Shock deceleration in interplanetary coronal mass ejections (ICMEs) beyond Mercury’s orbit until one AU

The CDPP propagation tool is used to propagate interplanetary coronal mass ejections (ICMEs) observed at Mercury by MESSENGER to various targets in the inner solar system (VEX, ACE, STEREO-A and B). The deceleration of ICME shock fronts between the orbit of Mercury and 1 AU is studied on the basis of a large dataset. We focus on the interplanetary medium far from the solor corona, to avoid the region where ICME propagation modifications in velocity and direction are the most drastic. Starting with a catalog of 61 ICMEs recorded by MESSENGER, the propagation tool predicts 36 ICME impacts with targets. ICME in situ signatures are investigated close to predicted encounter times based on velocities estimated at MESSENGER and on the default propagation tool velocity (500 km s ). ICMEs are observed at the targets in 26 cases and interplanetary shocks (not followed by magnetic ejecta) in two cases. Comparing transit velocities between the Sun and MESSENGER (vSunMess) and between MESSENGER and the targets (vMessTar), we find an average deceleration of 170 km s 1 (28 cases). Comparing vMessTar to the velocities at the targets (vTar), average ICME deceleration is about 160 km s 1 (13 cases). Our results show that the ICME shock deceleration is significant beyond Mercury’s orbit. ICME shock arrival times are predicted with an average accuracy of about six hours with a standard deviation of eleven hours. Focusing on two ICMEs detected first at MESSENGER and later on by two targets illustrates our results and the variability in ICME propagations. The shock velocity of an ICME observed at MESSENGER, then at VEX and finally at STEREO-B decreases all the way. Predicting arrivals of potentially effective ICMEs is an important space weather issue. The CDPP propagation tool, in association with in situ measurements between the Sun and the Earth, can permit to update alert status of such arrivals.


Introduction
Solar coronal mass ejections (CMEs) can trigger intense space weather events (Gonzalez et al., 2007). In order to predict potential consequences, it is important to understand the propagation of interplanetary CMEs (ICMEs) and magnetic clouds (MCs), which are a subset of ICMEs, in the interplanetary medium. An ICME magnetic ejecta is often preceded by an interplanetary shock (ICME shock) and a sheath.
The temporal delay between an eruption of a CME at the Sun and its impact at Earth, i.e. the maximum of geomagnetic activity, was found to be in many cases 80 h, which corresponds to a mean ICME propagation velocity of 500 km s À1 (Brueckner et al., 1998). Assuming that slow CMEs are accelerated by the surrounding interplanetary medium and that fast CMEs are decelerated can explain the observed travel times. Gopalswamy et al. (2000Gopalswamy et al. ( , 2001a found evidence of such acceleration/deceleration by comparing CME velocities (coronagraph observations) to ICME velocities (in situ observations).
The NASA STEREO mission (Kaiser et al., 2008) improved our understanding of the ICME velocity evolution in the interplanetary medium by means of stereoscopic images (Davis et al., 2009). The ICME speed along its Sun-to-Earth journey can be monitored with a triangulation technique based on STEREO observations (Liu et al., 2010). Based on a few case studies, CME acceleration seems to be confined to distances of 10 solar radii (R ), and fast CMEs become decelerated between mainly 20 and 50 R distance (Liu et al., 2016;Manchester et al., 2017).
This makes CME/ICME association complex (Bocchialini et al., 2018) and wrong associations lead to wrong transit velocity estimations. CME channelling (Möstl et al., 2015) and CME-CME interaction (Gopalswamy et al., 2001b), identified by radio type II burst intense continuum-like emission, make the CME/ICME association more tricky. In situ measurements farther from the solar corona permit to study the ICME shock deceleration in the interplanetary medium, avoiding the region where a drastic change can be observed in ICME propagation.
Following ICMEs between different vantage points in the solar system has been done up to Saturn (Prise et al., 2015) or Pluto (Witasse et al., 2017). Alternative to magnetic structures authors frequently use the Forbush decrease to identify ICMEs at various targets (Möstl et al., 2015;Freiherr von Forstner et al., 2018;Winslow et al., 2018). Witasse et al. (2017) studied an ICME that progressively slowed down from the Sun to Earth, Mars, Comet 67P, and possibly Pluto. Prise et al. (2015) noted unexpected ICME acceleration after an interaction with high-speed jets. Considering 15 events Freiherr von Forstner et al. (2018) noted ICME deceleration between the solar corona and L1 (170 km s À1 ) and between Earth and Mars (67 km s À1 , based on Forbush decrease observations).
Radial propagation of ICMEs or of ICME shocks in the solar system can be studied with the Centre de Données de Physique des Plasmas (CDPP) Propagation Tool (Rouillard et al., 2017) which is publicly available at http://propagationtool.cdpp.eu/. Witasse et al. (2017) successfully followed ICME propagation from the Sun to Pluto with this tool. We propose here to test the ability of the Propagation Tool by tracking ICMEs between MESSENGER and different probes up to 1 AU. All ICMEs are part of a catalog which is based on NASA MESSENGER observations (Winslow et al., 2015). The aim of this study is to estimate the ICME shock acceleration/deceleration between Mercury's orbit and the orbit of Venus (VEX observations) as well as out to a distance of 1 AU (ACE and STEREO observations). We also want to test the utility of the tool to reduce the number of false alerts of potential effective ICMEs when in situ measurements are available in real-time between the Sun and the Earth.
Our dataset and methodology are detailed in Section 2. The ICME hits of VEX, ACE and STEREO probes predicted by the Propagation Tool are compared to in situ observations in Section 3. We comment on the ICME acceleration beyond Mercury's orbit up to 1 AU in Section 4. Before concluding we put special attention to ICMEs observed by two spacecraft beyond Mercury's orbit (Sect. 5).

Data set and terminology
We based our study on a catalog of ICME observations by the NASA spacecraft MESSENGER at Mercury, which lasts from 19 May 2011 until 2 September 2014 (Winslow et al., 2015). Running the Propagation Tool in radial propagation mode for each of the 61 events in the catalog, we identify the ICMEs that probably hit another object besides MESSENGER up to one AU. These objects, or targets, are VEX that orbits around Venus, ACE, and STEREO-A and B spacecraft orbiting at %1 AU. In addition to the starting point, MESSENGER in the present case, the necessary inputs for the radial propagation option are: ICME radial velocity (default value v def = 500 km s À1 ), the angular size of the CME at the Sun (default value a CME = 45°) and the choice of the target. There is no measurement of the plasma bulk flow velocity available at MESSENGER, so we do not consider the drag option available in the Propagation Tool. That option permits to brake or speed up ICME velocity depending on the preceding solar wind velocity. The ICME shock acceleration (positive or negative) is studied by looking for an input radial velocity that matches with observation times at MESSENGER (t obsMESS ) and the targets (t obsTAR ). We then compare this velocity (v MessTar ) to the velocity at MESSENGER (v SunMess ) and at the targets (v Tar ).
When the Propagation Tool predicts a target hit by an ICME we refer to the MESSENGER observations and the target observations as a pair. For every pair, we check the data observed at the target around the predicted impact times (t predV500 and t predVmes ) for signatures of an ICME, considering all the datasets which are publicly available on the AMDA web tool (http://amda.cdpp.eu/).
ICMEs have distinct plasma and magnetic field properties which differentiate them from the surrounding solar wind. An ideal magnetic signature of a magnetic cloud (MC) is a force-free field, which corresponds to a steady rotation of the magnetic field vector as the MC is sweeping over the observer (Burlaga et al., 1981). ICME identification also relies on additional parameters such as a low plasma density, a low plasma temperature, high abundances of heavy elements, or typical signatures of energetic particles compared to typical solar wind conditions (e.g., Gosling et al., 1973;Cane & Richardson, 2003;Liu et al., 2005;Malandraki et al., 2005Malandraki et al., , 2007, and references therein). Typically, ICMEs match only some of these criteria (Richardson & Cane, 1993).
Cane & Richardson (2003) noticed that magnetic rotation in ICMEs often does not agree with a force-free field, and hence ''a likely ICME interval can be inferred from reduced fluctuations and some degree of organization in the magnetic field and is bounded by distinct magnetic field discontinuities which may be accompanied by abrupt changes in plasma parameters''.  Winslow et al. (2015) identified ICMEs from magnetic field observations only. They select events with a clear interplanetary shock at the time of arrival, followed by a sheath and a magnetic ejecta region. Assuming that an ICME's magnetic configuration is preserved during its propagation, we proceed in the same way at the target by looking for periods exhibiting a structured interplanetary magnetic field (IMF), preferably including a smooth rotation, and a preceding shock and sheath. When possible, this ICME identification is confirmed with in situ solar wind observations. For each pair, observations at the target are classified as No Data, Void, ICME and IP (cf. Table 1): No Data when no magnetic data are available; Void (VEX only) when ICME shock arrival could be hidden to VEX due to its proximity to Venus (see right panel of Fig. 1 and related text for more explanations); ICME when observing a magnetic ejecta and additional convincing plasma parameters (density, velocity, a-particle to proton ratio, temperature, as previously defined) are present; and IP when the observed interplanetary shock is not followed by any magnetic ejecta. We compared our results with the following ICME catalogs: HELCATS (https://www.helcats-fp7.eu/ catalogues/wp4_icmecat.html), STEREO (Jian et al., 2018) and ACE (http://www.srl.caltech.edu/ACE/ASC/DATA/level3/ icmetable2.htm; described in Richardson & Cane, 2010). We finally check for each pair ICME propagation between Mercury and the target in ENLIL simulations (Odstrcil & Pizzo, 1999) available online (http://helioweather.net/index.html).
We distinguish among the following times and velocities: -t obsMESS : ICME shock arrival time observed at MESSEN-GER (Winslow et al., 2015). t obsMESS is defined for all the records. -t obsTAR : ICME shock arrival time observed at the target.
It is defined for ICME and IP events. -v SunMess : shock transit speed at MESSENGER (Winslow et al., 2015). v SunMess is derived from the time delay between the CME onset at the Sun and its arrival at MESSENGER. We want to emphasize that v SunMess is not the measured velocity at MESSENGER but rather an average velocity of the shock between the solar corona and MESSENGER. It relies on a CME/ICME association. Winslow et al. (2015) do not specify how they select the solar source. v SunMess is defined for all the pairs. -v MessTar : this is a constant ICME velocity which is set in the Propagation Tool. v MessTar fits with both ICME shock arrival times at MESSENGER t obsMESS and at the targets t ob-sTAR . v MessTar is defined for ICME and IP events. v MessTar is the average velocity of the shock between MESSENGER and the target. -v Tar : observed ICME shock velocity by the target spacecraft. v Tar is defined for ICME and IP events, when solar wind velocity measurements are available at the target. -t predV500 : ICME shock arrival time at the target predicted by the tool based on a default propagation velocity of 500 km s À1 between MESSENGER and the target. It is defined for all the pairs. -t predVmes : ICME shock arrival time at the target predicted by the tool based on v SunMess . t predVmes is defined for all the records.
We focus on the ICME shock propagation rather than on the magnetic ejecta itself because the shock timing appears to be more precise and somehow less subjective than the timing of the boundaries of the magnetic ejecta.

Analysis of the predicted hits
Running the tool on the 61 catalog records, 36 pairs are found as detailed in Table 1. There are no magnetic field measurements at target for STEREO-A and VEX No Data events. VEX is close to its periapsis around the predicted hit times in the three Void events. One of these events is described hereafter (right panel of Fig. 1). There are two IPs. Among the 29 ICMEs, three ICMEs are not resolved at the targets because they interacted with other ICMEs between MESSENGER and the targets. Table 2 gathers for each pair, observed arrival times, time mismatches from predicted arrival times, and velocities.
Finally the False category is empty, which requires some comments. An interplanetary shock followed, or not followed, by an ICME was identified ±24 h from t predVmes and/or t predV500 for each pair where interplanetary magnetic field measurements are available. ICMEs are observed at the target with large time delays with regard to predicted times (columns Dt Vmess and Dt V500 in Table 1) in several cases (larger than 20 h for pairs #02, #18, #20, #26, #27, and #36). Pair #02 corresponds to a fast ICME: both v SunMess and v MessTar are high. A default velocity of 500 km s À1 leads for that case to a large Dt V500 value. ICME shocks are strongly decelerated between MESSENGER and the targets for pairs #18, #20, and #27 leading to a high Dt Vmess value. v MessTar and v Tar in pair #26 are close to each other: v SunMess might be overestimated. Pair #36 observations are detailed in Section 5.1.
Beyond Mercury's orbit, CME-CME interactions have implications for ICME propagation (Lugaz et al., 2012). It might concern CMEs following each other at Mercury (in pairs #01-#02; #07-#08; #12-#13; #23-#24-#25). In two cases we were not able to properly identify different ICMEs at the targets (#12-#13; #23-#24-#25). We take into account only one shock for each of these two cases. Table 2. Details of the 36 pairs. A number is assigned to each pair. Pairs are sorted first by target names and then in chronological order. id refers to the event number in the list of Winslow et al. (2015). t obsMESS , t obsTAR , v SunMess , v MessTar , and v Tar are defined in Section 2. Dt V500 = t obsTAR À t predV500 and Dt Vmess = t obsTAR À t predVmes . Note that Dt V500 and Dt Vmess are time differences between observations and predictions. Identified ICMEs and IPs are mentioned in a catalog (see Sect. 2) for 19 of 28 cases (cf. Table 2). As for the 9 other events of the 28, ICMEs have various aspects, from the textbook case to a very short magnetic ejecta. We also mention two ICMEs (pairs #11 and #17) without a sheath. It is quite obvious that the CME-driven shock is larger in longitudinal direction than the ICME itself (see Fig. 1 in Kilpua et al., 2017). ENLIL simulations show that for the two IP events STEREO-B is on the edge of an ICME that went through MESSENGER earlier. IPs are related to ICMEs and we keep the two events in our study of the ICME shock deceleration. These events are important for space weather alerts as, at Earth and probably elsewhere, the sheath of the ICME can be more effective than the ICME itself (e.g., Kilpua et al., 2017;Bocchialini et al., 2018).
We emphasize here that besides Propagation Tool predictions, timing considerations, in situ observations, and simulations, there is no certainty that the ICME observed at MESSENGER is the one observed at the target.
However this is not surprising to have a high number of matches as ICMEs propagate radially, and we also use a loose criterion for the ICME identification. Möstl et al. (2017) found that 25-35% of 1335 ICME arrivals were correctly predicted, assuming a constant speed and propagation direction and a longitudinal extent of 60°. Having MESSENGER as a starting point rather than the Sun can explain the higher rate of correct predictions, because our study involves only ICMEs which propagate inside the ecliptic plane. We also avoid cases when drastic changes in CME propagation occur close to the Sun (Möstl et al., 2015). Figure 1 presents a Correct-pair sample (left part of the figure) and a Void-pair sample (right part). t obsMESS , t predV500 , t predVmes , and t obsTAR are marked by vertical dashed lines (blue, red, green and violet, respectively).
The Correct-pair sample has STEREO-A as a target. Over the 60 h presented in Figure 1, one can notice a single shock observed in both, the magnetic field and the bulk flow velocity (panels a-c). The shock arrival time (22 August 2002 07:05) is t obsTAR (see vertical violet line). Later on (23 August 02:00 UT), one can notice the arrival of the ICME's magnetic ejecta, i.e. the smooth part of the magnetic field. t obsTAR is observed later than t predVmes and earlier than t predV500 , making an identification of t obsTAR easy. The large gap (26 h) observed between t predVmes and t predV500 results from the large ICME shock velocity estimated at MESSENGER (v SunMess ¼ 975 km s À1 ) compared to the Propagation Tool default velocity (500 km s À1 ). The difference between v SunMess and v Tar (540 km s À1 ) can result from a strong deceleration between MESSENGER and STEREO-A or from a bad estimation of the velocity at MESSENGER. Finding v MessTar (625 km s À1 ) between v SunMess and v Tar supports the deceleration assumption.
The Void pair sample has VEX as a target. The three components and the magnitude of the magnetic field are plotted in panels d and e, respectively. The VEX-Venus distance is plotted in panel f. In the vicinity of the three periapsis passages, B fluctuates and |B| intensifies (Fig. 1d and e), obviously due to bow shock crossings. The intensification observed during bow shock outbound is usually the largest (7 and 9 March 2012 periapsis). In the 8 March periapsis case, there is also a large |B| increase at the beginning of the fluctuation period that can correspond to the bow shock inbound. Even looking closer to the data it is not clear if it results from an interplanetary shock or from a bow shock crossing. It occurs close to the predicted shock arrival times (t predVmes and t predV500 ).
Farther from the periapsis, VEX can be considered to be inside the interplanetary medium. There is a remarkable shock visible on 7 March at 13:15 UT, followed by an ICME magnetic ejecta which starts at 20:00 UT. The timing of this shock leads to a mean ICME velocity v MessTar of about 1950 km s À1 , between MESSENGER and VEX, which is very high. As we are not certain that an interplanetary shock reached Venus during the March 8 periapsis, and that the 7 March shock corresponds to the ICME seen at MESSENGER, this pair is classified as Void.

ICME acceleration and deceleration in the interplanetary medium
We considered in that study the 28 ICME and IP events (cf. Table 2). Comparing the average velocities v SunMess (estimated in Winslow et al., 2015), v MessTar , and v Tar allow us to study deceleration (v SunMess > v MessTar ; v MessTar > v Tar ) or acceleration in the interplanetary medium. Mean velocity values for different subsets are gathered in Table 3.
When considering the 15 pairs for which v Tar is available, one can notice that v Tar is systematically lower than v MessTar . The mean v Tar value (653 km s À1 on average, cf. Table 3) is lower than the mean v MessTar (493 km s À1 ), by 160 km s À1 : ICME shocks are slowing down between MESSENGER and the targets. We also note that v SunMess is larger than v MessTar in 22 of 28 cases, and by 171 km s À1 on average: ICME shocks generally slow down between the solar corona and the targets.
We detail now the six cases where v MessTar > v SunMess . For three pairs (#15, #16, and #33), v MessTar exceeds v SunMess by less than 40 km s À1 . These cases are not certain cases of ICME acceleration as these value can be explained by an uncertainty in v: assuming a rigid cone structure of the shock in the radial propagation is probably not realistic. However, #15 and #33 pairs have the lowest v SunMess values of our dataset (419 and 341 km s À1 ). Gopalswamy et al. (2001a) noted that slow ICMEs have a larger probability to be accelerated than the fast ones. In the three other pairs (#28, #30, and #36) v MessTar is larger than v SunMess by more than 100 km s À1 . However, in those cases v MessTar is also larger than v Tar by more than 100 km s À1 which is indicative of a deceleration between MESSENGER and the targets. In these three cases, v SunMess might be underestimated because of a wrong CME-ICME association, thus making ICME acceleration questionable. Bocchialini et al. (2018) discusses the difficulties of correctly associating CMEs with ICMEs. There is no definitive evidence of ICME shock acceleration beyond Mercury's orbit in these six events. Figure 2 presents velocity variations as a function of the Sun-Mercury distance for the slow ICMEs (v SunMess < 600 km s À1 , top panel) and the fast ICMEs (v SunMess > 600 km s À1 , bottom panel). In the top panel, one can note that the six cases when v SunMess < v MessTar (meaning that the ratio of velocities is less than 1) are included in the nine slow ICME cases when an ICME acceleration by the surrounding interplanetary material is more probable. The lowest ratios are observed within 75 R (%0.35 AU). Major speed changes are expected to occur close to the Sun, especially within 50 R (Liu et al., 2016). If both ICME associations at MESSENGER and at the targets are correct, the region with positive acceleration can extent to 75 R . Looking at the influence of the distance of Mercury to the Sun on the v MessTar =v Tar ratio for the fast ICMEs (bottom panel of Fig. 2), strongest deceleration cases are observed below 80 R (%0.37 AU), when the distance of Mercury to the targets is near maximum due to Mercury's highly elliptical orbit.
In order to study shock deceleration between VEX and 1 AU, we sort Correct pairs into two subsets: the VEX subset (10 VEX pairs) and the 1 AU subset (18 STEREO-A and B, and ACE pairs). v MessTar is much higher on average in the VEX subset (831 km s À1 ) than in the 1 AU subset (585 km s À1 ). This alone does not confirm a deceleration of ICMEs between VEX and 1 AU because also v SunMess for the VEX subset is on average %370 km s À1 higher than in the 1 AU subset (1103 km s À1 vs. 731 km s À1 ).
A statistics on the differences between the observed arrival time t obsTAR and the predicted arrival times (t predV500 and t predVmes ) is presented in Table 4, for all Correct pairs. According to the propagation velocity analysis, the standard deviation of the propagation time is quite high (11-12 h). The standard deviation is lower than the 16.6 h found in Möstl et al. (2017). ICME major speed changes occur within 30 R (Liu et al., 2016). Avoiding that area by starting from Mercury's orbit might explain the lower standard deviation found in our data set.
The negative mean value of t obsTAR À t predV500 shows that the default velocity of the Propagation Tool leads to a delay of À5.5 h on average. The slowing down of the ICME after passing MESSENGER leads to an error of 6.5 h on average in the predicted arrival time based on t predVmes . Considering the absolute error, the default velocity and v SunMess predict the arrival time with a similar precision.

Events involving multiple solar system bodies
Three ICMEs (#10, #56, and #59) at MESSENGER are paired with more than one target (see Table 2). ICME #10 is paired with VEX and STEREO-B, and ICME #59 is paired with both STEREO spacecraft, as shown in Figure 3 and detailed hereafter. Figure 4 shows data from VEX (panels a-c) and STEREO-B (panels d and e). At VEX (respectively at STEREO-B) the arrival time of the ICME shock is marked by a dashed red (blue, resp.) line in panel b (e, resp.). VEX encounters the shock near its periapsis. The ICME is seen as an organized and intense magnetic field lasting roughly one day (6 November 2011). The picture is complicated by the proximity to Venus. At STEREO-B, the ICME lasts two days (7 and 8 November). The magnetic field is more organized than at VEX.

ICME W#10: Pairs #06 and #31
It is worth noticing that considering radial propagation between VEX and STEREO-B, the ICME shock velocity is about 590 km s À1 , which is lower than v MessTar at VEX Table 4. Observed versus predicted impact time of the 28 ICME shocks (mean values and standard deviation values).

Dt
Mean r h h t obsTAR À t predV500 À5.5 12.2 t obsTAR À t predVmes 6.5 11.4 Fig. 2. ICME velocity evolution vs. Mercury-Sun distance for ICMEs observed at MESSENGER with a velocity lower than 600 km s À1 (top panel) and larger than 600 km s À1 (bottom panel). Note that plots are ratio of average velocities (top panel) and ratio of average to in situ velocity (bottom panel). (940 km s À1 ) and at STEREO-B (705 km s À1 ), and lower than v Tar at VEX (780 km s À1 ). In that case, it seems likely that the same ICME is observed by VEX and STEREO-B. The different velocities indicate that the ICME's radial velocity is decreasing from MESSENGER to VEX and further decreasing from VEX to STEREO-B. Even if the solar wind velocities are not continuously available at STEREO-B, the data indicate that the shock arrives with a velocity of v Tar % 620 km s À1 . This is quite close to the 590 km s À1 mean velocity obtained between VEX and STEREO-B. The slightly lower value can be explained by the geometry of the ICME: VEX and STEREO-B are not radially aligned.
5.2 ICME W#59: pairs #30 and #36 Figure 5 presents the same ICME seen at STEREO-A (left) and at STEREO-B (right). The ICME is easily identified in STEREO-A data. Sudden IMF strength and solar wind velocity increases mark the shock arrival (11 August 2014 09:00 UT). The ICME's sheath lasts about 1 day. Then, the ICME's magnetic ejecta is observed from 12 August 08:00 to 13 August 20:00. The ICME displays a remarkably stable and rotating IMF and low solar wind density. One can also note the gradual decrease of both IMF strength and solar wind velocity. These properties correspond to a magnetic cloud.
The picture is more complicated at STEREO-B: it is difficult to identify a shock arrival time. We choose 11 August 05:50 because of concomitant jumps in IMF strength and solar wind velocity. The ICME magnetic ejecta starts on 11 August at 09:00. The IMF is less structured and fluctuates more contrary to STEREO-A observations. Furthermore, the solar wind density is low for a few hours before the IMF's ending on 14 August, 10:00.  A striking feature is the difference in observations from STEREO-A and B: the ICME magnetic ejecta is observed at STEREO-B earlier than at STEREO-A, even though the latter is closer to the Sun. Accordingly, ICME and shock velocities are larger at STEREO-B (>620 km s À1 ) than at STEREO-A (<500 km s À1 ). However, the magnetic ejecta signature is very confused at STEREO-B.
It is thus reasonable to ask whether the two spacecraft observe the same ICME. Looking at the Sun, STEREO-A and B coronagraphs observe only one CME on 8 August, and none on 9 or 10 August (https://cor1.gsfc.nasa.gov/catalog/cme/2014/ HongXie_COR1_preliminary_event_list_2014-08.html). This CME, also seen in SOHO/LASCO coronagraphs, is listed as the source CME of the ICME observed at MESSENGER. LASCO observes only two minor events prior and after this CME (https://cdaw.gsfc.nasa.gov/CME_list/UNIVERSAL/ 2014_08/univ2014_08.html). They come from a different source location. Due to the weak solar activity during those days, it is unlikely that STEREO-A and B were hit by two different ICMEs. We thus believe that STEREO-A is encountering the ICME at a place where it is still well structured while the ICME part encountered by STEREO-B experienced more interactions with the ambient solar wind. The larger velocity seen at STEREO-B might be the result of an interaction with a highspeed stream in the ambient solar wind that is affecting only the part of the ICME which is finally sweeping over STEREO-B.
This example demonstrates the azimuthal variations observed in an ICME at two different locations which are separated by less than 40°in azimuthal direction. This finding agrees with a study by Good & Forsyth (2016) who noticed a reduced number of ICMEs observed by two spacecraft when the angular separation is larger than 30°.

Conclusion
Based on 61 ICME events observed at MESSENGER, the Propagation Tool predicts an impact at VEX, STEREO-A, STEREO-B, and ACE in 36 cases. Out of these, 28 events are suitable for a comparison of v SunMess and v MessTar . ICME shocks are clearly decelerated in 22 cases, unlikely accelerated in three cases, and in three cases the velocity remains almost constant. The same ICME observed at MESSENGER is paired with two different targets in three cases. Different signatures of the same ICME are observed at 40°angular separation.
In the 13 pairs with available v Tar , ICME shock velocities always decrease between MESSENGER and the targets (v MessTar > v Tar ). The average speed difference is %160 km s À1 . With respect to the mean Sun-Mercury transit velocity, the ICME speed decreases by about 300 km s À1 on average (v SunMess À v Tar ). Freiherr von Forstner et al. (2018) studied a similar number of events (15 vs. 13 in our study) with similar mean v Tar values (474 vs. 475 km s À1 ). They noted a 170 km s À1 average deceleration between the solar corona and 1 AU and a 60 km s À1 average deceleration between 1 and 1.5 AU (Mars orbit). Our results show that ICME deceleration is still strong between Mercury and 1 AU. We believe the deceleration observed in the present study results from the accumulation of the solar wind plasma upstream of the ICME. Emission of (magneto)hydrodynamic waves in this collisionless environment causes the drag between ICME and solar wind (Cargill et al., 1996;Vršnak et al., 2013), as already observed (Lindsay et al., 1999;Gopalswamy et al., 2000).
We do not follow ICMEs all the way between MESSENGER and the targets. Our ICME/ICME associations rely on timing, ICME identification and on global simulations. Our dataset suggests that ICME shock deceleration between MESSENGER and the targets is the common case. There is no convincing observations of ICME shock positive acceleration since cases with v MessTar > v Tar are not part of our dataset. Potentially most accelerated ICME shocks (v SunMess =v MessTar < 0:9) and most decelerated shocks (v MessTar =v Tar > 1:8) are observed at MESSENGER when Mercury is located between 65 and 80 R (%0.30 and %0.37 AU). Major speed changes can thus occur in a larger region than the 50 R limit found by Liu et al. (2016).
Evidence of ICME deceleration between VEX and STEREO-B is found based on ICMEs observed by MESSENGER, VEX and STEREO-B. Our results suggest that ICME deceleration would occur as long as ICMEs propagate faster than the surrounding solar wind as noted by Witasse et al. (2017) at a larger distance.
We conclude that the CDPP Propagation Tool is helpful in studying ICME propagation between bodies in the inner solar system. The predicted arrival time is, on average, correct within a À5.5 ± 12.2 h margin. An ICME shock is observed at a target in 31 of 36 pair predictions. The lack of events classified as False follows from our choice to reject observations inside Mercury's orbit where drastic changes in ICME parameters are more frequent. The proportion of ICMEs observed at 1 AU that have not been observed at MESSENGER is left for a further study. By means of real-time in situ data obtained between the Sun and the Earth, uncertainties in shock arrival times can be reduced and predictions for ICME propagation can be improved in general. In our study the shortest propagation time between Mercury and VEX is about 8 h and 24 h between Mercury and 1 AU. Using spacecraft like MESSEN-GER as monitor for extreme solar events shortens the time to communicate an updated alert status as compared to observing directly the activity on the solar surface. However, it enables a more accurate prediction of propagation path and geoeffectiveness of a potentially hazardeous ICME. Our study demonstrates the utility of having in situ measurements between the Sun and Earth in an operational space weather context. In the future, ESA's Solar Orbiter, ESA-JAXA's Bepi-Colombo, and NASA's Parker Solar Probe missions will provide more opportunities to observe ICMEs at different locations in the inner solar system.