The framework presented here illustrates the potential of household simulation models to help understand famine dynamics when coping mechanisms are adopted to mitigate hunger and starvation. Although the prototype model presented does not include all system components, the simulations are consistent with literature on the high value of home gardens in improving food security, nutrition, and economic wellbeing [14, 15] and point to higher resilience afforded by food acquisition spread out over the year as opposed to concentrated in the single harvest events of annual field crops. The results also indicate a role for supplemental food sources—harvesting of wild foods in these scenarios—in improved outcomes, both in cases of chronic food shortfall with crisis disruptions of primary food acquisition. Caution should be taken in interpretting this result, however, because the prototype model does not yet include labor tradeoffs and the opportunity cost diminished farm or garden yields if time is dedicated to foraging rather than cultivation.

An important insight from the simulations is the potential for household-level rationing to either increase or decrease mortality, depending on other circumstances. While the specific quantitative results should be interpretted with caution, the simulated pattern reflects the physiological reality that rationing results in caloric deficiency, which can increase the chance of mortality, whether directly or indirectly (see Thorn and Fitzpatrick [44], current issue). In real-world famine systems, one trajectory that can occur is for food intake to be limited early in the crisis, before infrastructure breakdown or displacement has lead to increased vulnerability to disease and other stressors, and for the depleted stores of body energy to compound risks if subsequent broader breakdown occurs [16]. The potential for rationing as a short-term coping mechanism to threaten long-term survival exemplifies the potential costs of coping mechanisms as described in Howe’s [5] “watch” archetype.

More generally, the simulations suggesting that access to wild foods as a supplemental resource can delay onset of high mortality (by up to 3 months according to some simulations) illustrate the ways in which “the watch” dynamics can mask an ongoing crisis. Also, the role of supplemental food sources or rationing in enabling households to survive during chronic shortfall of food from primary sources may illustrate less visible differences in vulnerability among households. A household closer to “the edge” may appear to flourish under baseline conditions if supplemental resources are sufficient, but quickly succumb to disruptions if they are already fully exploiting available resource streams.

While the quantitative household model is highly preliminary, the model outputs do shed light on important metrics to be emphasized in humanitarian responses to food crises. The results show how household coping behaviors can enable households to survive a crisis, while simultaneously masking their degree of vulnerability. For households close to ‘the edge,’ rationing (Supplementary Figure S6) and/or foraging (Supplementary Figure S7) can enable long-term survival even if their baseline food production falls short of their annual needs. When existing vulnerability is compounded by a major disruption in food production, such households will be unable to exploit these strategies moving forward. This observation underlines the importance of behavioral metrics for vulnerability and resilience, such as the Coping Strategies Index (CSI) and the simplified reduced Coping Strategies Index (rCSI), which uses coping behaviors as indicators of chronic or acute food insecurity [17, 18]. The CSI and rCSI are commonly used—typically alongside other metrics—both in primary research [19,20,21,22,23] and as monitoring metrics for humanitarian organizations such as the World Food Program, UN Food and Agriculture Organization, and the IPC acute food insecurity classification [24].

A notable challenge in measuring food insecurity is that despite strong supporting evidence for various metrics, the metrics themselves are not always correlated, and the IPC advocates contextualization of metrics [4]. Maxwell et al. compared the performance of the CSI, rCSI, Household Food Insecurity Access Scale, Household Hunger Scale, Food Consumption Scale, and other metrics and concluded that areas of weaker correlation reflected the difference in underlying aspects measured, combined with the challenges of establishing arbitrary cutoffs, and importantly that different metrics are most at different levels of insecurity [24]. The IPC manual notes that “a defining characteristic of Phases 3 and 4 is that food consumption might reflect a lower phase, but only because households are using negative crisis or emergency coping” [4].

A more complete systems dynamics model for household resource management could help to elucidate the progression through known indicators of food insecurity, aiding in the development of more consistent combined metrics for crisis severity, using the underlying temporal process to help diagnose changing conditions, including the inconsistent overlap between existing metrics highlighted by Maxwell et al. [24]. Focusing on household dynamics could be particularly valuable for developing metrics incorporating diversity of livelihood and social position within a famine system, better reflecting Sen’s entitlements framework [13]. The relatively simple modeling framework described here is well suited for incorporation of into agent-based models, such as have previously been developed for various rural economies [25,26,27,28,29,30], would enable simultaneously accounting for differences in livelihoods, finances, and social position, as well as interactions among diverse households.

Before applying the framework and modeling approach described here to real-world systems, however, it will be crucial to incorporate additional components Fig. 2 into the quantitative model, particularly labor tradeoffs and market dynamics.

As noted above, the prototype model does not account for the interference of time spent foraging with future garden or field productivity, nor is there a mechanism for the reduction in labor capacity expected as the body’s energy reserves are depleted. The only adverse consequences of the simulated coping mechanisms are increased mortality due to weight loss, and the draw-down of resources (food in storage, fat stores on the body, and wild food sources), which reduces the ability of the household to absorb future shocks. Thus, the model does not fully capture the interference with future harvests described in the “watch” archetype from Howe [5]. A more complete model for the “watch” should include rules for calculating labor effort available and allocating labor between food acquisition activities (Supplementary Figure S8). A challenge in implementing labor tradeoffs will be determining the correct definitions for time-delayed returns on labor investment, whereas investment of labor in foraging will produce food in the immediate term, loss of field or garden labor early in the growing season is expected to reduce harvest at the end of the season, which could be defined either by a simple mathematical definition converting labor input into expected harvest or by a detailed crop model, such as those reviewed in [31], explicitly accounting for management actions taken at different stages of plant development. Incorporation of crop models (and ecological models) would provide the added benefit of facilitating appropriate coupling productivity for field, garden, and wild systems based on climate conditions, but would also require more specific model inputs with regard to land use and yield potential.

Relatedly, the logistic growth model used to simulate wild food regeneration does not account for biodiversity or differences in the nutritional value of different plant tissue types. Whereas the logistic model assumes immediate growth responses to biomass removal, the plant tissues highest in macronutrients (mostly starch and protein, with some oils) include seeds, fruits, and tubers [32]. These tissues are unlikely to contribute to plant population growth or regeneration until a future generation or season. Examination of the biology of specific famine food plants would be valuable in model refinement.

In addition, while the current model centers on direct subsistence food production, this livelihood is not common, and the model furthermore focuses on households in a closed famine system, a situation typically combined with additional stressors. In his classic book Poverty and Famines, Sen notes that individual outcomes during a famine depend upon economic and social inequalities, which he frames as different forms of entitlements [25]. Furthermore, Corbett identifies numerous financial and transactional strategies for dealing with periods of hunger, including borrowing resources, selling assets, and engaging in outside labor [11]. Inclusion of market dynamics could therefore enhance model realism (Supplementary Figure S8). Such dynamics could either be predetermined (such as with a cosine model for seasonal fluxuations in price) or be dynamically simulated by adapting existing economic models [34,35,36]. The focus of the model on energy balance and direct food acquisition may also not reflect the realities of closed famine systems. Closed famine systems are most often occur in the context of armed conflict [6], a scenario in which violence, infrastructure breakdown, displacement, and environmental exposure may be of particular importance to mortality. The model presented may be most appropriate for the early stages of closed famine systems. Modifications for more open systems, for example, by including market dynamics, could broaden the applicability.

Another facet of open famine systems is the ability of household members to relocate to places with employment prospects or better access to humanitarian aid [6]. To incorporate population mobility into the described modeling framework, one approach would be to model the household members as two connected ‘stocks’ representing different locations or livelihoods with access to different livelihoods and risk profiles.

The prototype also assumes all household members have the same physiology, whereas real households are of course composed of people of different ages, genders, and body types that contribute to household livelihood in different ways. Modification of the model to include diverse household members would improve realism.

The current metabolic submodel is also extremely simplified. As noted by Thorn and Fitzpatrick (current issue), a realistic starvation model should account for loss of lean weight, not just body fat, and to model starvation during a famine, interactions with personal and maternal history, disease, malnutrition, and other stressors should be considered. Adaptation of more physiologically based starvation models such as Hall [33] could improve the household modeling framework in its current form and inform further model development incorporating physiological determinants of capacity for labor.

Finally, while coping strategies in the current model are largely reactive, real-world coping mechanisms are well documented to be pro-active, drawing on traditional knowledge to plan well in advance of when food supplies run short [7]. The system of threshold-based decisions described in the current modeling framework might not sufficiently capture the sophistication of real-world decision-making. To appropriately apply the framework described here to a real-world system, mathematical functions for coping responses should be descriptive of documented human behaviors, ideally including location-specific ethnographic studies of household resource management.