Weather prediction relies on the large number of observations carried out over the entire globe and made available to meteorological centers via the Global Telecommunication System (GTS). The observations are used to determine the state of the atmosphere at a particular time, the so-called analysis. The analysis serves as the initial state of numerical weather predictions. Starting from this state, the differential equations that govern the dynamics of the atmosphere are integrated by the numerical model into a prediction. As the number of observations is much less than the total number of model unknowns, the computation of the initial state requires some additional information that is in practice provided by the previous weather forecast – the so-called “first guess”. The operation that mixes the first guess and observation is called assimilation. This data-fusion process computes the differences between the first-guess and the observations, and modifies the first guess so that both match. The modification of the first-guess must be done very carefully, it must not introduce unrealistic features that would destabilize the forecast or violate the fundamental laws of the atmospheric physics (for instance, a large-scale modification of the wind field requires that a large scale modification of the pressure field is also brought otherwise the geostrophic equilibrium will not be verified anymore). At Météo-France, the global model ARPEGE is run 4 times a day at 00 UTC, 06 UTC, 12 UTC and 18 UTC. Each run contains an assimilation. The first step of the assimilation consists of a screening procedure that aims at discarding the observations that are too far off the first-guess and thus appear to be gross errors. For many information types, the number of observations that are tagged as gross errors can be very large and represent more than 90% of the total number of observations.

The nature and volume of wind observations available for weather forecast is detailed in Table 1. They are grouped in three major classes (column 1) depending on whether they are obtained from a satellite, or are surface measurements acquired by surface based sensors (Surface), or are altitude winds.

Surface winds are measured by in situ sensors (anemometers) based either on land or at sea on ships or buoys. The geographical distribution of these types of observations is highly irregular (Fig. 1). Most of them are obtained in highly populated regions of the globe, and large, desert areas (open sea, Sahara, polar regions, tropical rain forest…) are almost completely empty. Although the technology involved is rather simple, their total number is relatively small (about 100 000/day) because telecommunications are necessary to deliver the measurements in a time period short enough (in less than a few hours in practice) so that they can be assimilated. This requirement restricts the number of possible sites to a rather small number of stations equipped with wired or wireless telecommunication systems (buoys, for instance, transmit their data via expensive satellite communication links). In addition, many of these observations do not pass the screening phase of the assimilation. This does not mean that they are not of good quality – indeed, many of them are of good quality – but they suffer from an inherent drawback common to all surface measurements: they are influenced by local effects – topography for instance – that the forecasting models are not able to represent correctly. Thus, they bring information that the model cannot use properly and they may even destabilize the forecasting process.

Altitude wind observations are more numerous as their number exceeds 200,000. However, this figure must be considered with care as 97.5% of the observations are provided by commercial aircrafts specifically equipped for weather observation. This information is of great value for aviation as it characterizes the weather conditions prevailing on major flight routes or in the approach areas of airports (for instance, it enables precise predictions of gas consumption or time of arrival), but for weather prediction, these observations suffer from a major limitation: most of them are along a limited number of flight routes (Fig. 2) and characterize the wind at the typical flight altitude of ∼10 km.

TEMP messages contain the observations made by radiosounding systems. A radio sonde is a small and light device that combines several probes, usually for temperature, humidity and pressure. The sonde is attached to a balloon inflated with nitrogen. As it is lighter than the air, the balloon goes up in the atmosphere with the sonde underneath. Along the ascent, the sonde transmits the temperature, pressure and humidity measurements by a radio frequency link. The position of the balloon relative to receiving station at the surface is determined by a tracking system which nowadays is generally a differential GPS (Global Positioning System). The wind is derived from this information. Today, radio-sondes are considered the reference for altitude winds because they are accurate and reliable. In spite of this, their number is very limited around the Earth (Fig. 3) because they are expensive (the cost of the radio-sonde is of the order of 100 €, and a radiosounding requires the work of an operator for more than an hour). Most of the radio sounding stations are in rich countries, and operated only 2 or 4 times a day. In continental France for instance, seven radio sounding stations launch balloons two times every day at 00 and 12 UTC.

Like TEMP, PILOT gives wind information obtained by tracking a balloon, but this time the balloon has no sonde attached to it. The wind is derived from the time derivative of its position determined by optical or radar tracking.

Two different sources of information on wind are available from space. One is called SATOB. SATOB winds are derived from satellite images. Distinctive features – clouds or water-vapor structures – are tracked across successive images of the same scene in either the visible, the infra-red or the water vapor channel. Their apparent movement is then converted into a wind information. The altitude of the information is computed from the infrared brightness temperature of the cloud top (or water vapor structure). The images come mostly from geostationary satellites (Meteosat in Europe, GOES in the USA…) that offer the advantage of always looking at the same point of the Earth and a fast repetition rate (one image every 15 minutes with Meteosat Second Generation). For several years now, the concept has also been applied to images from polar orbiting satellites – in particular TERRA and AQUA from the National Aeronautics and Space Agency of the United States of America.

SATOB provides a large set of wind data – about 1,500,000 pieces of information every day – with a good geographical coverage (with the exception of the high latitudes that are not seen by the geostationary satellites – Fig. 4) and an altitude range from the near-surface to altitudes as high as 100 hPa (or ∼15 km). However, they suffer from major limitations. SATOB does not provide vertical profiles of wind but only one measurement at the cloud top altitude for a given latitude and longitude. It is thus not possible to document possible wind shears along the vertical. Wind measurements are restricted to cloud regions – no information is available in cloud-free regions – and the number of measurements is reduced during night time when no visible images are available anymore.

A second source of information from space is the various scatterometers that are permanently flying around the Earth in low-level, polar orbits. Scatterometers are active remote sensors operating in the microwave frequency domains (frequencies between 5 and 15 GHz). They measure images of the scattering cross-section of the surface (that is, the ability of the surface to reflect the radio wave back to space). Over the ocean, the radar cross-section varies as a function of the wind force and direction relative to the azimuth of the radar beam. By combining several beam directions, the wind force and direction can be estimated.

Fig. 5 shows the European polar-orbiting meteorological satellite METOP-A with the Advanced Scatterometer ASCAT on board. The three double antennas of ASCAT are visible below the spacecraft. They measure the sea-surface cross-section in two swaths centered at an inclination of 36 degrees to the left and the right of the satellite track. Each swath is 550 km wide, the horizontal resolution of sea-surface wind measurements is 50 km.

Scatterometers provide a wind information in regions – the oceans – that are otherwise poorly documented by other wind observation systems. They have the advantage that they can make measurements in all types of weather conditions. For instance, they are able to measure winds below cyclones and thus provide observations of great value for the forecasting of such dangerous weather hazards. However, the observations are restricted to the surface, they give no information at all on the vertical structure of the atmospheric dynamics.

In the late 1990s, the European Space Agency (ESA) decided and started a mission for a space-based Doppler lidar. The decision came after several studies that refined the scientific requirements and showed the system was technically challenging, but feasible, and a selection process that culminated in 1999 when the program entered the phase B of its development cycle. The program is called ADM-Aeolus, ADM standing for Atmospheric Dynamics Mission. It is planned to be launched during the second semester of 2011.

ADM-Aeolus is the second mission of the Earth-Explorers program of ESA (information on the program can be found online at http://www.esa.int/esaLP/LPearthexp.html, or in (ESA, 1998). Its aim is to measure winds at the global scale throughout the entire depth of the meteorological atmosphere (Andersson et al., 2008; Stoffelen et al., 2005). The mission is based on a Doppler lidar. The wind is measured by estimating the frequency Doppler shift Δν between a laser pulse emitted by the instrument and the light that the atmosphere scatters back to the satellite. According to Doppler's equation, the Doppler shift is proportional to the component v_r of the wind along the laser beam (the so-called radial velocity):

Here, λ is the laser wavelength (λ = 355 nm for ADM).

The measuring geometry of ADM is depicted in Fig. 6. ADM will fly along a polar orbit at about 400 km above the Earth-surface. The laser beam is always directed perpendicular to the flight track with a nadir angle of 35 degrees (optimal compromise between the sensitivity to the horizontal wind that increases with the nadir angle and attenuation of the laser beam with the distance along the line-of-sight that is weaker when the lidar is pointing closer to the nadir). The perpendicular line-of-sight offers the advantage that the satellite speed has no impact on the frequency Doppler shift (its component on the laser beam direction is null). However, it also follows that ADM will measure a single component of the wind, the east–west component – and will be almost insensitive to the north–south component.

The narrow laser beam (the spot has a diameter of ∼200 meters at the surface) hits the Earth at a distance of ∼300 km to the side of the flight track. Laser pulse returns are accumulated during 7 seconds so the wind profiles have a horizontal resolution of 50 km. The laser is switched off between two 7-second burst, leaving a gap in the horizontal sampling of ∼150 km between two profiles. The observational requirements set at the beginning of the mission design are listed in Table 2. The vertical resolution of the actual lidar will vary from about 250 m in the planetary boundary layer (better than the 500 m of the observational requirements) to 1 km in the free troposphere and 2 km in the stratosphere, for an accuracy of 2 ms⁻¹ to 5 ms⁻¹.

ADM implements two channels of detection, one dedicated to the analysis of the frequency shift of the light backscattered by the molecules, the other one for aerosol returns. The need for two detection channels arises from the fact that the spectrum of the molecular backscatter is broadband (proportional to the square root of the temperature, the width of the molecular spectrum is several Gigahertz) while the aerosol return is narrow-band (a few hundreds of megahertz in width). The frequency shift of the molecular return is determined by passing the light through two Fabry-Perot filters centered on either side of the emitted frequency (Fig. 7). This is the so-called double-edge technique described in Garnier and Chanin, 1992. When the Doppler shift is zero, the number of photons N_A and N_B counted at the output of the two filters are equal. When the Doppler shift moves the molecular spectrum to the right (left), N_A (N_B) increases while N_B (N_A) decreases. The Doppler shift is obtained by inverting the relationship between the instrument response defined as R = (N_A-N_B)/(N_A + N_B) and the Doppler shift. The inversion is done by applying calibration curves. Details can be found in Dabas et al., 2008.

For the narrow-band aerosol return, the Doppler shift analysis is based on a fringe-imaging technique: the atmospheric return received by the lidar is passed through a Fizeau interferometer. A fringe is formed between the two plates of the interferometer at a position proportional to the frequency Doppler shift. The fringe is imaged on a CCD by a proper optical arrangement. The position of the image on the CCD is computed from the number of photons counted by the CCD pixels (in ADM, the CCD is a square of 20 lines by 20 columns of pixels), it gives the position of the fringe in the Fizeau which in turn provides the Doppler shift estimation.

ADM-Aeolus will measure vertical profiles of one wind component, giving information on the possible wind shears. A major limitation of the system is that it cannot penetrate clouds (at the exception of the thin ones). It follows that there will be no wind information inside and below thick and extended cloud covers. When the cloud cover is broken, a limited number of narrow laser beams should be able to go through and sound the air masses below. Provided there are enough of them, it should be possible to measure their wind profiles. At last, optically thin clouds like cirrus are causing no problems. Not only are they only weakly attenuating the laser beam, but they also provide strong Mie returns that the lidar can use in order to measure accurate winds at the cloud altitude. Overall, the number of individual wind measurements made every day should be of the order of 50,000 to 100,000 on average (depending on the cloud cover). They will be delivered to the meteorological centers in less than 3 hours so they can be assimilated by most numerical weather prediction systems ((see Tan et al., 2008) for details on the processing and assimilation of ADM wind data).

For the weather centers that wish to assimilate ADM data, a portable version of the level-2 processors of the mission has been developed for deriving all the information needed for the assimilation of the wind data, and are available to any center that requests for it.

Processors and data will also be made available to the research community. Data will include wind measurements but also optical products like vertical profiles of the aerosol backscatter and extinction coefficients of the atmosphere. These latter products are obtained from the intensities of the molecular and aerosols returns through a complex processing scheme described in (Flamant et al., 2008). They will complement the large base of atmospheric data products collected by the CALIPSO mission (see http://www-calipso.larc.nasa.gov/ for details).

Although wind is a basic variable of the atmospheric state, it is rather poorly observed by existing observation systems. Ground-based systems deliver surface winds in mostly rich, populated regions of the Earth, leaving large gaps in the coverage above the oceans or deserts. In the upper levels, the major sources of data outside space are the reliable, but costly radio sondes and equipped commercial aircrafts. Both suffer the same type of limitations as ground-based systems, their geographical coverage is highly irregular, limited to the flight routes for the latter. First observations of wind from space were made by tracking clouds in visible satellite images. Called SATOB, they were extended to the infrared domain and the tracking of water vapour structures; they are today a major source of information for the wind. Their geographical coverage is nearly uniform (except for the high latitudes unseen by geostationary satellites), but the vertical structure of the atmospheric dynamics remains unresolved as SATOB relies on 2D images. For a given atmospheric column, there is a maximum of two wind measurements available (two wind components). The scatterometers complement the observation system by providing sea-surface winds. Their observations have proven to be very useful – for hurricane forecast for instance – they cover regions, the oceans, that are otherwise poorly sampled by ground-based systems.

Today the major gap in the wind observation system is the altitude, cloud free regions of the atmosphere. The space-based wind lidar ADM-Aeolus of the European space agency should fill when it is launched in 2011, as it is primarily targeted to the cloud free regions of the atmosphere and provides whole vertical profiles of one wind information.