Fourteen active dust devils were observed by the High Resolution Stereo Camera (HRSC) on Mars Express, which enable the first analysis of the forward speed of dust devils on Mars determined from orbit. Results show speeds on the order of 20 m/s, which compares favorably with values of the wind profiles estimated from the Martian Climate Database for higher altitudes. Smaller dust devils of 1 km height and very small diameters were found moving only at 1.5 to 6.0 m/s in agreement with surface wind speeds.
 The imaging of fast temporally variable surface features from orbit, for example, dust devils on Mars, was previously only possible in snap-shots until Mars Express arrived at the planet. The unique capabilities of the HRSC experiment on Mars Express include three stereo channels, covering the same surface area at different times [Neukum et al., 2004]. The nadir channel views vertically downward, while one stereo channel views slightly forward and the other slightly backward (at an angle of 18.9° from nadir). Thus, it is possible to detect the direction and speed of motion of individual dust devils by their change in position seen in the three images.
 Although no systematic search for dust devils was performed with HRSC images, our main focus was on the northern lowlands, in the spring and summer for images taken between 0600 and 2000 hours local time.
 We found active dust devils in Arcadia Planitia, Syrtis Major, Thaumasia Planum and Amazonis Planitia. Speeds, diameters and heights of the dust devils observed are listed in Table 1. The first three dust devils discovered by HRSC were in Arcadia Planitia, at mid-afternoon, Ls = 337° (local winter), in the same area where Thomas and Gierasch  found the first dust devils on Mars from Viking images. Unfortunately only the backward looking and nadir images covered the dust devils in Arcadia Planitia sufficiently to determine their speeds.
Table 1. Dust Devil Speeds and Sizes Without Parallax Errora
S1 represents the forward looking channel, ND the nadir, and S2 the backward looking channel.
 It is usually assumed that dust devils move with the ambient wind [Sinclair, 1969]. The fact that these three observed dust devils have the same direction of motion and move approximately the same distance in the same time seems to confirm this assumption. Another three dust devils were found in the backward looking stereo image. These vortices developed in the two minutes between the nadir image and stereo image which indicates that dust devils develop rapidly. The “older” dust devils were more diffuse in this image so that diameters could not be determined, in contrast to their better-defined appearance in the nadir image which indicates that they are short-lived atmospheric features.
 In Syrtis Major, only one possible dust devil was detected, as seen at Ls = 115° (local winter) in the mid-afternoon. This feature showed no vortex structure and looked more like a cloud, which was estimated to be moving at about 1 km. No shadow was seen.
 Five dust devils were discovered in Thaumasia Planum at Ls = 118° (local winter), also in the mid-afternoon, oriented in a row (Figure 1 and Animation S1). Vortical structure can be identified in four dust devils (1, 2, 4 and 5); dust devil 3 looks rather like a big dust cloud where no vortex can be seen. Four of the dust devils (2, 3, 4 and 5) were detected in all three stereo images. One (dust devil 1) was seen only in the nadir and backward looking image because it was out of the field of view of the forward-looking channel.
 Dust devils were also detected in Amazonis Planitia at Ls = 142° (local summer) at noon, in the area where most of the dust devils have been seen on Mars [Cantor et al., 2002; Fisher et al., 2005]. These five dust devils are different to the other examples; they were quite small with diameters of 50 to 100 m (near the limit of image resolution) and heights of 0.9 to 1.4 km. In addition, the change in position from one image to another was not large; the speed is small and has a large error. However, the speeds are in good agreement with expected values.
 We estimate the error in the recorded dust devil position (centre of plume) to be 5 pixels (62.5 m at highest resolution in nadir image). The uncertainty in dust devil position also leads to an uncertainty in time when the dust devil was imaged. Errors in position and time affect the value of the derived dust devil speeds and are taken into account in Table 1. These errors are not of the same kind as time and range errors in measurements on Earth where one observes the dust devils in field studies in mostly horizontal views [Snow and McClelland, 1990]. The size, height and distance travelled must be estimated in a relatively short time. For Mars with HRSC, we have three images and mostly three clear positions of the dust devils.
Toigo et al.  simulated the evolution and development of Martian dust devils with background horizontal wind speeds. Dust devils developed in their simulations in the “highest wind speed” case (wind speed approximately 25 m/s at the surface) and in the “no background wind” case [see Toigo et al., 2003, Figure 1]. Although their study was based on limited data, they suggest that dust devil development is not necessarily connected to background wind speed. Rennó et al.  and Rennó and Bluestein  showed that the potential intensity of convective vortices solely depend on the thermodynamic properties of their environment. Kanak et al.  arrive at the same conclusion that the existence of Martian dust devil-like vertical vortices does not depend on the mean background wind, sources of angular momentum or surface inhomogeneities. Vertical vortices formed in their simulation at the intersections of convective up-drafts where local maxima of vertical velocities were found.
 Relatively high dust devil speeds have been derived from HRSC images. The intersection of the shadow with the plume (interpreted as the vortex) defines the surface point which is tracked in each image. Otherwise, if the high speeds result from observations of the plume at higher altitudes and not directly from the surface, the parallax error has to be taken into account because of the tilted viewing direction (18.9°) of the stereo channels (Figure S1 and Text S1).
 We used wind speed profiles derived from the Martian Climate Database [Lewis et al., 1999] for a given time, season and region for comparison with the derived speeds and heights. Figure 2 shows an example for dust devil 4 in Thaumasia Planum. This dust devil was approximately 3.4 km high and travelled at 19.6 m/s, corresponding well with the wind profile speed at the altitude of 3.4 km. It seems that the speed was measured at the upper part of the dust devil and therefore represents a value of the wind speed at this altitude, with a speed of only a few meters per second at the surface. This is consistent with most of the observed dust devils except for dust devil 1 and 5 in Thaumasia Planum and those in Amazonis Planitia.
 According to the predicted wind profiles of the database, dust devil 1 and 5 in Thaumasia Planum should move at 4 to 7 m/s but also those dust devils have a forward speed of 16.8 and 20.8 m/s like the other three dust devils in Thaumasia Planum. It is remarkable that the five dust devils move almost in-line in the same direction although several kilometres separated from each other (Figure 1). They have maybe formed along an air mass boundary, and move forward at that high speed because of the high wind speeds existing along fronts and shear lines.
 The dust devils in Amazonis Planitia have heights of 0.9 to 1.4 km, suggesting a wind speed of 9.5 to 13.5 m/s from the Martian Climate Database for this altitude, but the observed speeds are 1.5 to 6.0 m/s. It seems that in this case the real forward speed of the dust devils was measured and not the wind speed at a higher altitude. This is confirmed by the analysis from the parallax error which leeds to an error in the distance. Speeds are between 10.6 to 14.4 m/s for those dust devils. The speed results showed no major changes for dust devils in the other regions. Values between 12.1 to 24.8 m/s correspond well with former results without parallax error. The parallax has no effect because the dust devils moved much a larger distance then those in Amazonis Planitia.
 Assuming that the results describe surface values of dust devil speed, there are two possibilities for the interpretation: First, dust devils do move with the ambient wind, which would suggest that the ambient wind has a higher value than the assumed 5 m/s. The analysis of Sinclair , Snow and McClelland  and Toigo et al.  show that dust devil occurrence on Earth is also possible at higher ambient wind speeds (>10 m/s) and thus a higher forward speed is possible. Alternately, if the ambient surface wind speed is a few meters per second, then the dust devils move much faster with a component in the dust devil motion direction which leads, besides high rotational and vertical speeds (within the dust devil), to a strong forward speed component. Nevertheless, dust devils seem to move in the direction of the ambient wind even if they move at lower speeds than the ambient wind speed [Sinclair, 1969]. This is shown by the dust devils in Arcadia Planitia and Thaumasia Planum, all of which moved in about the same direction.
 The HRSC observations of dust devils are consistent with the hypothesis of Cantor and Edgett  that dust devils evolve wherever the atmospheric conditions are suitable. This is in contrast to the previous assumption that dust devils develop mainly in flat desert-like areas, such as the northern lowlands. We found and analysed 17 dust devils in different regions, including Amazonis Planitia (a desert plain several kilometres below the reference aeroid) and Thaumasia Planum (south of Valles Marineris, 3 km above the reference aeroid), and conclude from that dust devil occurrence is widely spread over Mars. All dust devils were found between noon and 3 pm Mars local time in agreement with other observations on Earth and Mars [Sinclair, 1969; Wennmacher et al., 1996; Balme et al., 2003]. However, in contrast to common believe [e.g., Fisher et al., 2005], dust devil activity is also seen in local winter.
 The analysed large dust devils travelled at speeds of 20 m/s on average. Those are interpreted as near-surface values (underlined by the impression of the images). If it is a surface value, the consequence is that either the ambient wind speed must be higher than usually considered or the dust devil itself is travelling with a high forward speed independent of the environmental conditions.
 (CS) is supported by DFG under grant PA 525/4-2 in the Priority Program “Mars and the Terrestrial Planets”, (MP) by Bundesministerium für Forschung und Technologie via DLR under grant 50QP9909. (RG) was supported by NASA through a contract from JPL. (EH) was supported by the European Community's Improving Human Potential Program under contract RTN2-2001-00414, MAGE. We thank the HRSC Teams at DLR Berlin and Freie Universitaet Berlin. We acknowledge the effort of the HRSC Co-Investigator groups who have contributed to this investigation in scientific discussions.