Mobile phone data is an important asset in understanding human mobility, widely investigated and used in different fields. Especially in the urbanized developed countries, mobile phones have become a real proxy for their human users. Therefore, the idea of using mobile phone data to understand the impact of the Covid-19 pandemic on urban mobility, and of the sanitary constraints associated with it, imposed itself as an evidence. In the SafeCityMap project, we used mobile phone data provided by two different industrial partners: SFR and Roofstreet. The two datasets cover a geographical region focused on the city of Paris, for a time period in early 2020, before the first French lockdown, and a second time period during the lockdown. Besides the city of Paris, the Roofstreet data also covers the geographical regions of three neighboring departments of Paris. The two data sources are different (a mobile operator and a platform providing location-based services), but they use mobile phones to localize users, to measure attendance in certain areas and mobility between areas. Practically, to briefly summarize the differences between the two datasets, we can say that the Roofstreet data is more fine grained geographically, but the number of users and their density is significantly lower in this dataset. The reference areas are at the size of the INSEE IRIS zones (992 such zones in the city of Paris) in the Roofstreet dataset (Figure 9), while the SFR datasets further aggregates the IRIS zones in 326 areas over the city of Paris (Figure 1). We first conducted an analysis of the two datasets, to understand global trends and properties. Both datasets show a significant drop (division by 3-4) in the number of observed users in the zones once the lockdown is established, as shown in Figure 2 and Figure 10. In the SFR data, for example, two zones lost 94% and 70% of their attendance compared to a normal-activity time (Figure 8(a)). The events that involve user movements seem to be particularly impacted by the lockdown: the top attendance values in terms of number of flows during the lockdown barely reach the mean values in normal times (Figure 12). However, despite this significant reduction in the number of observed user flows, we can still observe some hotspots on the map of Paris, as visible in Figure 6(b) and Figure 18(b). Interestingly, in the SFR data, some areas in Paris (mainly located at the frontier with other departments) even show an increase in the number of observed flows during the lockdown (Figure 8(b)). To further study the impact of the lockdown, we build on previous work we conducted and we define a so- called signature for each area. These signatures are then clustered using an unsupervised machine learning technique, with the objective of detecting geographical zones with similar behaviors in terms of user atten- dance. By applying this methodology, we find two major classes (i.e. behaviors) in the SFR data and three major classes in the Roofstreet data. In each case, a large number of unique patterns (i.e. classes containing 1 or 2 zones only) is detected as well. By comparing the obtained classes with INSEE land use data, we label them as Activities, Residential, and Others, plus a specific Transit class that appears in the Roofstreet data. The representation of these different classes on the Paris map is shown in Figure 31 (for SFR) and in Figure 34 (for Roofstreet). In the SFR data (Figure 31), we notice that most areas in the city become Residential once the lockdown is established, but a few Activities and specific areas still exist, even during the lockdown. Therefore, we zoom in and analyze the evolution of the signatures to better understand what is happening in areas not showing a residential pattern during lockdown. We discover that some of them are located in leisure areas (e.g. Jardin des Tuileries), former activities areas without a residential component (e.g. Opera Bastille, university cam- puses), train stations, places of power, hospitals, and EHPAD areas. The signatures clearly show the human attendance patterns in these areas and how they changed with respect to a normal period. Nevertheless, the signatures do not directly contain any mobility-related information. We therefore take a complementary graph-based approach that also allows to follow mobility flows and their temporal dynam- ics. Practically, after building a time-varying weighted mobility graph representing urban zones and their neighbor-zone connectivity, we study three different types of graph centrality, which quantify the impor- tance of each zone in the graph according to the habits in mobility of the population: i) the betweenness centrality captures the tendency of people to follow shortest paths, ii) the closeness centrality captures the locality of people movement, and iii) the degree centrality captures topologically central hubs. The analysis of the betweenness centrality before and during the lockdown allows us to detect a signifi- cant change in the paths mostly preferred by the population. While the central Paris areas were highly ranked in terms of betweenness centrality before the lockdown, downtown areas are outlined during the lockdown (Figure 49). When considering the neighboring departments of Paris (which relativizes the importance of downtown Paris), we see in Figure 50 that, in normal times, the top-ranked zones are located in the north and northeast business areas outside Paris (Saint-Denis and Bobigny) as well as in areas with big parks in Paris (Boulogne and Vincennes). Interestingly, such zones are still top ranked during the lockdown and some big park areas in Paris even show an increase in the number of observed flows. The decentralization during lockdown can also be observed in terms of closeness centrality, which cap- tures how localized population mobility is. Indeed, closeness results reveal a noticeable reduction in the flows through the zones within Paris, while the Eastern region of Paris becomes a major hub during the lockdown (Figure 52). This is an indication of the shorter displacement of the population, consequence of the lockdown distance restrictions. The last metric, the degree centrality, logically shows similar trends before and during the lockdown, since the topology of the city does not change in this short time period. A particularity of degree centrality is the possibility to capture the attracting and dissemination power of zones regardless the total number of flows or the size of the areas. Therefore, results from the two data show lockdown mobility restrictions do not impact the connectivity of zones in normal times (Figure 53). We observe that some Roofstreet data centrality results are different from SFR ones due to dataset dif- ferences in terms of covered population, considered departments, and geographical granularity. Such differ- ences bring the benefit of isolating mobility behaviors restricted to the city of Paris, as well capturing changes when surrounding departments are considered. Generally speaking, results demonstrate that human mobil- ity changed shape (i.e., it became more local), but it has not disappeared during the lockdown, although a significant reduction in the flow of people circulating between different zones is observed. Finally, we consider that the three centrality metrics discussed above capture complementary informa- tion with respect to human mobility. Since human mobility is clearly linked with the epidemics propagation, we use the three centralities to compute a specific metric per zone of the city. We denote this as impact- factor, shown in Figure 55 and Figure 56, and use it to quantify the global importance of zones according to the mobility behavior of the population. We notice that zones being highly impacted by normal population mobility habits are mainly located in the central Paris. In SFR data, such highly impacted central-zones have their impact-factor value decrease during lockdown, but still appear as the most important zones in the city of Paris. On the other hand, when considering transiting flows between Paris and neighborhood departments with Roofstreet data, we notice central zones become less important during the lockdown and a shift in terms of impact-factor values appears at the north-border zones in Paris (Figure 56(a)). Business zones and parks still concentrate the zones with highest impact factor in normal and lockdown periods (Figure 57). We consider the impact-factor metric can give an insight regarding the urban areas noticing a relatively significant mobility level, hence an important role in controlling the epidemics. Still, this claim needs to be investigated from the epidemiological point of view and constitutes not-validated intuitions, which is left for the 2nd-phase of SafeCityMap. All current results represent daily computations of the mentioned features, while investigation using a more precise time window (i.e., slots of 2h, 4h, or 6h) is also left for future.