The region of Marrakech-Safi extends over a very diverse territory in central Morocco, for a total of 38,167 km² (Haut Commissariat au Plan, 2020). The climate of the region is predominantly semi-arid. The Atlas mountain range is the source of a large number of oueds (i.e., seasonal creeks), the largest of which is the Tensift. Several plains, piedmont and lowlands areas occupy large extents of the region and are intensively utilized for agricultural production (irrigated and rain fed). The region is administratively divided into seven provinces—El Kelaa des Sraghna, Rehamna, Al Haouz, Chichaoua, Essaouira, Safi, Youssoufia—plus the prefecture of Marrakech. There are 251 municipalities in the region, with the municipality of Marrakech being further divided into five arrondissements (i.e., districts). The municipal level was chosen to define the spatial resolution of the current analysis, which consists therefore of 255 different units. With 4,520,569 inhabitants in 2014, Marrakech-Safi is the third most populated region in the country (Haut Commissariat au Plan, 2020). The city of Marrakech alone accounts for more than 20% of the region's population, while 57% of the inhabitants live in rural settlements (Haut Commissariat au Plan, 2020). More than 42% of the regional workforce is employed in agriculture (Haut Commissariat au Plan, 2020), and overall the region is responsible for up to 75% of national agricultural exports¹. In recent decades, the region has experienced altered precipitation patterns and increased mean temperatures (Messouli, 2015; Fniguire et al., 2017; Choukrani et al., 2018), leading to several droughts, with progressively shorter intervals between events (Ezzine et al., 2016). Fniguire et al. (2017) showed that at higher time scales (12 and 24 months intervals), significant drought events (defined as rainfall deficiency) became more frequent and of longer duration after 1975. Analyzing agricultural yields, Messouli (2015) showed a correlation between meteorological droughts and reduced cereals output in the region between 2000 and 2015 (especially in the years 2006–2008). In particular, rain fed agricultural systems (which are mostly small scale subsistence farms, with little or no access to groundwater or reservoirs and low levels of technology and of market integration, World Bank, 2018) are characterized by an extreme sensibility to drought events of even short duration (Ezzine et al., 2014; Hadri et al., 2021a). The high seasonality of precipitation in the region is also a contributing factor to the occurrence of floodings. Ezzine et al. (2016) report that between 1982 and 2015 there have been at least 17 flood events that have caused severe human and material losses in the region, with at least one event per year since 2008. Arguably, the most devastating flood event occurred on August 17th 1995, when an exceptionally strong flash flood in the Ourika Valley (located in the High Atlas), claimed the lives of more than 200 people (Bennani et al., 2019). Figure 1 shows the main geographical features of the region, namely its land use-land cover and administrative divisions (A), and the superimpositions between flood and drought hazards extents and a set of assets of interest, namely population (B), road infrastructure (B), rain fed farmlands (C), and irrigated farmlands (D).

The definition of the scope for a multi-risk assessment requires terminological clarification. In recent years, the risk science community has been widening the scope of risk assessments to accommodate for complex settings. For instance, under the umbrella term of “multi-hazard”, numerous studies have tackled risk in the context of multiple hazards, either in the sense of hazard interactions and cascades (Gill and Malamud, 2016) triggering hazards, i.e., hazards linked by causal relationships (Kappes et al., 2012a), compound hazards, i.e., two or more independent hazards occurring at the same time or in rapid succession (Zscheischler et al., 2020) or even interpreted as “cumulative” hazards, meaning all dependent and independent hazards potentially affecting a territory, also at times referred to as all-hazards-at-place approach (in reference to a concept introduced by Hewitt and Burton in 1971—Gill and Malamud, 2014). For more complete overviews of hazard interactions, see Pescaroli and Alexander (2018) and Gill and Malamud (2014). Somewhat evolving from and complementing the multi-hazard perspective, recent years have seen the introduction of the notion of “multi-risk” analysis, which shifts the focus from multiple hazards to the presence of multiple risks and related potential losses (Schmidt et al., 2011). However, agreement on a precise definition for multi-risk has yet to emerge (Gallina et al., 2016; Curt, 2020). Zschau (2017) reserves the term multi-risk only for assessments which adopt a multi-hazard risk framework that also consider all possible interactions and dynamics in the hazard and vulnerability components. Gallina et al. (2016) suggest that the concept of multi-risk should revolve around the sub-components of multi-hazard and multi-vulnerability, the latter being a measure of the different vulnerabilities of multiple exposed elements. Moreover, they introduce a distinction between “multi-hazard risk assessment” and “multi-risk assessment:” the first is a combination of a multi-hazard analysis on one side and the sum of existing vulnerabilities on the other, an example of which can be found in Depietri et al. (2018), where the vulnerability indicators considered are hazard-independent. The “multi-risk assessment” approach, on the other hand, explores interactions between hazards and vulnerabilities at the single risk level before proceeding with the multi-risk aggregation. Despite the diversity of terminologies and methodological approaches, researchers insist on the usefulness of multi-risk assessments for policy-making: as remarked by Scolobig et al. (2017), governance of risk can benefit from the adoption of a multi-risk perspective since it can provide a holistic view of risks interactions and impacts. Moreover, a comprehensive understanding of all hazards and risks affecting a specific area is recognized as an essential prerequisite for effective risk management (UNDRR, 2019; Hagenlocher et al., 2020). In particular, in the specific case of drought and flood events, Ward et al. (2020) have shown how risks connected to these extremes of the hydrological cycle are linked not just in terms of hazard propagation, but also in the conditions of socio-economic vulnerability that can cascade from successive events. Therefore, effective risk management practices should acknowledge the relationships and dependencies between multiple hazards and risks, as hazard-specific Disaster Risk Reduction measures could potentially increase risks connected to other hazards, a condition that Ruiter et al. (2021) have termed “a-synergies.” In their assessment of urban exposure and vulnerabilities in multi-hazards environments, Galderisi and Limongi (2021) elaborate further on the role of multi-risk information for risk management, suggesting that it could provide a much needed input to overcome the structures in which risk management actors operate, often isolated in sector-specific silos with little opportunity of cross-information. Somewhat reminiscent of the all-hazards-at-a-place perspective, they posit that multi-risk analysis should be “spatial-centered,” i.e., focusing on a particular geographical unit of analysis and all possible risks it might be subjected to, rather than “hazard-centered” (which they argue implies skewed attention toward hazard' extent and magnitude as a parameter to define an area of interest).

For the current study, multi-risk is interpreted as a cumulative combination of relevant single-risks at the municipality level, and whose sub-components of hazard exposure and vulnerability are assessed through a semi-quantitative approach.

The operationalization of the multi-risk assessment for the present study builds on the IPCC AR5 framework, which represents risk as the interaction between hazard, exposure and vulnerability (IPCC, 2014), a conceptualization that was recently confirmed in the IPCC AR6 report (IPCC, 2022). We first identified the most relevant risks connected to drought and floods in the study areas through a review of scientific literature and reports covering the impacts of past events. The literature review included peer-reviewed articles and international, national and local reports with a focus on flood and drought risk. The review was continuously expanded from a core set of studies identified through a systematic review on the Scopus database in April 2019 (37 sources; for additional information on systematic literature review see Supplementary Material), to which additional sources were added through snowball searches (46 sources). Table 1 reports the four types of single-risks that were found to be particularly relevant for the case study area.

During the course of a dedicated workshop with 31 local experts and stakeholders (held in October 2019 in Marrakech), drivers of risk were discussed and identified for each single-risk, following the “impact chains” methodology (adapted from Fritzsche et al., 2014; Hagenlocher et al., 2018a; Zebisch et al., 2021). Impact chains are analytical tools designed to highlight the relational nature of drivers of risk through a participative process and have been increasingly used in climate risk assessments at various spatial scales in Europe (Buth et al., 2015; Greiving et al., 2015; Arabadzhyan et al., 2020; Lückerath et al., 2020), Benin (Wetzel et al., 2022), Bolivia (Zebisch et al., 2021), Burundi (Schneiderbauer et al., 2020), Pakistan (Zebisch et al., 2021), and Morocco alike (GIZ, 2014a,b). This exercise resulted in the creation of draft conceptual models describing the interrelationships between drivers of risk, which were subsequently complemented by additional literature research and bilateral consultations (Cotti et al., in progress): moreover, the models were used to inform the selection of indicators of hazard exposure and vulnerability for the multi-risk assessment (see Sections Multi-risk Hazard Exposure and Multi-risk Vulnerability). While a large number of drivers was identified (ranging from ecosystem services, socioeconomic factors and governance issues), spatial indicators for the semi-quantitative assessment at the desired level of analysis (i.e., municipalities or provinces) were only available for a subset of them. Table 2 summarizes the drivers of vulnerability and exposed elements which could be included in the assessment and their applicability to the four single-risks considered for the study. The full list of indicators (including data sources, spatial resolution, and reference year) is available in the Supplementary Material.

The spatial indicators of hazard exposure and vulnerability were collected and used to inform the aggregated sub-components of multi-risk hazard exposure and multi-risk vulnerability.

The multi-risk spatial assessment is based on the single-risks drivers identified by the local stakeholders and complemented by literature review. Following this approach, a number of drivers were found to be common for all four single-risks considered: while the complete list of shared drivers is extensive (Cotti et al., in progress), for the current application we will discuss only those which could inform the spatial assessment (i.e., drivers that could be represented by spatially-explicit indicators), as per Table 2. Among these, the driver “presence of early warning systems” was recognized as relevant for both drought-related and flood-related risks. The importance of flood early warning systems was made apparent in the region after the Ourika flood of 1995 (see above), which lead to the installation of a Flood Forecasting and Warning System (FFWS) in the basin, later expanded to five additional neighboring watershed (Louodyi et al., 2022). Louodyi et al. (2022) report that the presence of the FFWS has positively affected the awareness of the local population toward floods. However, the system is currently not implemented for the remainder of the region. For drought-related risks, several public institutions in Morocco monitor different indicators of hazard (such as precipitation levels) or impacts (such as vegetation conditions), but the information thus produced is still not efficiently made available to farmers and stakeholders (World Bank, 2018). Related to this, the driver “access to information” was also found to be relevant for all types of risk, indicating that the production of timely information on hazards and risk on one side must be supported with the increase of people's access to it on the other. The remaining drivers shared across all risks hint at more deeply-rooted causes of socio-economic marginality, i.e., “financial poverty” and “education levels”: these can lead to increased vulnerability in relation to the different risks here considered, e.g., in terms of lower capacity to recover financially after a flood-related loss or the lack of knowledge about drought-effective agricultural practices for farmers. In fact, while Morocco has made important steps toward the eradication of poverty and social disparity in recent decades, multiple pockets of disparity are still present, especially in rural areas (World Bank Group, 2017).

By incorporating data and information from a variety of different sectors and involving multiple stakeholders in the design phase, the multi-risk assessment here presented can support the efforts of creating a multi-sectoral approach to risk for Marrakech-Safi, the lack of which was identified by international analysts as one of the key limitations of current risk management practices in Morocco (World Bank, 2013; Kahime et al., 2018). The multi-risk analysis of the Marrakech-Safi region has shown that every municipality of the region is facing multiple risks associated with floods and droughts, but also that the contributing factors vary extensively across the territory. The various levels of disaggregation of the results by single-risks and by risk sub-components (i.e., hazard exposure and vulnerability) provide informative insights into the multi-risk profile of every municipality (see Supplementary Material for full results at the municipality level). The widespread presence of agricultural assets (i.e., farmlands) and the large extent of drought hazards determine a strong influence of drought-related risks on the overall multi-risk profile, and in particular on the hazard exposure component. This is a well-known challenge in assessments dealing with multiple hazards, each of which might be characterized by different spatial and temporal scales, and therefore assessed through different metrics (also in terms of related losses; Kappes et al., 2012a). In our study, while the influence of drought-related risks is dominating for the hazard exposure component, this is in line with the actual predominance of agricultural land uses in the regions (both rain fed and irrigated farming), whose loss or reduction in productivity due to the occurrence of drought would have severe economic consequences for the livelihoods of a consistent portion of the population. The predominance of rain fed agricultural systems in the region is reflected in their large share in shaping the overall multi-risk scores for the majority of municipalities. However, the influence of irrigated agricultural systems is notable for those parts of the region which host them (most notably, the provinces of Al Haouz and the prefecture of Marrakech): this is of particular interest as, according to the hazard analysis, similar return periods for a drought event capable of affecting irrigated systems (SPI12) can be expected in other areas of the region currently not developed for irrigated agriculture, such as the south-west (province of Essaouira). Conversion to irrigated agriculture has been heavily promoted and supported by the Moroccan government in the last decade, mainly through the Plan Maroc Vert (2008–2019), a strategic policy plan designed to address the multiple challenges facing agricultural and rural development in the country through the introduction of modern techniques and support for the creation of agricultural enterprises, with the overall goal of developing an export-oriented agriculture sector capable of leading the country's development (Faysse, 2015). Aiming to overcome the historical dualism between unequipped subsistence farming and modern agricultural exploitation, the plan has in fact been largely successful in increasing the number of agricultural enterprises using irrigation in Marrakech-Safi, increasing the total irrigated surface from 16,900 ha in 2008 to 99,983 in 2018². Paradoxically, however, this might have contributed to a further depletion of the country's groundwater resources, as overall water demand increased (Molle and Tanouti, 2017), against a consistent reduction of precipitations in the same period (Messouli, 2015).

The results must also be interpreted in consideration of the limitations in the available datasets and methodology. A complete multi-risk assessment would theoretically follow an all-hazards and all-risks approach, i.e., taking into consideration all possible hazards and all risks (i.e., types of negative consequences) connected to them in a given territory. However, such comprehensive analysis was beyond the scope and possibilities of the current work. The availability of data is also a major potential hurdle to the development of a multi-risk assessment (Mignan et al., 2014): in our case, the necessity of performing a regional assessment constrained the analysis to datasets that offered completeness and coherency for the whole territory, thus limiting the possibility of representing all drivers identified as relevant by the stakeholders during the impact chain construction and focusing only on those who could be spatially quantified. Moreover, the vast scale implied at times renouncing to higher levels of details that might be available at more local level. In particular, the available meteorological data did not allow the consideration of flood events generated by intense, very localized and short-lived precipitation phenomena, which are the origin of flash flood events that have historically plagued the region. The flood hazard modeled for the current study is therefore more akin to a riverine downstream flood, i.e., large-scale events caused by precipitations of long duration increasing the runoff of tributary basins which in turn accumulate into a major and progressive rise in discharge of the mainstream water course (Keller et al., 2012). In the case of Marrakech-Safi, the Tensift river may be subject to this process, given its dependence on several tributaries and large downstream river bed (Louodyi et al., 2022). In the case of a 100-yr flood event, the model used in this analysis shows that this type of flood can potentially cause major effects on the growing urban settlements in the floodplains of the major watercourse of the region. However, the most destructive floods in the region have predominantly originated from intense precipitations occurring locally into narrow basins with extremely steep slopes, thus producing exceptionally fast accumulation of rainfall and large discharge (Daoudi and Saidi, 2008; Bennani et al., 2019; Louodyi et al., 2022). Furthermore, flood exposure results is also conditioned by the population distribution model used: in particular, Smith et al. (2019) warn that the use of the WorldPop dataset might generate overestimation of flood exposure, especially in rural areas, due to the fact that in the model built-up areas are not masked: this means that even pixels that do not belong to areas classified as urban will receive a population prediction, albeit very small, and no pixel will receive a value of zero (equal to absence of population; Smith et al., 2019). While this can create uncertainty, on the other hand a population distribution model that strictly confines the presence of people to the built environment may introduce an equally strong assumption on their exposure to flood, making it dependent on their proximity to buildings. This seems to be the case with historical flash floods in Marrakech-Safi: while injuries and physical harm have also been caused by collapses of houses and other buildings (Ezzine et al., 2016), the deadliest events concerned people gathered in outdoor areas such as in the tourist facilities in Ourika. These exposure dynamics are not captured by any population distribution model currently available for Marrakech-Safi, and their absence limits the ability of the risk assessment to exactly predict the extent of possible impacts.

The case study here presented provided also a testbed to explore the opportunities, benefits and challenges of expanding common methodologies for risk assessment toward multi-risk. While the selected approaches (i.e., spatial analysis for the hazard exposure components, construction of a composite indicator for the vulnerability component and matrix-based aggregation for risk) are common in the risk assessment literature, in this application they were employed to capture a more complex dimension of risk. In particular, our approach was aimed at constructing a framework in which both multi-risk and single-risks can be taken into account, by emphasizing the interaction of sub-components of risk at the single-risk level before proceeding to create the aggregated output. This approach has the advantage of offering a more detailed characterization of multi-risk for the units assessed. In particular, the use of combination-specific libraries of vulnerability indicators together with combination-specific weights offers a sophisticated angle to differentiate how vulnerability varies across space both in terms of relevant drivers (which ones apply in each unit?) and in terms of their relevant performance (which ones contribute the most to the final score?), thus providing a more layered set of information to policy-makers. Moreover, to compensate for this additional complexity, the choice of constructing the multi-risk assessment around a set of single-risks was deemed suitable also in terms of interactions with stakeholders: in fact, the single-risks perspective allowed experts to present their in-depth insights and local knowledge, and at the same time inform the underlying multi-risk framework without presenting an exceptionally abstract (and potentially alienating) framework of reference. This approach could be of particular interest from a policy perspective, as it offers the possibility of representing a multi-risk dimension while at the same time preserving the possibility of decomposing it back into the single-risks which are its constituents. In other words, it can offer to scientists and policy-makers the possibility of understanding single-risks under a wider framework, allowing comparison and therefore prioritization of risks, a characteristic considered an important added value by researchers and practitioners (Zschau, 2017). On the other hand, this approach did not allow to capture stakeholders' and experts' perception on the importance and priority of each single-risk against the others, which could offer additional insights on the construction of the multi-risk framework (Carpignano et al., 2009). From a policy perspective, the “cumulative” single-risks approach might also offer a practical entry point of interest, since policy-makers' mandates are often associated with defined administrative units. The use of composite indicators for the multi-risk vulnerability component, however, might prove quite complex to disentangle, especially in presence of variable weighting schemes: in fact, the comparison of the contribution of single-risk vulnerabilities to the final multi-risk vulnerability across municipalities can be more easily assessed when equal weights are applied, as it would only require to select the relevant indicators and compare their scores. However, weights are an important component of composite indicators (Becker et al., 2017), and especially so when they aim to capture the opinions of local experts and stakeholders, thus introducing a judgement value to the construction of the final composite indicator (Saltelli, 2007). It remains therefore an open question how best to match the complexity of single-risks combinations with experts' judgement in the context of multi-risk. In the field of indicator-based assessments, certain researchers even argue for avoiding the aggregation into a single output, and rather promoting the comparison between different components (Saltelli, 2007): for multi-risk applications, however, the multiple layers of aggregation involved could prevent from obtaining a synthetic overview based only the single, disaggregated components. An alternative path to explore, however, lies in the relational conceptual models constructed for the single-risks, i.e., the impact chains, which offer a map of interconnections between drivers of risks: in fact, the characteristics of multi-risk (including connections between drivers) are more fully represented in the conceptual models rather than in the spatial analysis, which is constrained by the non-relational nature of the index construction process. However, the weighting scheme employed for the index partially represents interactions by calculating different scores according the combination of risk (instead of for each risk separately). While for this application they only informed the selection of relevant indicators for the static composite indicator, future applications should explore modeling approaches to correctly represent the complex relationships between drivers of risk in a multi-risk context, expanding on recent efforts at the single-risk level (Wetzel et al., 2022).

Moreover, given the complexity of the scope of the multi-risk assessment in general, future applications should also consider additional interactions with experts and stakeholders in both the design and the validation phases of the analysis, so as to identify what output can best support the decision-making process.

The multi-risk assessment methodology applied for the Marrakech-Safi region has shown the potential benefits of employing a multi-risk framework based on a selection of relevant single-risks. Stakeholders and experts' consultations can draw on concepts close to local knowledge and experience, while at the same time contributing to a more complex all-encompassing framework. The semi-quantitative assessment is based on a simple methodology, replicable for other data-scarce environments. However, further research is required. In particular, additional stakeholders' validation of the final products (including the aggregation matrix used and potential prioritization between different risks) could provide additional fine-tuning of the results. Of extreme relevance, moreover, is an in-depth assessment of policy-makers understanding of the multi-risk framework and its actual utility in informing disaster risk reduction measures.