS) . The swell properties provides a unique insight on the storm history and intensity, stressing out the need to uniquely identify each event and relate each observation from whatever source to these events to offer the most extensive view of each event and allow for proper characterization and classification.
The identification and propagation of these swells is illustrated through the "fireworks" animations.
Association between storm events and swells is an ongoing effort and a product spanning over the full Envisat era (from ASAR wave mode data) will be soon available. Meanwhile a reprocessing of the ERS-1 and ERS-2 is ongoing at Ifremer to provide a new time series of swell observations using the same inversion method than for Envisat. It will then be investigated if data quality and sampling is sufficient to apply a swell generation area detection algorithm as for Envisat.
Cross-source storm database
As mentioned in the above section, building a proper database for storm characterization and classification relies on interconnecting the observations from various sources (offering a different view of the same phenomenon) and relating them to unique events. The rationale applied to achieve this task is :
- to first identify the most exhaustive catalog of events, which is done as shown before :
- identifying and tracking features from high temporal frequency model outputs
- identifying from observations events that may not have been captured by numerical models (StormWatch)
- then building of the cross-source database is performed both ways :
- from observation to event : detecting events from analysis of satellite imagery or in situ data (StormWatch, swell tracking from SAR or buoys, ) and finding the event in the catalog matching each observation
- from event to observation : extracting in the satellite or in situ data all the observations intersecting the path of the storm events documented in the catalog, which is the fastest way to populate a cross-sensor database once we have enough confidence in the storm catalog
- for each storm observation by any source, metadata are extracted to document it (time and geolocation, file and subset into file, properties on the seen event computed from the data) and stored into a database, allowing then fast identification and access to the relevant data
The current storm database includes :
- tracks from JRA25 reanalysis
- wind speed from ERS-1 & ERS-2 scatterometer
- wind speed from QuikSCAT scatterometer
- wind speed from ASCAT onboard METOP-A scatterometer
- swell from Envisat/ASAR
- SAR images from Envisat and Radarsat-1 (over tropical storms only)
It is planned to extend this database to available altimeters, scatterometers (SeaWinds, OceanSAT-2, NSCAT) and radiometers (WindSat, AMSRE, SMOS).