Print Version

Chapter II: Satellite Derived Winds

Table of Contents

Atmospheric Motion Vectors

Knowledge of atmospheric motion is essential for many applications. Information on high-level atmospheric winds is of great importance for forecast models as the current state of the atmosphere has to be specified before the future state can be predicted. Winds in the upper levels can be observed using radiosondes or aircraft measurements, but those observations are limited in time and space. As satellites provide worldwide and continuous data, they are the ideal data source for regular upper atmospheric wind information.

Atmospheric motion vectors are computed by tracking clouds or moisture features from a sequence of satellite images and are thus the only observation type that provides good coverage of upper tropospheric wind data over oceans and at high latitudes. Geostationary satellites are best suited to tracking clouds as they provide data from the surface and the atmosphere with a high temporal resolution and from a non-varying satellite perspective. In the IR and VIS ranges AMVs are derived from clouds, whereas in absorption bands like the water vapor channels they are derived from moisture fields.

The routine production of AMV data started as early as in the 1970s. There have been many improvements in computation and data usage since those days: the image resolution has increased and the time interval between satellite images has decreased. The available number of satellite channels used for extracting cloud motion vectors has also increased since the 1970s. In the beginning, AMVs were produced using data from geostationary satellites only, while nowadays winds are also computed from polar orbiting satellite data.

Cloud Motion Winds (CMW)

The tracking of cloud or moisture features from satellite data does not automatically provide in-situ winds since clouds can be stationary (e.g. lee clouds) or move in a direction not coinciding with the local wind field (e.g. cumulonimbus clouds). Therefore it is very important to distinguish between motion vectors and winds. Motion vectors provide the direction and propagation speed of a selected cloud feature, while winds represent the current velocity of an air parcel at a certain height.

In practice this differentiation is achieved by a careful selection of atmospheric tracers and the use of model winds as a first guess for wind speed and direction at a given atmospheric level. For example the displacement of a front line results in atmospheric motion vectors while using a cloud fiber as tracer suits the computation of cloud motion winds better. Both applications have strengths as will be shown in some practical applications in chapter 3.

The computation of cloud motion winds requires model data anyway, not only for the first guess comparison and validity check, but also for the height assignment of the derived winds. This is achieved by assigning the brightness temperature of the tracer cloud to a given height by using a vertical temperature profile from a model.

Table 1 gives an overview of the agencies currently producing and disseminating cloud motion winds and the satellites used for that purpose. Altogether the data provides a nearly global coverage, see Figure 3.

Table 1: List of agencies currently producing cloud motion winds (green: geostationary satellites, red: polar orbiting satellites).

EUMETSAT Europe Meteosat-10 (GEO), Meteosat-7 (GEO), MetOp-A (P), MetOp-B (P)
NOAA/NESDIS, CIMSS USA GOES-13 (GEO), GOES-15 (GEO), Aqua (P), Terra (P), NOAA-15 (P), NOAA-18 (P), NOAA-19 (P), NPP (P)
JMA Japan MTSAT-2 (GEO) HIMAWARI 6 - 8
IMD India Kalpana (GEO), INSAT-3D (GEO)
CMA China FY-2D (GEO), FY-2E (GEO)
KMA South Korea COMS (GEO)

Figure 3: Location of all AMVs used in the data assimilation for the UK Met Office model in 2013 (Source: UK Met Office).

Methodology

The methodology is explained with help of AMVs derived from the SEVIRI instrument onboard MSG. Various satellite channels are used to calculate cloud motion. The IR window channels (IR10.8 μm, IR12.0 μm) and the visible channels (VIS0.6 μm, VIS0.8 μm) are used for cloud tracking. In contrast, both WV channels (WV6.2 μm, WV7.3 μm) provide information for cloudy and cloudless areas. For the latter, moisture patterns or gradients are tracked instead of clouds.

The algorithm to derive cloud motion winds mainly consists of the following steps:

  1. Initialization of the data
  2. Tracer selection
  3. Tracking
  4. Height assignment
  5. Quality control
  6. Orographic flag computation

Ad 1: Initialization of the data: In this preprocessing step the algorithm provides all the auxiliary data needed in the following stages for computing the winds (e.g. pixel latitude and longitude, solar zenith angle, reflectance and brightness temperatures for all used channels ...). This preprocessing step is crucial as the exact location of the feature is critical for all the further calculations.

Tracer selection: In this first step a cloud or moisture feature that can be tracked easily, has to be identified. Some clouds make better tracers of the atmosphere's motion than others. The EUMETSAT target extraction scheme mainly depends on the contrast and the entropy as selection criteria. Therefore suitable targets have the highest contrast and the largest amount of standard deviation (entropy).

Ad 3: Tracking: The motion of the cloud/moisture feature is extracted by analyzing two or more consecutive images (see Figure 4 ). This task usually uses a large amount of computational resources. Thus, various methods can be used:

  • Euclidean distance
  • Cross correlation in the spatial domain
  • Cross correlation in the Fourier domain

At EUMETSAT a cross correlation is used.

Figure 4: Tracking a cloud pattern for deducing an AMV.

Ad 4: Height assignment: In this step a pressure level is assigned to the vector based on a temperature/pressure profile derived either from radiative transfer calculations in the environment of the target or from model data. When computing AMVs from WV images in a region with clear sky conditions, the brightness temperature from the latter is used for the height assignment.

Height assignment is the most important, but also the most difficult step as it is very error-prone. An error of +/- 50 hPa can induce a major wind speed error especially at levels with high wind shear (see Figure 5).

Figure 5: The impact of a height error on the wind speed at two different levels (Source: Salonen et al.).

There are several difficulties with the height assignment:

  • Which pixel of the target area should be selected for height assignment?
  • Which level is the most representative of cloud motion? Normally the cloud top temperature is assigned to a pressure level. Furthermore, each layer of the atmosphere emits a different amount of radiation, which is true especially for absorption channels. However in the end the AMV is treated as single-level data in NWP models.
  • The assigned pressure level is normally calculated as a function of the cloud type. Semi-transparent or sub-pixel clouds are often the best tracers as they provide good radiance gradients, but their height assignment is very tricky as the measured radiance contains a large and unknown contribution from lower cloud layers and/or the earth surface.

At EUMETSAT height assignment is usually done by the two following algorithms:

  • Equivalent Black Body Temperature (EBBT): the measured brightness temperature of the tracer is compared to forecast temperature profiles in order to infer the level of best agreement. This level is chosen as the observation height. The EBBT method shows good results for opaque clouds. In semi-transparent or sub-pixel clouds this technique often assigns insufficient height to the clouds. A big advantage of this method is that it is available almost everywhere.
  • Cross Correlation Contribution Method (CCC): defines AMV pressure by only considering the pressure in the pixels that contribute to image correlation best.

Ad 5: Quality control: Quality control is part of the post-processing procedure. This step is crucial in determining whether a derived motion vector fulfills the criteria of satellite-derived wind. During this process each motion vector is automatically set to a quality flag. This quality flag is the result of several independent quality tests:

  • Forecast consistency: speed and direction of measured and forecasted wind at the same altitude are compared to each other
  • Spatial consistency: neighboring wind vectors (same time-step, same level) are compared
  • Temporal consistency: the wind vectors of consecutive time steps are compared to each other
  • Temporal pressure consistency: altitude and pressure set to the 2nd and 3rd time step are compared to each other
  • Image correlation (WV6.2 μm, IR10.8 μm): at EUMETSAT this is done between WV and IR images

The weighted sum of all individual quality tests represents the final Quality Index QI which ranges from 0 to 1. Distinct QI thresholds determine whether a motion vector can be used as wind equivalent or not.

Ad 6: Orographic flag calculation: This is another post-processing step in which tracers affected by land are detected and finally rejected. For example, tracers whose flow is blocked (e.g. barrage cloudiness) or affected by mountain ranges (e.g. lee waves) are not representative of atmospheric motion.

Figure 6 Shows the result of the entire algorithm described above: a satellite derived wind product from EUMETSAT. The different colors of the wind vectors represent different height levels.

Figure 6: Example of satellite-derived wind product from EUMETSAT.

Polar Satellites

Initially AMVs were computed using only data from geostationary satellites. Thus, AMV data was available from 55° latitude north to 55° latitude south. Theoretically AMVs can also be derived in regions near the poles, but given the strongly inclined satellite viewing angle the quality of the wind vectors thus obtained would be quite bad. Hence there was a constant lack of observations in polar regions. In the early 2000s, NOAA started to routinely produce polar wind vectors by using MODIS data. Nowadays AVHRR and VIIRS data are also used for computing cloud motion winds and so they are available at latitudes greater than 70°. NWP models in particular benefit from this development as polar regions had long been data scarce areas. Polar wind vectors are mainly used in data assimilation.

Extracting satellite winds over polar regions brought not only improvements for the forecaster community but also new challenges. A large time interval of about 50 to 100 minutes between consecutive images, small areas for tracking, viewing angle problems, parallax shifts and varying pixel size are only some of the problems. For tracking a feature and thus determining the motion vector, normally two (EUMETSAT) to three (CIMSS/NOAA) consecutive images are used. As the orbit of polar satellites is shifted for each overpass due to the Earth's rotation, only a small area where three consecutive overpasses overlap is available (see Figure 7). Due to the long temporal gap between consecutive images, only the most persistent clouds can be tracked. Matching the features is also more demanding as the cloud shape might change in such a long period. Nowadays, image pairs instead of image triplets are used for deriving motion vectors. On the one hand, the global coverage of cloud motion winds is increased by the addition of polar satellites, but on the other hand the quality of the wind information thus determined might decrease a bit.

Height assignment for polar winds

In polar regions the tropopause is lower than in middle or tropical latitudes. Therefore the first step in height assignment is checking that the tracer cloud is not situated above the tropopause. The level of the tropopause is determined by using forecast fields. The altitude of the tracer cloud is then restricted to a layer between the ground and the forecasted tropopause level. AVHRR measurements represent a special case as no appropriate water vapor or CO2 channels are available. Because of this only the EBBT method can be used for height assignment. If available, EUMETSAT uses the IASI cloud top height product for determining the cloud top pressure. However, due to the coarse sampling of the IASI instrument this data does not cover the whole area scanned by the AVHRR instrument.

Figure 7: AMVs can be derived from polar-orbiting satellite imagery where the successive overpasses overlap (shown in grey) in polar regions (Source: D. Santek (CIMSS)).

Combination of GEO/LEO or MetOp-A/B data

Geostationary and polar data help to achieve a good global coverage of cloud motion winds. Nevertheless a data gap exists between roughly 55° and 70° latitude and therefore synoptic phenomena like polar jet streams are partly not reflected in AMV data. This problem can be solved by combining either geostationary and polar satellite data or data from two LEO satellites situated in the same orbital train, such as MetOp-A and MetOp-B.

An advanced method is used to blend imagery from LEO and GEO satellites in order to produce composite CMWs. Data from geostationary satellites (GOES) is blended with data from polar satellites (NOAA-15 through NOAA-19, MetOp-A, NASA's Terra and Aqua). When computing these CMWs, high-spatial-resolution LEO data is the first choice if available. If not, high time resolution GEO data is used. The satellite winds are produced using three half-hourly composites. These mixed-platform wind vectors seem to be comparable to other available CMWs in terms of the root-mean-square error. Recent experiments have shown that these composite wind vectors tend to have a positive impact on the forecasts in the southern hemisphere. In the northern hemisphere the impact seems to be neutral.

On 17 September 2012, EUMETSAT launched the MetOp-B satellite. This tandem configuration of two satellites - MetOp-A and MetOp-B - operating on the same orbital train with a phase difference of 46 to 55 minutes provides the opportunity of creating not only polar but global AMVs from LEO data. When compared to CMVs from geostationary satellite data only minor differences regarding the assigned height of the derived winds are noticeable. As MetOp satellites do not operate water vapor channels the methods for height assignment are restricted as mentioned earlier.

Scatterometers

Scatterometers are active remote sensing instruments for deriving wind direction and speed from the roughness of the sea. They are used by low Earth orbiting satellites and act like radars: they transmit electromagnetic pulses and detect the backscattered signals. Spaceborne wind scatterometry is an indirect method of measurement. It has become an increasingly important tool for monitoring climate, forecasting (marine) weather and studying the atmosphere-ocean interactions.

Scatterometers measure the radar cross section of the ocean surface. A Geophysical Model Function (GMF) provides the radar cross section as a function of the equivalent neutral wind vector at 10 m anemometer height (not the actual wind!), incidence angle, relative azimuth angle, radar frequency, and polarization.

Depending on the wavelength of the radiation a distinction can be made between C-band and Ku-band scatterometers. The two radar types are compared with each other in Table 2. The wavelength of the radar is chosen according to the sampling scale. Thus, Ku-band radars with a wavelength of about 2 cm are sensitive to rain.

Table 2: Comparison of C-band and Ku-band scatterometers.

C-band Ku-band
Frequency 5.255 GHz 13.4 GHz
Wavelength 5 cm 2 cm
Limitations less detection in the higher wind range (> 60 kt), sea ice, coastal coverage - Land contamination sensitive to rain, coastal coverage - Land contamination
Scatterometer ASCAT-A and ASCAT-B QuickScat/RapidSCAT/HY2A
Polarization VV-pol Dual-polarization
Sampling 12.5-25 km 25 km- 50 km
Geometry Static Rotating antenna
Swath Double (about 550km each) Single

History

During World War 2 it was discovered that radars also picked up some kind of clutter over the oceans. It took some 20 more years to discover that this noise was related to wind velocity. Scatterometry started in the 1970s – the first satellite with a scatterometer onboard was Seasat-A (NASA). However, widespread usage of scatterometers started in the 1990s. Nowadays scatterometer data is operationally used especially for data assimilation and for marine nowcasting.

Figure 8 shows an overview of the current and proposed satellite missions carrying scatterometers.

Figure 8: Overview of finished, current and proposed satellite missions with scatterometeres onboard. The only exception is 'RapidSCAT', which is a scatterometer on the international space ship (ISS). (Source: http://ceos.org/ourwork/virtual-constellations/osvw/).

Backscatter

Scatterometers are radars and send pulses of radiation at the surface. The backscattered signal varies according to the wind speed and its effects on ocean roughness. Higher friction velocity means higher roughness of the sea, and thus a stronger backscattered signal. It is important to know that scatterometers are not sensitive to large (tidal) waves, but only to the wind-driven centimeter-scale capillary waves. These ripples are superimposed on the larger waves.

 

 

Figure 9: Bragg scattering in different Beaufort classes and therefore different states of the sea. Blue arrows show the incoming radar beam, red arrows the scattered radiation.

The scattering mechanism that occurs when electromagnetic radiation and water waves interact with similar wave lengths is called Bragg scattering. The Bragg scattering explains the effects of the reflection of electromagnetic waves on periodic structures whose distances are in the order of the wavelength.

Unlike with a flat surface, radiation is scattered in all directions rather than just being reflected. Only this Bragg scattering makes it possible to derive wind speed and direction as it changes the total amount of energy that returns to the satellite.

Bragg scattering is dependent on the incidence angle. This means that there is an optimal range of the incidence angle for scatterometer measurements. Typically angles between 30° and 60° are used as they provide the largest sensitivity to changes in wind speed. The design of scatterometers takes these facts into account which means that scatterometers only see the ocean surface from the side, never directly from above.

In Figure 9, Bragg scattering is shown for different states of the sea. The surface roughness modulates the backscattering. When the sea is glassy like at Beaufort force 1, the radar beam hardly scatters at all and merely bounces off the surface. The radar beam is refracted and keeps going. Only a small proportion is reflected back to the instrument. As the wind picks up and the waves get higher, more and more energy is reflected back to the scatterometer.

Ultimately the scatterometer measures the backscattered and normalized radiation from an extended target area (wind vector cell) sigma naught σ0.

Geophysical Model Function (GMF)

The (semi-)empirical GMF compares ocean surface wind speed and direction to the backscatter cross section measurements. The geophysical parameter that should be estimated is the wind vector. For ASCAT measurements, three known values of sigma naught are available and two unknown values - wind speed and wind direction - have to be determined. The GMF is a function of several variables:

σ0model = GMF (U10N, θ, φ, ρ, λ)
U10N ... equivalent neutral wind speed (wind at 10 m height for given surface stress assuming the marine boundary layer is neutrally stratified)
θ ... incidence angle
φ ... wind direction with respect to the direction of the radar beam
ρ ... radar beam polarization
λ ... microwave wavelength

The dependency on the azimuth angle allows determining not only the wind's speed but also its direction. Using multiple observations from different azimuth angles improves the accuracy but a minimum of two observations is required.

Scatterometer instruments

By measuring the backscattered signal by the ripples and waves, scatterometers derive wind direction and speed from the roughness of the sea. The backscattering is dependent on the incidence and azimuth angles: the dependence on the incidence angle helps to derive wind speed, while the dependence on the azimuth angle allows deriving wind direction. Two kinds of instruments are used for this measuring principle:

  1. fan beam scatterometers (ASCAT)
  2. rotating pencil beam scatterometers (RapidSCAT)

Ad 1.:

ASCAT is a so-called fan beam scatterometer. This instrument is a radar with vertically polarized antennas. It consists of two sets of three antennas. Thus each point of the surface is seen and measured by the scatterometer three times. The radar beams are directed 45 degrees forward, orthogonally, and 45 degrees backwards relativ to the satellite's flight direction, on both sides of the satellite ground track (see Figure 10). Each scanning swath has a width of 550 km with a scanning gap of 670 km in between (see Figure 11). This scanning gap is a result of the optimal incidence angle of Bragg scattering. The antenna structures are typically about three meters in length and require large unobstructed fields of view on the spacecraft. Due to the antenna geometry, different measuring points on the surface have different incident angles.

Figure 10: Antenna design of a fan beam scatterometer.

Each beam provides measurements of radar backscatter on a 12.5 km and/or 25 km grid, in such a way that each swath is divided into 41 (41*12,5) or 21 (21*25) so-called Wind Vector Cells (WVCs). Each WVC is viewed thrice because of the three different viewing directions caused by the antenna geometry. Thus one obtains three independent backscatter measurements - triplets of sigma naught σ0 - separated by a short time delay. A mathematical model ("Geophysical Model Function" GMF) is then used for deriving wind speed and wind direction on the ocean surface out of the triplets of sigma naught.

Figure 11: Measuring geometry of a fan beam antenna onboard of ASCAT.

The ASCAT fan beam instrument operates at a frequency of 5.255 GHz (C-band). This makes it fairly insensitive to rain, but C-band radars have limitations when it comes to extreme wind speeds larger than 60 knots.

Ad 2.:

The rotating pencil beam scatterometer uses a rotating radar. The radar emits signals at certain incidence angles. It never looks directly beneath the spacecraft as those incidence angles are not suitable for deriving wind speed and direction. The radar sends out two rotating beams with different polarizations (vertical and horizontal). For example RapidScat, the scatterometer which is on the ISS, scans at an incidence angle of 49 degrees for the inner beam (horizontal polarisation) and 56 degrees for the outer beam (vertical polarization) as shown in Figure 12. These two polarisations have slightly different backscatter characteristics and thus the quality of wind speed and direction estimations is improved.

Figure 12: The working method of RapidScat as an example of rotating pencil beam scatterometers (image source: www.nasa.gov/Rapidscat).

Figure 13 shows the measurement geometry of a rotating pencil beam scatterometer. The two different incidence angles and the different polarizations allow a very precise determination of wind speed and direction. Each footprint on the ocean surface is observed twice: when it is located in front and behind the satellite track. Thus up to four observations of the same spot are made in the mid-swath when horizontally and vertically polarized beams are considered individually (see Figure 13). In the far-swath, only the vertically polarized beam provides surface observations. The best quality of measurements can be obtained in the mid swath due to good azimuth diversity.

Measuring backscatter with the help of a rotating pencil beam scatterometer has two advantages over the fan beam scatterometer: first, due to its compact design it is much easier to accommodate on satellites or spacecrafts. Second, no 'nadir-gap' appears. However, due to the poor azimuth diversity the measurements in the nadir swath may not be unambiguous.

Figure 13: Measurement geometry of a rotating pencil beam scatterometer (Source: J. Sienkiewicz).

Wind retrieval algorithm

Several steps are needed to convert the measurements of sigma naught into wind data:

  1. NWP collocation: first, land and sea ice regions have to be excluded from the wind retrieval areas
  2. Quality control: the sigma naught values are quality checked
  3. Wind computation by GMF inversion: the Geophysical Model Function (GMF) is used for converting the sigma naught values to wind speed and direction. The GMF has an empirical or semi-empirical form.
  4. Field-wise ambiguity removal: each wind vector cell has four possible solutions for wind direction and speed. The correct solution is determined by using NWP forecasts and wind field spatial patterns.

Limitations

In general, the quality of scatterometer winds is quite good. But there are some situations where the data can be compromised:

  • sea ice and land contamination (see Figure 14)
  • large spatial wind variability (for example in the vicinity of fronts and low pressure centers, e.g. downbursts)
  • rain - especially in Ku-band systems (see Figure 15)
  • higher wind speeds (>60 knots) - especially in C-band systems

Figure 14: Example for sea ice and land contamination from the QuikScat scatterometer (source: http://www.remss.com/missions/qscat).

Figure 15: Example for rain contamination from the QuikScat scatterometer (source: http://www.remss.com/missions/qscat).

Rain effects

Ku-band scatteromenters are particularly sensitive to rain. For C-band scatterometeres the rain effects are less significant.

  • radar signal and the backscattered information are attenuated by rain
    - σ0 and the retrieved wind speed decreases
  • radar signal is scattered by the rain drops
    - σ0 and the retrieved wind speed increases and directional information can be lost
  • the roughness of the sea surface is increased due to the splashing of rain drops - this effect mainly occurs at low wind speeds when the sea surface is calm
    - σ0 and the retrieved wind speed increases
    - directional information may also be lost
  • variable roughness of the sea surface due to wind downbursts and/or a gust front
    - confused sea state: wind speed and direction might be unclear
    - area influenced by the downburst might be bigger than the area affected by the splashing of rain

Rain contaminated winds can be detected in the majority of cases as those wind vectors are oriented perpendicular to the satellite's track. Rain also increases wind speed most of the time.