Famine can be conceptualized as an emergent state within a complex, dynamic system characterized by the interaction of social, economic, political, and environmental factors in non-linear patterns [1,2,3,4,5]. Howe [1] and Fortnam and Hailey [2] emphasize that famine dynamics exhibit cascading effects and self-reinforcing feedback loops that may propel populations toward a famine state, marked by starvation, acute malnutrition, disease outbreaks, and mass mortality [1, 2, 4,5,6,7,8,9,10].

Despite growing recognition of famine’s complexity, current analytical approaches lack systems-based methodologies and analytical tools [2,3,4]. The Integrated Food Security Phase Classification (IPC) relies on outcome severity indicators and static, threshold-based classification, potentially overlooking temporal, spatial, and processual dimensions of famine risk [1,2,3,4,5,6,7,8,9, 11, 12]. IPC Famine (Phase 5) classification requires randomly sampled survey evidence that outcome indicators exceed specific thresholds [13, 14], yet gathering reliable data for these measures is frequently challenging in conflict-affected contexts [3,4,5, 11, 15]. The IPC classification also involves multiple procedural stages, each adding complexity and potential for delay (see Supplement A for full IPC protocols, thresholds, and procedural detail).

The conditions signaling famine risk simultaneously degrade the evidence base. Conflict, restricted access, displacement, and isolation limit the feasibility of representative surveys precisely when evidence is most needed [1,2,3,4,5, 11, 12, 15, 16].

Quantitative models designed for predicting IPC Phase classifications, including famine, face critical limitations: IPC thresholds do not capture complex dynamic interactions, and classifications rely heavily on expert judgment rooted in imperfect information, increasing risk of errors, biases, and analytical noise [4, 9, 15, 17,18,19,20]. Such models essentially predict expert judgments rather than objectively measurable realities.

Fuzzy Cognitive Mapping (FCM), developed by Kosko in 1986, is a practical method for modeling complex systems through weighted causal networks [21,22,23,24,25,26,27,28]. FCM’s critical strength lies in capturing feedback loops and emergent properties both visually and mathematically, enabling comprehensive analysis of dynamic systems [21,22,23,24,25,26,27,28]. FCM integrates diverse data types, supports scenario analysis, and generates system visualizations [23,24,25, 27, 28]. Despite widespread recognition that famine operates as a complex system, FCM has not been systematically evaluated for famine risk analysis. This paper applies FCM to a retrospective case study to assess its utility for tracking famine risk and identifying the critical pathways and feedback loops driving system deterioration toward famine (see Fig. 1).

Example fuzzy cognitive map

Ayod County, South Sudan, situated within the Sudd in an agropastoral livelihood system, experienced severe flooding, disease outbreaks, and conflict between 2016 and 2017 [16, 29,30,31,32,33,34,35,36,37]. By early 2017, starvation and livelihood collapse led to near-complete reliance on wild foods, disease outbreaks, and widespread malnutrition, resulting in an IPC Acute Malnutrition (AMN) Phase 5 classification in May 2017 [38, 39]. A household survey in June 2017 reported 29% of households at IPC AFI Phase 5 Catastrophe [39], and the non-trauma crude death rate reached 1.89 (95%, CI: 1.26–2.84) per 10,000 per day—slightly below the famine threshold of 2.0 [40]. Local populations referred to it as “The Year of the Thou,” signifying complete dependence on tree leaves for sustenance [39]. Despite these indicators, Ayod was not classified as a famine by the IPC [41].