Heading toward a famine: modeling seasonality of malnutrition in Yemen, 2018–2023

Abstract

Yemen continues to experience one of the most severe child malnutrition and food insecurity crises on the planet. Seasonal synchronization of malnutrition often serves as an early marker of emerging famine. In this study we compiled data collected by the humanitarian agencies on several potential drivers of malnutrition such as food insecurity, conflict events, and environmental indicators (rainfall, temperature, and normalized difference vegetation index) in Yemen from 2018 to 2023. We characterized the seasonality of malnutrition outcomes and explore spatiotemporal synchronization using non-linear time series analyses. Our findings demonstrate significant seasonal profiles in child malnutrition, which could be synchronized with the peak in food insecurity and peaks in environmental indicators. The quantification of seasonal characteristics in term of seasonal peaks and their amplitude pave the way to build strong foundation for deeper understanding of famine evolution from a complex system perspective by defining processes and triggers that amplify malnutrition and escalate faming formation.

Supplementary Information

The online version contains supplementary material available at 10.1057/s41271-026-00649-2.

Key messages

• We demonstrated significant seasonal patterns in child malnutrition in Yemen in 2018–2023 and the potential for spatiotemporal synchonsization of factors that could amlify seasonal malnutrition.

• The emergence of harmonized datasets and ensemble modeling approaches could further refine the malnutrition seasonality modeling.

• The insights into the dynamic processes governing crises have important implications for understanding, modeling and forecasting seasonal acute malnutrition and subsequent famine formation, evolution, and prevention.

Supplementary Information

The online version contains supplementary material available at 10.1057/s41271-026-00649-2.

Introduction

The United Nations sustainable development goal (SDG) 2, which aims to achieve ‘Zero Hunger’ by 2030, seeks to end all forms of hunger and malnutrition. However, progress toward this goal has been slow [1–3], particularly in conflict-affected regions like Yemen [4]. Yemen is currently experiencing one of the highest child malnutrition rates in the world, with over 2.7 million children under five, which is more than 50%, suffering from acute malnutrition [5]. The ongoing civil war dividing the country into two regions with different levels of restrictions and financial hardship, compounded by the COVID-19 pandemic, has worsened Yemen’s malnutrition crisis [6–8], which was already fueled by environmental stressors [9–11], disease outbreaks [12], economic hardship [13], and persistent food insecurity [14].

Yemen continues to face one of the largest humanitarian crises in the world [5] and is viewed as at a high risk of famine. However, previous research has not sufficiently explored the seasonal oscillations of malnutrition and their role in famine formation. In relation to famine, the understanding of malnutrition seasonality is important because it allows us to establish a seasonal baseline for excess mortality that characterizes famine, gain insight into the spatiotemporal synchronization of processes that govern famine formation, and define the direct and indirect drivers contributing to famine escalation [15].

Seasonality, or systematic periodic annual oscillations, is a crucial yet often overlooked factor of child malnutrition [16–18]. The previous research has indicated a strong likelihood of seasonal fluctuations in regions where child acute malnutrition is prevalent and persistent [16]. While a few studies from other countries with high malnutrition burdens have explored how seasonal variations in acute malnutrition are related to conflict and environmental factors [19–22], there remains a limited understanding of the seasonality of malnutrition in Yemen.

To better understand the synchronization of malnutrition drivers, as potential crisis precursors, we adapted the general framework by Young et al., widely adapted in development settings to reflect seasonal shocks and categorize drivers into three levels: basic, underlying, and immediate (Fig. 1) [23]. This framework enables us to assess malnutrition from multiple perspectives (from individual and household factors to community, national, and global forces), systematically compile and harmonize data from various sources, and characterize seasonality of malnutrition, as a systematic periodic oscillation in an outcome of interest over the course of a calendar year.

Fig. 1.

A conceptual framework of acute malnutrition and factors contributing to seasonality of malnutrition. This framework classifies drivers of malnutrition into immediate, underlying, and basic levels [23] that can be aligned to sources of data at global, regional, local, communal, and individual levels

Exploring spatiotemporal synchronization, or co-occurrence of events and processes across both space (different locations) and time (different points in time) allowed us to deeper understand the interconnections of seasonal patterns in malnutrition and its drivers and derive the implication of spatiotemporal synchronization for famine modeling. In our early work, we introduced the concept of disease synchronization and argued that in complex system sciences, synchronization is defined as a system’s property in which the dynamics of individual elements of a system are correlated in time due to non-linear interactions between elements [24, 25]. This vision is also motivated by the conceptual model of famine evolution by Paul Howe [26], our observations of synchronizations of infectious outbreaks in Yemen, 2016–2019 [27], and a deeper understanding of synchronization of complex human networks [28].

In this communication, we argue that the synchronization of malnutrition and other measurable factors could be demonstrated with publicly available records and well-established methodology to measure seasonal characteristics [16, 29–31] in understanding the relationships between malnutrition and their driving factors. We aimed to compile a dataset using publicly available secondary data sources [32] and estimate seasonal characteristics of malnutrition in Yemen and its drivers within the acute malnutrition framework [23]. We hypothesized that seasonal fluctuations in malnutrition spatiotemporally co-occur with the seasonality of basic drivers, including environmental, economic, and differ based on conflict-related drivers. The presented perspective of seasonal synchronization help us to advance famine modeling and set up a novel framework to examine forces that drive famine formation, evolution, and resolution.

Data and methods

In Yemen, different institutions and humanitarian agencies have generated data for purposes such as demographic and health surveys, program monitoring, intervention evaluations, nutrition surveillance, and population censuses [32]. However, the challenge lies in the harmonization of these datasets, given the different methodologies and objectives behind their collection [33, 34].

Acute malnutrition indicators: We utilized two primary data sources to assess the seasonality of child acute malnutrition in Yemen. The first dataset was extracted from the Global Nutrition Cluster, which involved 45 partners coordinating nutrition activities and was created by the Office for the Coordination of Humanitarian Affairs (OCHA), covering district-level data from January 2019 to December 2022. We selected the number of newly admitted children under five with severe acute malnutrition (SAM) without medical complications enrolled in outpatient therapeutic programs (OTPs) as an indicator. These data were aggregated at the governorate level for analysis. The second source was obtained from the WHO Yemen Nutrition Surveillance Infographics, covering the period from July 2019 to December 2023. This system, established by WHO in collaboration with the Ministry of Public Health and Population (MoPHP) and funded by the World Bank’s Emergency Health and Nutrition Project, employs a facility-based passive screening methodology to monitor child malnutrition. Data from district hospitals were collected monthly across Yemen’s governorates, and we considered the number of children under five with SAM as another indicator.

Underlying drivers: Food insecurity data were sourced from the World Food Programme’s (WFP) HungerMapLIVE platform, which provides daily reports based on mobile vulnerability analysis and mapping (mVAM) surveys. We extracted daily governorate-level data from the PDF reports and aggregated them into a time series from October 2021 to December 2023. Two indicators were included: (1) the number of people with insufficient food consumption, based on the Food Consumption Score (FCS), and (2) the number of people using crisis or above crisis-level food-based coping strategies, derived from the Integrated Food Security Phase Classification (IPC) Phase 3 (Crisis) threshold.

Basic drivers: Conflict data were obtained from the Armed Conflict Location & Event Data project (ACLED). We filtered data based on violent event types, including violence against civilians, explosions/remote violence, and battles, aggregating these events into monthly counts per governorate. Monthly food price estimates by product and market were collected from the World Bank, and we included wheat flour as the primary indicator of food price in our analysis. Environmental data, including rainfall and the normalized difference vegetation index (NDVI), were extracted from the WFP Climate Explorer on the vulnerability analysis and mapping (VAM) platform. Rainfall estimates were derived from the CHIRPS dataset, and NDVI data were sourced from the MODIS platforms Terra and Aqua. These data were provided as three decades per month per governorate, and we averaged across the decades to generate monthly values. Temperature data, specifically daily average air temperature, were extracted from NASA’s Earth Data Portal Giovanni, and monthly averages were computed for each governorate.

Descriptive analysis

From 22 Yemeni governorates we excluded the governorate of Socotra from the analysis due to its distinct cultural, climatic, and geographic characteristics compared to the mainland governorates in Yemen and limited data. To enable comparability across governorates, the malnutrition outcome indicators, specifically the number of children with severe acute malnutrition (SAM) either admitted for treatment or screened, were adjusted by dividing the actual number of affected children by the total population of each governorate population. Data were collected from January 2016 to December 2023. For datasets where information was not available for the full period, the analysis began from the earliest available date after January 2016 to the latest available date before December 2023. We assessed data availability and missingness for each variable and governorate (see Supplementary Material, Table S1). We grouped governorates into two regions: the northern region, primarily controlled by Houthi forces, and the southern region, governed by the internationally recognized government and local factions. These regions have experienced varying levels of conflict, displacement, infrastructure damage, and geographic and climatic differences, all of which may influence data patterns.

Seasonality analysis

Seasonal patterns were analyzed using harmonic regression models for each governorate separately. Harmonic terms were employed to identify potential seasonal peaks within the calendar year. Each variable was fitted with trend (linear, quadratic, and cubic terms), one peak (2π sine and cosine harmonic terms–Model 1), and two peaks (2π and 4π sine and cosine terms–Model 2) to evaluate the presence and shape of a seasonal pattern. To build predictive seasonal curves we applied Model 1 and estimated peak timing using the δ-method when the best-fitting model indicated a single peak, specifically when at least one harmonic term coefficient was significant; otherwise, we selected Model 2 and stated the presence of seasonality only if at least one of the coefficients of second harmonic terms were significant. For models with two significant peaks, peak timing was determined using global and local maximum methods. To account for distributional features, we applied linear regression (Models 1a and 2a) to temperature and food insecurity data, a Generalized Linear Model (GLM) (Models 1b and 2b) with a log-link function for NDVI, rainfall, and wheat price data, and a Negative Binomial Model (NB-GLM) (Models 1b and 2b) for malnutrition outcomes and conflict events.

$$\text{Model 1a}:E\left(Y_{i,t,g}\right)={\beta }_0+{\beta }_{1}T+{\beta }_2{T}^2+{\beta }_3{T}^3+{\beta }_{4}sin\left(2\pi \omega T\right)+{\beta }_{5}cos\left(2\pi \omega T\right)$$

$$\text{Model 1b}:log\left(E\left(Y_{i,t,g}\right)\right)={\beta }_0+{\beta }_{1}T+{\beta }_2{T}^2+{\beta }_3{T}^3+{\beta }_{4}sin\left(2\pi \omega T\right)+{\beta }_{5}cos\left(2\pi \omega T\right)$$

$$\text{Model 2a}:E(Y_{i,t,g})={\beta }_0+{\beta }_{1}T+{\beta }_2{T}^2+{\beta }_3{T}^3+{\beta }_{4}sin\left(2\pi \omega T\right)+{\beta }_{5}cos\left(2\pi \omega T\right)+{\beta }_{6}sin\left(4\pi \omega T\right)+{\beta }_{7}cos\left(4\pi \omega T\right)$$

$$\text{Model 2b}:log\left(E\left(Y_{i,t,g}\right)\right)={\beta }_0+{\beta }_{1}T+{\beta }_2{T}^2+{\beta }_3{T}^3+{\beta }_{4}sin\left(2\pi \omega T\right)+{\beta }_{5}cos\left(2\pi \omega T\right)+{\beta }_{6}sin\left(4\pi \omega T\right)+{\beta }_{7}cos\left(4\pi \omega T\right)$$

where $Y_{i,t,g}$ represents the time series of monthly value for $i$-indicator, $t$-month, and $g$-governorate; sin(2πωT), cos(2πωT), sin(4πωT), and cos(4πωT) are the harmonic terms with $\omega =1/12$ to represent the annual cycle in radians; the variable T = {1…N_i,g} represents the continuous sequence of calendar months of N_i,g—length, used in linear ($T$), quadratic ($T^2$), and cubic ($T^3$) terms to account for non-linearities in trends [16, 19, 29]. The seasonal curves for each governance and outcome were predicted from these models and normalized to produce compact visualizations.

Results

Monthly time series of indicators in southern and northern governances of Yemen are presented in Fig. 2. In panels A and B, seasonal patterns for both child malnutrition indicators, including children with SAM admitted for treatment (from nutrition cluster data) and screened at local facilities (from WHO data), were similar between the southern and northern regions of Yemen, yet consistently higher in the north (42.67 ± 7.15 vs 37.65 ± 7.07 per 1,000,000 of the total population, p < 0.001, for children under five with SAM admitted for treatment; and 118.61 ± 67.78 vs 37.65 ± 22.36 per 1,000,000 of the total population, p < 0.001, children under five screened with SAM, respectively).

Fig. 2.

Monthly time-series line chart for Yemen (2016–2023) presented in nine stacked panels, each representing different indicators, such as malnutrition outcomes (panels A and B), food insecurity (panels C and D), conflict events (panel E), food prices (panel F), and environmental factors (panels G, H, and I) for goverances in North (red line) and South (red line) Yemen

Food security indicators (panels C and D) also demonstrated comparable seasonal patterns between the regions. The southern region had a higher overall prevalence of people with insufficient food consumption. However, the prevalence of people using crisis or above crisis-level food coping strategies was lower compared to the northern region in Yemen. The northern region experienced significantly more conflict events than the southern region between 2017 and 2020 (panel E). A marked and sustained increase in the price of wheat flour was observed in the southern region beginning in mid-2020 (panel F). Regarding environmental factors, the northern region consistently exhibited higher NDVI values, more intense rainfall peaks, and lower temperatures compared to the southern region (panels H and I). A detailed peak timing estimates in indicators for southern and northern Yemen are presented in Fig. 3.

Fig. 3.

Heat map of model-predicted daily values across governorates in Yemen. The predicted values were rescaled to range from 0 to 1 using min–max normalization shown as intensity to facilitate the comparison. In the heat map, darker color hues represent higher predicted values, while lighter hues indicate lower predicted values. The map is divided into two columns: data from northern Yemen are displayed on the left, and data from southern Yemen on the right. The nine stacked panels correspond to different indicators, such as malnutrition outcomes, food insecurity, conflict events, food prices, and environmental factors. Within each panel, governorates are ranked by their peak values for the respective indicators. On top of the heat map, black triangles denote the timing of the primary significant peak for each governorate. If a governorate exhibits a secondary peak for a given variable, gray dots indicate the timing of this secondary peak

We detected a significant seasonal pattern for malnutrition outcomes across most governorates, with the majority exhibiting a single annual peak (panels A and B). The peak time typically occurred from late July to early October, with the northern region displaying an earlier peak compared to the southern region. Food insecurity indicators exhibited a similar peak timing (panels C and D). About 30–50% of governorates in both the north and south exhibited a secondary peak in insufficient food consumption during January and February. Food prices exhibited the least seasonality among all indicators analyzed (panel F) and conflict events did not display seasonal patterns (panel E). As expected, environmental indicators (panels G, H, and I) demonstrated strong seasonal patterns, with peak timings closely co-occurring with food security indicators and malnutrition outcomes. Specifically, the seasonal peaks for NDVI aligned closely with the malnutrition indicator of children under five with SAM in both the northern and southern regions. Rainfall peaks consistently occurred earlier than NDVI peaks, with the southern region showing a more dispersed pattern; and temperature exhibited the most concentrated seasonal peak between June and July. In a detailed comparison of all indicators within governorates (see Supplementary Material, Figure S1), we found that the secondary peak in rainfall likely co-occurred with significant secondary peaks in food insecurity in Marib, Shabwah, and Al Hudaydah governorates.

Discussion

This study provides sensitive, fully interpretable models and comprehensive analysis of the child malnutrition seasonality in Yemen guided by the acute malnutrition conceptual framework [23]. We reflect on the findings from three vantage points: 1–recognizing the seasonal synchronization of malnutrition patterns across space; 2–recognizing the synchronization of seasonal patterns (or the lack of) in malnutrition drivers, indicating internal temporal sequences and time lags across environmental indicators, food prices, coping strategies, and food insecurity; 3–recognizing the implications of synchronization for understanding the evolution of malnutrition and famine.

Spatial synchronization. Our analysis detects a significant seasonal pattern of malnutrition across most governorates in Yemen, with the majority exhibiting a single annual peak occurring between late July and early October, which aligns with the lean season [35]. We also observed some discrepancies where southern Yemen has a higher percentage of people with insufficient food but a lower percentage of people using crisis or above-crisis-level food coping strategies compared to the north. Higher food insufficiency in southern Yemen could be associated with the food price gap between south and north Yemen, with north Yemen having higher food prices [36]. However, humanitarian aid and more effective local coping mechanisms, such as community aid programs and remittances, may play a more significant role in mitigating the need for extreme coping strategies. In contrast, in northern Yemen, conflict and logistical challenges may limit access to aid. Reports indicate that the Houthis’ systematic obstruction of aid agencies, combined with a collapse in donor funding [30], has compounded the situation, forcing more people to resort to severe, crisis-level coping strategies.

Temporal synchronization. Within regions, the results reveal relatively consistent seasonal patterns in the timing of environmental factors, prices, coping strategies, food insecurity, humanitarian assistance, and malnutrition over time. Temporal synchonization could suggests potential sequences with short time lags. For example, challenging environmental indicators couls lead to price increases and more severe coping strategies that only temporarily protect against a rise in food insecurity and then malnutrition. Humanitarian assistance attempts to alleviate this seasonal rise. There are several points to explore further.

First, the syncronized peak timing between admissions for treatment of children with SAM and screenings of acute malnutrition from different organizations (OCHA/UNICEF and WHO) carries several important implications. This synchronization might indicate that humanitarian assistance and monitoring systems are effectively capturing the same seasonal peaks in malnutrition, presenting an opportunity for earlier interventions to prevent SAM through more coordinated, integrated, and timely interventions across organizations. In general, by ensuring resources are mobilized before critical periods when SAM rates will be at their highest, these efforts can improve the efficiency of response measures, reduce duplication, and enhance the overall impact of humanitarian assistance on malnutrition. Additionally, it suggests that aligning and integrating surveillance and treatment data could allow for more accurate forecasting and early warning systems [37], enabling preemptive actions that prevent further deterioration in child health during peak malnutrition seasons. The close alignment of seasonal patterns between food insecurity and acute child malnutrition supports the notion that food availability and household food security are critical underlying drivers of malnutrition in Yemen. Similar time series analyses conducted in Tanzania [38] and Malawi [39] also found a significant seasonal association between child malnutrition and food insecurity, further supporting this connection among vulnerable populations across African and South Asian countries.

Second, the strong seasonal patterns and their temporal alignment to malnutrition in environmental indicators, particularly NDVI, underscore their influence on malnutrition and food security in Yemen. A study conducted among dryland African countries found that NDVI most correlate with child acute malnutrition compared to other environmental factors like precipitation and temperature [20]. The synchronization between NDVI peaks and malnutrition outcomes in both regions in our study could be explained by the alignment of humanitarian aid and health campaigns with periods of highest malnutrition risk, leading to increased screening and treatment during the peak in vegetation; on the other hand, households may be more likely to bring their children to health centers for screening at the end of the planting season. The northern region, with its more intense rainfall and earlier NDVI peak values, experiences an earlier peak in malnutrition. This finding aligns with studies in sub-Saharan African countries, which have demonstrated a positive association between climate factors—such as rainfall and vegetation patterns—and the prevalence of acute malnutrition [40, 41], likely due to the more rapid depletion of food resources post-harvest.

Third, while conflict events did not exhibit a strong seasonal pattern, they remain a critical factor exacerbating food insecurity and malnutrition in Yemen. In our analysis, secondary peaks in food insecurity coincided with the peak of conflict-related events in governorates such as Hajjah, Ibb, Sa’ada, Aden, and Shabwah. This aligns with findings from other studies, which demonstrate that conflict was positively and statistically associated with the increase in child wasting [10, 42]. However, the absence of a clear seasonal pattern for conflict suggests that its impact is unpredictable, creating challenges for designing timely interventions.

Implications for famine modeling. The key contribution of these findings lay in recognizing the implications of seasonal synchronization for understanding the evolution of malnutrition and famine. An emerging theory of famine formation [26, 43] suggests that these crises arise when pressure (a combination of shocks and vulnerability) is kept in place by a hold (something such as access constraints that prevents release from pressure) leading to self-reinforcing dynamics (a linked set of changes resulting in food insecurity and undernutrition) that tip over into a famine system (a rapid rise in hunger-related mortality) before rebalancing (a return to the baseline mortality). The spatial and temporal patterns that have been discussed contribute specifically to explaining the synchronized sequences that occur as the pressure leads to self-reinforcing dynamics. That is, for these contexts, it provides evidence of a pattern in which the pressure, or the environmental factors, trigger self-reinforcing dynamics, or sequential changes in prices and coping strategies leading, with a time lag, to rises in food insecurity and then malnutrition. This pattern, therefore, sheds light on and makes visible these complex interactions. It also allows us to see that, in this case, the delay in humanitarian assistance constitutes a hold.

In addition to the description of the Yemen case, the findings suggest three broader directions for exploring these components of famine evolution and modeling famine dynamics. First, in our early work, we proposed methodology to detect three forms of synchronization: 1–serial synchronization that captures whether two or more outcomes (or locations) share a similar temporal pattern; 2–phase-phase synchronization that derives associations between peak timing estimates and identifies co-occurrences in timing of two or more outcomes (or locations); and 3–phase-amplitude synchronization that detects seasonal behaviors of outcome-location pairs and examines whether the intensity of a seasonal peak varies in relation to its peak timing. We believe the synchronization of famine-related seasonal drivers is indicative of a complex dynamic because it generates non-linear, emergent crises. Instead of additive effects, synchronized stresses interact through feedback loops, creating rapid escalations that are far harder to predict and manage.

Second, famine evolution depends on the balance between competing forces that drive the progression toward famine and those that offer buffers and counterbalance [43]. We speculate that detection of seasonality in the basic and underlining drivers could be indicative of key features of a complex system in the context of crises, including competing forces and their balance, rapid escalations triggered by disturbances, and feedback loops that can either amplify or stabilize outcomes. Some factors could behave as stressors (or pressure forces) and drive the system away from equilibrium toward scarcity, malnutrition, and crisis; some can further escalate the conditions once certain thresholds are crossed, even without additional external shocks, Further investigations into seasonality synchronization might help us to contrast the self-reinforcing (positive) dynamics with self-correcting (negative) dynamics that can stabilize famine conditions.

Third, the ability to detect complex spatiotemporal alignments of conditions leading to famine help us with forming and testing hypotheses related to spatial dynamics sharing common seasonality features. Future research should focus on understanding the implications of seasonal synchonization and drivers of seasonal peaks in food insecurity and malnutrition.

Limitations

While this study provides valuable insights into the seasonality of malnutrition in Yemen and the implications for famine modeling, several limitations must be acknowledged. In this study we used only publicly available data, and data availability varied across governorates and time periods, potentially affecting the completeness and accuracy of the seasonal analysis. Moreover, the reliance on secondary data means that the quality and consistency of data sources may vary, and discrepancies between data collection methodologies could influence the results. In conflict sitiations and under regimes where politics control the data flow, access to data (and also data quality, biases, accuracy) forms the narrative [44]. More work need to be done to share critical data with broader research community to foster collaborative innovation in data harmonozation and modeling [33, 34]. We recognize that the selected outcomes may not be the best for predicting the actual rate of SAM, and likely to be underestimated. Region-specific infrastructure differences can also affect the proportion of children suffering from SAM who eventually receive treatment. The risk of bias was minimized in two ways in this analysis: first, by focusing on seasonality and not rate prediction, and second, by performing performed region-specific analysis to negate confounding at regional level. Limited and outdated census data could obscure estimates of malnutrition. Typically, the malnutrition estimates are derived as the total number of children screened divided by the total number of ‘expected’ children with SAM multiplied by programmatic coverage or derived using the population of children under five multiplied by the most recent available point prevalence estimates and incidence rates. However, if both nominators and denominators are fluctuating seasonally, then the ratios obscure those fluctuations [29]. Ideally, several metrics of malnutrition including the total share of children that are suffering from acute malnutrition are needed to guide both practitioners and data modelers.

Our findings call for investments and legal provisions in continuous monitoring and evaluation and preserving historic records related to humanitarian assistance outcomes, environmental conditions, socioeconomic challenges, and conflict-related factors. Longitudinal studies are needed to explore the long-term effects of climate change, conflict, and economic instability on the seasonality of malnutrition. The emerging harmonized datasets and the ensemble of modeling approaches could further refine the agenda for understanding, modeling and forecasting seasonal acute malnutrition and faming formation, evolution, and prevention [43, 45–47].

Conclusions

This study sheds light on the seasonal dynamics of child malnutrition in Yemen, emphasizing the importance of understanding the seasonality of factors involved in the evolution of crises, seasonal rises in malnutrition, and the risk of famines with high mortality. Our approach to quantification of seasonal charateristics in term of seasonal peaks and their amplitude pave the way to build strong foundation for deeper understanding of famine formation and evolution using a complex system perspective. In particular, it provides new insights into the spatiotemporal synchronization of key features of complex dynamics. This type of modeling might have the potential for strengthening early warning systems and real-time monitoring of evolving crises. Collaboration among humanitarian organizations, local stakeholders, and modelers is essential for implementing data collection efforts that are both timely and contextually appropriate.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

B.Z. drafted the main manuscript and created all figures. A.M., K.K.H.C., and E.N.N. provided feedback and revised various drafts of the manuscript. E.B. provided the necessary data for analysis. All authors reviewed and collectively approved the final version of the manuscript for submission.

Funding

None.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Conflict of interest

The authors declare no competing interests.