Modeling how and why aquatic vegetation removal can free rural households from poverty-disease traps

When households do not harvest the aquatic vegetation, the vegetation remains stable at the system’s carrying capacity (Fig. 2A). Because the snail vector population scales with the vegetation that provides it habitat and nutrients, household infection reaches a high steady state (Fig. 2 B and C). Households spend most labor on their farm and use moderate amounts of fertilizer in food production. High infection rates limit labor supply, however, leading to low income and a poverty-disease trap. These patterns are very similar across the wealth distribution.

If the household can harvest aquatic vegetation, however, all household types allocate only a small fraction of their labor to that task, but with considerable impact. Households’ clear some vegetation from the water source, leading to a stable vegetation level well below the carrying capacity, consistent with field experimental data finding that 10 or fewer individuals could clear a village’s water access points in a day (13). Even with continued household fertilizer use, modest effort allocated to aquatic vegetation harvest maintains a reduced aquatic vegetation stock, driving down the household infection rate, especially for villages characterized by poorer households with low or moderate land endowments (Fig. 2D). The differences between villages with smaller farms and poorer households versus larger farms and relatively richer households in this setting are driven by differences in optimal fertilizer use. Higher incomes relax households’ budget constraints, permitting increased fertilizer purchases, given that the expected marginal revenue product of fertilizer significantly exceeds its price in this setting, at prevailing application rates. However, more fertilizer use results in increased runoff, resulting in slightly higher levels of aquatic vegetation and thus schistosomiasis infection rates for villages with larger farms and better-off households.

Most household labor remains allocated to food production, but lower infection rates mean greater labor availability. This greater labor, in addition to the added nutrients returned to the soil from the compost, leads to higher median incomes than in the baseline case without vegetation harvest (Fig. 2E). These results highlight that the attractive economic returns to compost created from the harvested aquatic vegetation (13) can help disrupt disease ecology dynamics, both reducing infection rates and boosting incomes in a favorable reinforcing feedback loop. The model helps us understand the underlying mechanisms that explain how and why the intervention seems to work.

Fertilizer use is higher when vegetation is harvested. Since compost and fertilizer are substitutes, one might expect fertilizer use to decrease as farms begin harvesting aquatic vegetation. However, such substitution effects are often dominated by income effects, especially when fertilizer use is suboptimal relative to its expected profitability due to farmers’ financial liquidity constraints, as the prior household modeling literature has long established (31–34, 36, 37). Aquatic vegetation harvest increases incomes by increasing household labor availability and food productivity. Those higher incomes then stimulate greater household food demand and relax financial liquidity constraints to fertilizer purchase. So long as the expected returns to fertilizer use significantly exceed the price of fertilizer, as seems true in the northern Senegal context, then farmers apply fertilizer if they can afford it, Thus, the income effect can be—and as parameterized based on the available data from this context, is—stronger than the substitution effect and fertilizer use increases as farmers compost harvested aquatic vegetation. The higher dynamic equilibrium of greater food production and incomes alongside lower infection rates gest sustained by farmers regularly devoting some of their increased labor availability from lower infection rates to clearing water access points.

Our simulations are consistent with the empirical association of fertilizer use with infectious disease exposure (12, 13). The optimal level of fertilizer for a household may depend on the level of infection. To test this directly, we reran the simulations starting with households at different infection rates and keeping all other model parameters the same. We calculated the median optimal first-year fertilizer use and plotted it across the different starting infection conditions (Fig. 3). Optimal fertilizer use is negatively associated with infection rate, as predicted. The decrease is meaningful in magnitude; very high infection levels are associated with almost a 50% decrease in optimal fertilizer use compared with low infection levels This result again reinforces the central point that some innovation is needed to break communities out of their current high schistosomiasis infection, low agricultural productivity equilibrium.

We explore the sensitivity of our results to the effect of fertilizer runoff on vegetation ($\rho$), the vegetation recolonization rate ($n0$), the vegetation growth rate ($r$), and the price of fertilizer ($p_u$). We also conduct a sensitivity analysis of the price of the household good ($p_h$).^* For the sensitivity analysis, we focus on changes to parameters in the system and consider the median household land holding of two hectares.

The core model results described above are generally robust to changes in the effects of fertilizer runoff on vegetation growth, recolonization rate, and growth rate, and economic incentives modeled through changes in the price of fertilizer and the household good. Slightly higher levels of infection and lower labor availability result when the fertilizer runoff effect (Fig. 4) and vegetation recolonization rate increase (SI Appendix, Fig. S1). At lower levels of vegetation growth, the vegetation stock is smaller, infection prevalence is lower, household labor availability is higher, and income is slightly improved (Fig. 5).

As expected, cheaper fertilizer leads households to use more of it, which results in modest increases in infection prevalence (SI Appendix, Fig. S2). Finally, our results show no meaningful changes when the price of the household good changes (SI Appendix, Fig. S3). Together, these results show that the patterns in our main results are consistent across a range of reasonable values for underlying agroecosystem and market conditions, and thus provide a robust structural way to capture the relationship between aquatic vegetation growth, the microeconomic decisions of households, and poverty and disease outcomes. The returns to compost in food production are routinely large enough to induce aquatic vegetation harvest if people are aware of the household benefits. However, the impact of aquatic vegetation harvest can be muted at higher levels of fertilizer use because vegetation growth spurred by fertilizer use offsets some of the gains made by harvesting aquatic vegetation.

Back-of-the-envelope estimates suggest that economic gains from universal awareness of the potential benefits from aquatic vegetation harvest could average at least 3,200 USD per village or 3.25 USD per person annually, perhaps double that if one includes the estimated long-term earnings gains from improved child health and education. These are modest but significant gains for poor, rural communities. The estimated regional gain from aggregating across all villages in West Africa within five kilometers of surface freshwater that are likely to host the vegetation, snails and schistosomiasis is 138.5 million USD per year.

We developed a microstructural model of a poverty-disease trap by linking a nonseparable agricultural household model to one of schistosomiasis disease ecology dynamics through household labor availability, labor allocation choices, and optimal fertilizer use. The household-centered approach allows us to analyze how poverty-disease traps can exist under current conditions and how and why simple, low-cost interventions like aquatic vegetation harvest can help break those traps. Under the status quo, without aquatic vegetation harvest, infection prevalence is consistently high and household labor availability and income are steadily low. When we allow for vegetation harvest in the model, simulating what might happen after an agricultural extension and public health information campaign to promote aquatic vegetation removal, we see consistently lower infection levels and higher incomes. Introducing aquatic vegetation removal to this single equilibrium poverty trap model induces different household decisions that can lead to higher dynamic equilibrium incomes. The effect of aquatic vegetation harvest is greater in combination with measures that reduce nutrient runoff that spurs aquatic vegetation regrowth. Continued household fertilizer use limits the gains for those with the highest land holdings, signaling that this seems an intervention especially well-suited to communities with smaller farms. Thus, aquatic vegetation harvest has the potential to allow households to reduce the cycle of schistosomiasis infection and reinfection that characterize the poverty-disease traps currently confronting many rural households in northern Senegal and many other communities in the low-income tropics.

While removed aquatic vegetation can be dried and used for animal feed or used as a feedstock for biodigesters in producing off-grid energy, economic analyses in our study setting indicate that compost is the first-best use among the options evaluated by researchers (13). Preliminary results suggest that a latent market for compost (and livestock feed) exists.

One limitation of our modeling strategy is that we only explore the representative household’s choices, but water sources and water access points serve many households at one time. In the case of fertilizer use, one household’s decision to use lots of fertilizer will inevitably impact the common water source, increasing the aquatic vegetation and schistosomiasis reservoir for all households who use that water source. This provides an opportunity for households to harvest more vegetation, but it also poses a greater infection risk due to other households’ decisions. A natural extension of the current model would build out these interhousehold and spatial interactions into a dynamic general equilibrium bioeconomic model to trace out within-village spillover effects. Our results suggest that differential fertilizer use—and perhaps differential water contact rates—is an important piece of the system. Documenting these village externalities may prove helpful to fully understanding and tackling the poverty-disease trap.

Additionally, our representative household model does not account for any growth in a household or village over time. In the more general case of bioeconomic modeling of infectious disease transmission, one might worry about habitat changes affecting the transmission patterns (38). With aquatic vegetation removal, villagers are unlikely to create new fresh surface water habitats for snails as the rivers, lakes, and irrigation canals are relatively fixed over time and space. The snail habitat within the aquatic system is endogenous to the model as we directly model the amount of aquatic vegetation or snail habitat with the system. In our setting, we model aquatic vegetation removal, a form of habitation conversion that reduces the likelihood of contact between the disease hosts and humans. Furthermore, nutrient runoff is the key driver of snail habitat, and we explicitly link fertilizer use to aquatic vegetation in the model. We could consider increased cultivated land over time, which would be equivalent to adding a positive time trend to fertilizer use to the model. In northern Senegal, it is unclear whether cultivated land is expanding and thus we hold land constant. If cultivated land increased, households would need to more frequently remove aquatic vegetation to reduce schistosomiasis exposure.

Climate change and biodiversity loss may also influence the link between schistosomiasis and aquatic vegetation removal (39). Temperature does impact schistosomiasis transmission (35). Rising global temperatures may therefore alter the relationship between aquatic vegetation removal and schistosomiasis infection over our 20-y time horizon. While we abstract away from the impact of rising temperatures and other drivers of global change in this model, future work should explore how external drivers could interact with current feedback loops within human and environmental systems.

Field trials to date do not detect any significant ecological damage from aquatic vegetation removal (13). Unlike malaria control efforts, such as removing mangroves (40), aquatic vegetation removal is currently highly localized and requires only removing vegetation in the immediate vicinity of water access points. Most of the river or lake environment is untouched. Even so, ongoing work is explicitly monitoring for unintended ecological damage as this intervention scales.

While this model focuses on understanding how and why this intervention works in the specific context of the Saint Louis and Louga regions in northern Senegal, the modeling approach and principles of the intervention we evaluate likely apply to other settings with endemic infectious diseases. In fact, our back-of-the-envelope calculations scaled the benefits to all of west Africa, assuming that the benefits of aquatic vegetation removal observed in Senegal are similar in other locations. The estimated regional gain from aggregating across all villages in West Africa is 138.5 million USD per year. Note that these gains arise just from the value of the direct productivity effects of using the harvested aquatic vegetation to boost soil nutrients and thus agricultural output and the added earnings from fewer workdays lost to illness. This estimate is likely quite conservative in that it abstracts away from the longer-term earnings gains arising from improved health and educational attainment of children from reduced schistosomiasis infection (41–44). In fact, our per person annual gain in earnings is similar to estimates of the longer-term estimates of earnings gains from deworming, even without accounting for spillover gains on untreated children (43). Given that aquatic vegetation removal has similar gains as deworming (13), one might reasonably hypothesize that adding these long-term earnings gains among children to the nearer-term gains arising from vegetation removal and conversion into compost roughly doubles the long-run estimated benefits. Additionally, we identify and structurally model a key potential mechanism to reduce infectious disease burdens in the low-income tropics, demonstrating the importance of understanding feedback loops between household economic decision-making and the underlying natural environment, which has applications to other neglected tropical diseases and to other complex relationships between human and environmental systems. Consequently, we hope that the empirical calibration of our integrated human–biological–socioeconomic model provides a framework to evaluate environmental interventions to control other vector-borne diseases, such as malaria, Chagas disease, African Trypanosomiasis, or Leishmaniasis.

Poverty-disease traps are widespread; thus, understanding solutions is important. Research in Kenya finds significant impacts of deworming on child learning and that of their siblings (41, 42) and labor market outcomes later in life after deworming (43, 44). Thus, there are likely large potential long-term benefits of aquatic vegetation removal not modeled nor discussed here given the 20-y time horizon we impose in the modeling. Policy makers, community leaders, and development agencies should consider aquatic vegetation removal as an effective form of schistosomiasis infection control that can also boost incomes and overall quality of life for millions of people.

The bioeconomic model has two submodels. The first describes the disease ecology dynamics, more specifically, how the schistosome, aquatic vegetation, and snail populations interact, and relates these populations to human infections. The second, an agricultural household submodel, describes how utility maximizing households make decisions about how to allocate their land, labor, and income. We describe the key parameters and model equations here. A full description of the model with equations can be found in SI Appendix, Text S2.

The household’s problem is a variant of the nonseparable agricultural household model in which consumption and production decisions become inextricably linked by multiple market failures that typically characterize poor rural villages like those in our setting (30). The economic model begins with a representative household that maximizes utility, defined over consumption of food, an aggregate nonfood household good, leisure, and the health status of household members. We assume that utility is well-defined, increasing, and concave in all its arguments. We model the household’s nutrient intake via food consumption. The health production function is Cobb–Douglas for food consumption and the fraction of household members infected downscales the health status variable as the fraction infected increases. Health status increases with food consumption, representing the value of more nutrient intake. The household can only influence health status through more food consumption or a lower infection prevalence; one cannot buy good health. Because aquatic vegetation is a common pool resource, there is no market for aquatic vegetation, either in the water or as harvested vegetation. The multiple market failures in health status and aquatic vegetation together create nonseparability between the household’s production and consumption decisions. To simplify the model, we also assume no market exists for land rentals or sales and from cash labor markets as land or labor transactions are uncommon in the study area. Households allocate their time among cultivating food, harvesting aquatic vegetation, and leisure and commit their land to their own agricultural production. These assumptions do not qualitatively change model outcomes.

If households choose to harvest aquatic vegetation, they turn it into compost, which increases agricultural productivity (13). Households produce food using land, labor, fertilizer, and compost from harvested aquatic vegetation. Recent experimental evidence finds that compost and urea fertilizer are virtually perfect substitutes (13). Harvesting vegetation only requires labor.^† The household employs a constant elasticity of substitution (CES) food production function while aquatic vegetation harvest follows Cobb–Douglas production technology.^‡ More details on the agricultural household model are in SI Appendix.

To simulate the status quo ex ante, we also present a simplified version of the model without aquatic vegetation harvest, in which households cannot use labor to harvest aquatic vegetation to produce compost. Our core comparisons thus simulate the equilibrium effects of making villagers aware of the prospective value of composting harvested aquatic vegetation.

The disease ecology model tracks the populations of aquatic vegetation (Ceratophyllum, $N$), miracidia (larval schistosomes that infect snails, $M$), infected and susceptible snails ($I_2$ and $S_2$), cercariae (larval schistosomes that infect humans, $P$), and infected and susceptible humans ($I_1$ and $S_1$). We adapt an existing schistosomiasis disease ecology model (28) to fit the Senegalese context and down-scale the parameters from a large community to one that matches the household-level simulations. Additional details on the disease ecology model are in SI Appendix.

Relative to the human lifespan, the schistosomiasis infection cycle is relatively short. Cercariae live around 10 h, miracidia live around 25 h, and snail infections last around 100 d (45). Very few or none of the existing cercariae or miracidia population will survive over the course of the year, which creates a challenge to match timescales across the household and disease ecology submodels. One could convert the continuous time disease ecology submodel to discrete time to match the household submodel through significant linearization and assumptions about annual changes in miracidia, cercariae, and snail populations. But that can cause meaningful aggregation errors. We therefore instead use a continuous time disease ecology submodel that better matches the timeline of the schistosomiasis infection cycle. We simulate annual changes by simulating the system of differential equations forward 365 d, where all parameters are given in daily rates. We then export the annual output to the discrete time household model that operates at annual time steps.

Furthermore, we could instead model household decisions on a smaller timescale to match the disease ecology submodel (5, 46). However, we explicitly want to model agricultural households making decisions over an entire cropping season. Thus, we use an annual household model to best capture the microeconomic decisions that are the foundation of our model.

The disease ecology submodel and the household submodel link to one another in two ways. The first is through the infection status of the household, which directly affects household utility and impacts the household’s labor availability and thus income and the budget set that constrains purchase of fertilizer as well as food and consumption goods. The second is through the household’s use of urea fertilizer and its aquatic vegetation harvest, each of which changes the vegetation population within the water source. Thus, infection status can affect income, which in turn can affect fertilizer use and runoff that fuels aquatic vegetation growth.

The disease ecology submodel provides population estimates of infection, which we scale down to individual- and household-level infection rates through stochastic infection realizations drawn from an independent Bernoulli distribution for each household member at the start of each time period. The distribution’s mean is the infection rate predicted by the disease ecology submodel, the population infection prevalence. After the first period, we also take random draws for curing infection, where the mean of the Bernoulli random variable was set at 0.25, which captures the fact that households in this region experience sporadic mass deworming campaigns (47). The lack of smooth time paths in labor availability and household infections (Fig. 2) arises from the stochastic process that generates household infections and periodic deworming within the model.

Since each individual household is only one small part of a village and villages only access a small portion of the entire aquatic system, these households do not individually influence the disease ecology submodel. Since household behavior does not individually impact disease ecology, the household does not consider the equations of the disease ecology submodel in its own optimization. In this way, the household solves a series of static, single period optimization problems as in prior bioeconomic models (31).^§ In this framework, the disease ecology submodel shows how the state and the average infection rate change over time. In each period, we solve the household’s static optimization problem and then use the household’s choices to determine the stock of aquatic vegetation and the realizations of infection to determine the current infection prevalence. With these new starting populations, we simulate the disease ecology model 1 y forward to give the state of infection in the next time period. The model is then solved by the following iterative process for each period in the simulation:

Additional details on the linkages between submodels are in SI Appendix.

We limit simulations to 20 y to explore the within-generation results of the model to see what happens when aquatic vegetation harvest is introduced, in particular, if vegetation harvest becomes a sustained behavior, resulting in new levels of (reduced) equilibrium infections and (higher) household incomes. This time frame is long enough to capture any short-term changes in the equilibrium level of schistosomiasis infection but allows us to abstract away from long-term changes, including through impacts on children’s educational attainment, or in human fertility behaviors that would further complicate the model.

We simulate the model in Julia 1.6.2 and aggregate and analyze the model output in Stata 16. For each household type, we conduct 1,000 stochastic simulations to capture different optimal paths based on the realized random infection draws. Household types are determined by land holdings, which are set at the 25th, 50th, and 75th percentiles of land holdings in the Saint Louis and Louga regions based on the Harmonized Survey on Household Living Standards in Senegal 2018–2019 (SI Appendix, Table S1) (48). Land holdings are proxies for wealth in this context and these simulations. Comparisons across land holding types give insight into how wealth levels impact the optimal decisions of the household. We track the following key outcome variables: household labor availability, labor allocated to food production, leisure, fertilizer use, the vegetation load in the water source, the household’s level of infection, and the household’s income. We then take the median of 1,000 simulations for each outcome at each time period for each household land endowment.

To begin, we eliminate the household’s option to remove vegetation and produce compost by mechanically setting the marginal product of labor in aquatic vegetation harvest to zero. This lets us model how households currently behave and establish starting levels of infection and income under current conditions.

After solving the model and running the simulations described above, we generate back-of-the-envelope estimates of the net income gains from aquatic vegetation removal at scale. We emphasize the many strong simplifying assumptions in these estimates, which are meant to give a sense of the magnitude of the prospective gains from diffusing aquatic vegetation removal broadly across West Africa.

To generate those estimates, first we calculated the difference in median income with and without aquatic vegetation removal for each of our representative households after 5 y of vegetation removal. Then, we scaled these estimates to the village level. We assume that villages have 985 people (the average village size for this region for villages within five kilometers of a water source) (49–51) with an average household size of 10, so on average 98.5 households. Then, we assume that the 0.5, 2.0, and 5.5 ha households represent 25%, 50%, and 25% of total households in the average village, respectively, enabling us to scale up to village level the median increase in household income from aquatic vegetation removal conditional on land holdings. At year 5, aquatic vegetation removal produces estimated additional average income gains of 3,197 USD per village per year. These estimates assume that all households know about aquatic vegetation removal and the potential economic benefits from making compost.

Finally, we used remote sensing imagery analysis to scale to all of West Africa. We estimate that there are 43,320 villages across West Africa that are within 5 km of surface freshwater and likely to host the vegetation, snails, and schistosomiasis (49–51). This number was then multiplied by the estimated net gains per village estimated from Senegal.

A replication package with the code and data used is available at https://github.com/mdoruska/Bioeconomic_Model.

For helpful comments and discussions, we thank Ed Barbier, Brian Dillon, Katie Fiorella, Chris Haggerty, Nicolas Jouanard, Kira Lancker, Shanjun Li, Sean Moore, Alex Perkins, Ivan Rudik, Alex Sacks, Burton Singer, Meghan Forstchen, Alexander Timpe and seminar participants at Cornell University. We thank Florio Arguillas and the staff at the Cornell Center for Social Sciences for their assistance verifying and compiling the replication package associated with this paper. Any errors are the authors’ sole responsibility. This research was supported by NSF Grants DEB-2109293 and BCS-2307944 and the Indiana Clinical and Translational Sciences Institute.