Articles

PUBLISHED: 20 march 2017 | VOLUME: 1 | ARTICLE NUMBER: 0108

The rise and fall of malaria under land-use change

in frontier regions
Andres Baeza1,2†
​ *, Mauricio Santos-Vega3†, Andrew P. Dobson4,5​ and Mercedes Pascual3,5​

Land-use change is the main force behind ecological and social change in many countries around the globe; it is primarily driven by
resource needs and external economic incentives. Concomitantly, transformations of the land are the main drivers for the emer-
gence and re-emergence of malaria. An understanding of malaria population dynamics in transforming landscapes is lacking,
despite its relevance for developmental and public health policies. We develop a mathematical model that couples malaria epide-
miology with the socio-economic and demographic processes that occur in a landscape undergoing land-use change. This allows us
to identify different types of malaria dynamics that can arise in early stages of this transformation. In particular, we show that an
increase in transmission followed by either a decline, or a further enhancement, of risk is a common outcome. This increase results
from the asymmetry between the relatively fast ecological changes in transformed landscapes, and the slower pace of investment
in malaria protection. These results underscore the importance of reducing ecological risk, while providing services and economic
opportunities to early migrants for longer periods. Consideration of these socio-ecological processes and, more importantly, the
temporal scale on which they act, is critical to avoid potential bifurcations that lead to long-lasting endemic malaria.

H
uman-induced land-use changes driven by road construc-
tion, deforestation, agriculture and irrigation are strongly
linked to outbreaks of a range of infectious diseases, from
emergence of new pathogens to altered transmission in established
infections1,2. The most striking example of this interaction is the
early emergence of Plasmodium falciparum malaria, which is intrin-
sically linked to the development of frontier agricultural settle-
ments3,4. The rapid adaptation of Anopheles vectors and Plasmodium
parasites to environmental changes, added to the multiple ways in
which humans quickly modify landscapes, continue to contribute
to the expanded distribution of malaria throughout the developing
frontiers of the world3.
Frontiers can be defined as large, previously undeveloped, geo-
graphic areas that are experiencing large-scale patterns of land
transformation5. The economy of frontiers is labour-intensive
and characterized by low investment in public services for early
migrants6, often resulting from decisions to maximize short-term
economic return, incentivized by external forces such as govern-
ment investment and colonization policies6. Throughout the tropics,
land transformation is the main driver of economic development;
the speed with which it progresses hinges on environmental, demo-
graphic and social changes7,8.
Deforestation in frontier regions such as the Amazon has cre-
ated open areas with abundant sunlight and rainfall water. Altered
microclimatic and hydrological conditions have, in turn, modified
demographic and life-cycle parameters of the mosquitoes, such as
longevity and larval developmental rates, as well as the developmen-
tal rate of the parasite within the vector9–13. Changes in the human
demography of frontier regions can also affect malaria transmis-
sion by changing human–mosquito interactions, mainly through
increasing contact and biting rates14.
So far, studies on malaria and land-use change have focused
primarily on understanding the ecological effects that accompany
land transformation15–17. Much less is known about the interaction
between epidemiology and the economic and demographic pro-
cesses that accompany these ecological effects. Similarly, economic
analyses used to support large-scale environmental transformations
of frontier regions rarely consider long-lasting investment in reduc-
ing vector-borne diseases, and the dynamical interactions between
the economic processes and associated ecological and epidemio-
logical feedbacks.
This study develops a novel hybrid model that incorporates
processes both from ecological epidemiology and economics
to define a framework in which we can examine the feedbacks
between economic development and the transmission of malaria
and other vector-borne diseases. We have specifically used it to
provide a synthetic perspective of the interplay between malaria
population dynamics and the economy in areas of the tropics expe-
riencing rapid land-use change. The model couples the epidemiol-
ogy (mosquito–parasite–host) and labour-intensive economy of a
landscape experiencing land transformation from an undeveloped
state towards a more developed state. It is based on observations
from ecological, economic and demographic changes documented
for two case studies: (1) the deforestation of the Brazilian Amazon
region18; and (2) the expansion of irrigated agriculture in the desert
of northwest India19. We characterize and explain the different types
of malaria population dynamics that can occur in these regions at
the early stages of land transformation, and identify the dominant
factors that drive these interactions. The structure of the model is
shown in Fig. 1 and its equations and assumptions are described in
the Supplementary Information; here, we describe the main insights
gained from numerical solutions for a broad range of characteristic
parameter values. We specifically focus on emphasizing the short-
term, or transient, temporal trajectory between two equilibria: pre-
and post-land transformation (Fig.  1). Thus, instead of focusing
only on the long-term equilibrium behaviour of the system, which

National Socio-Environmental Synthesis Center, University of Maryland, Annapolis, Maryland 21401, USA. 2School of Sustainability, Arizona State
1

University, Tempe, Arizona 85281, USA. 3Department of Ecology and Evolution, University of Chicago, Chicago, Illinois 60637, USA. 4Department of
Ecology and Evolutionary Biology, Princeton University, Princeton, New Jersey 08544, USA. 5Santa Fe Institute, Hyde Park Road, Santa Fe, New Mexico
87501, USA. †These authors contributed equally to this work. *e-mail: andres.baeza@gmail.com

nature ECOLOGY & EVOLUTION 1, 0108 (2017) | DOI: 10.1038/s41559-017-0108 | www.nature.com/natecolevol 1

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles
a b c d
Steady state Pertubation Short term Long term
Land transformation
Undeveloped (U) Transformed (D)
KD
KU
υD
υU X W
X W Immigrants
(1 – ν) ρ(CD)
βU μ
ν
βD μ We first applied a multinomial logit regression analysis that uses the
S I R S I R
ετ ετ
σ
λ(CD)
σ m
Figure 1 | Graphical description of the processes of land transformation
and their association with malaria. a, A region without intervention
where malaria transmission is generally at a stationary state (blue line)
or seasonally varying equilibrium (red line). b, The beginning of the land
transformation is represented, with land use depicted by the opening of
a road, which is in turn driven by the external decisions of governments
or other parties. These changes can create environments that are more
favourable to mosquitoes, including many vectors of malaria. From an
economic point of view, newly converted areas are likely to be more
productive because of technological changes and/or the intensity of
exploitation. c, Given these opposite forces potentially acting in the initial
stages of land use, we hypothesized that malaria risk and its prevalence can
follow different short-term trajectories (shown with different lines). d, These
trajectories would in turn influence the population dynamics of malaria and
the final state of the system, when the landscape is completely transformed.
The lower figures illustrate the dynamic model used to describe malaria
dynamics at different stages of development. β and υ represent the force of
infection from susceptible (S) to infected (I) human, and from healthy (X) to
infected (W) mosquitoes, respectively. Infected individuals gain immunity by
naturally recovering at rate μ and lose it at rate σ. In transformed landscapes,
a new economically protected class is included. The rate at which individuals
become economically protected, ρ, is a function of capital (C ). This
d
protection can be lost at a rate inversely proportional to capital (λ(Cd)).
may take several decades to establish, we describe transitions on the
timescale of months and years, when most of the economic, demo-
graphic and environmental interactions are changing simultane-
ously (Fig. 1).
Numerical simulations were undertaken with different param-
eter values, sampled using a Latin hypercube method in a large
range of transmission intensities (see Methods); six different types
of malaria population dynamics were identified (Fig. 2; see Methods
and Supplementary Table 7). These were labelled from the most to
least common as the following: (1) convex decrease; (2) convex
increase; (3) concave decrease; (4) concave increase; (5) monotonic
decrease; and (6) monotonic increase. Convex decrease, the most
2 nature ECOLOGY & EVOLUTION 1, 0108 (2017) | DOI: 10.1038/s41559-017-0108 | www.nature.com/natecolevol
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Nature Ecology & Evolution
common outcome, reflects a situation in which malaria prevalence
increases rapidly and then falls to levels lower than observed pre-
transformation (Fig.  2 panel 3). In this case, the maximum inci-
dence is achieved at intermediate times between the start and end of
the simulated land transformation. This transient regime implies a
worsening of the disease burden before an improvement is obtained.
The second most commonly observed type of dynamics was the
convex increase, which is also a situation with an initial increase
in incidence, followed by a decrease that is not sufficient to reduce
disease burden back to the low levels observed pre-transformation
(Fig.  2f). The two dynamics labelled concave (decrease and
increase) are the opposite of the convex dynamics; that is, they
generate an initial reduction in incidence, followed by an increase
(Fig. 2c,e). These two types were observed least frequently (Fig. 2h).
Finally, none of the 10,000 simulations showed a constant mono-
tonic decrease or increase in incidence (the examples in Fig. 2a,d
are from simulations labelled ‘concave decrease’ or ‘convex increase’
that look very similar to monotonic decrease or increase dynamics
due to their long period of transition).
To investigate the most influential factors explaining the differ-
ences in the trajectories, we conducted a set of sensitivity analyses
on some characteristic quantities of the simulation outputs that
defined each pattern depicted in Fig.  2 (Supplementary Table 7).
P observed dynamics as a nominal variable and the parameter values
that generated these dynamics as the ‘independent’ predictors (see
Methods). The results imply that differences in malaria dynamics
are explained by a combination of socio-economic factors directly
related to malaria investment and the ecological variables acting in
early stages of the transformation (Fig. 3a,b; Supplementary Table 8).
We further investigate the mechanisms that more precisely divide
the convex trajectories from the other patterns, by using a gen-
eral linear model analysis assuming a Boolean response variable
y =​ {0,1}, with 1 for convex dynamics and 0 otherwise. The results
suggest that environmental conditions predominantly create the
initial state of high malaria risk in early stages of land-use change
(Fig.  3b; Supplementary Table 9). These ecological effects then
interact with the socio-economic factors that tend to reduce risk on
slower, and longer, timescales (Fig. 3a). The tension between these
stabilizing and destabilizing forces generates the convex rise and
fall in transmission (Figs 2b and 4). A more detailed analysis of the
simulations resulting in convex patterns (Fig. 3b; Methods) suggests
that the magnitude of this transition phase is determined by the
rapid ecological changes that enhance malaria risk in newly trans-
formed areas (higher carrying capacity and biting rate, lower level of
immunity of the migrants) and by the socio-economic factors that
increase capital flow to malaria protection in the short term (treat-
ment), as well as the long term (long-lasting physical protection).
Finally, we statistically compare the levels of malaria incidence
pre- and post-transformation to explain the long-term, asymptotic
decay in malaria over longer timescales (Methods; Supplementary
Table 10). Specifically, we compare the frequency of simulations
with a convex decrease and that of simulations with a convex
increase (Supplementary Table 10). These long-term differences
are mostly explained by the asymmetric and synergistic interaction
between economic parameters, such as the rate of economic invest-
ment in malaria treatment, the price of the crops, the saving rate and
the difference in transmission due to the ecological conditions in
the newly transformed land (Fig. 4; Supplementary Table 10). Lower
levels of investment in protection and control against the disease
determine the post-transformation malaria prevalence as well as
the persistence of malaria during land transformation (Fig.  4a,b).
These asymmetries in the rates at which increased economic pros-
perity is invested into malaria protection interact with the under-
lying ecological risk to generate non-linear dynamics, leading to
increased malaria risk over periods longer than a decade (Fig. 2f).
Nature Ecology & Evolution Articles
a d g
Monotonic decrease Monotonic increase

0.3 0.8 Decrease

Increase
0.0 0.0
1 4 7 10 2 7 13 20 27
Time (yr) Time (yr)

b e h
Convex decrease Concave increase
Cases per person

0.6 0.5 Convex

Monotonic
Concave
0.0 0.0
2 5 8 12 05 20 35 55 70
Time (yr) Time (yr)

c Concave decrease f Convex increase i Convex

0.40 0.20 decrease

Concave Monotonic decrease

increase Concave decrease
0.20 0.00 Convex
2 5 8 12 1 4 7 increase
Time (yr) Time (yr)

Figure 2 | The typology of malaria incidence under land transformation. a–f, The six different types of malaria population dynamics (total cases per
person) obtained for the early stages of land transformation in the simulations. The green area represents the equilibrium stage before the process of
transformation begins. g, The pie chart shows that the most common outcome is a reduction in risk. An increased risk encompasses all the simulations
that resulted in higher incidence post- than pre-transformation (examples shown in d–f). h, The pie chart shows that the most common dynamics are
those described as convex (examples in panels b and f). i, More specifically, this pie chart shows that the most common type of dynamics is the one that
displays an increase in prevalence in early stages of development, followed by a decay (convex decrease). When this decay generates a higher endemic
level than before the land transformation, we label this behaviour as convex increase.

This non-linear transient phase arises even when economic growth
is itself monotonic and positive.
Discussion
The term ‘frontier malaria’ has been adopted to describe temporal
and spatial changes in malaria risk that follow environmental, social
and demographic changes of previously undeveloped areas that expe-
rience large-scale land-use transformations18. Our model attempts to
capture the intertwined effects of these socio-ecological modifica-
tions, to explain the transient trajectories of malaria incidence over a
decade or more. We show that a transient stage of enhanced malaria
risk followed by a decay over a decade is a common outcome of the
feedbacks between economic and epidemiological processes. These
results are consistent with empirical observations in the Amazon
frontier, where many areas have experienced extensive deforestation
due to an increase in new settlements following a colonization policy
in effect since the 1980s20. These studies have documented the rise of
malaria risk due to an increase in breeding sites, biting rates and con-
tact rates. They also describe the subsequent reduction in risk over
longer timescales to improved housing, and better infrastructure for
public health; this is posited to produce long-lasting benefits of stron-
ger economic conditions and community cohesion21–23. Our results
are also consistent with the patterns described in the semi-desert
region of northwest India, where large investments in water distri-
bution (canals and dams) have transformed dry regions of Gujarat
and Rajasthan into a green desert with higher carrying capacity for
mosquitoes and higher contact rates with humans. Higher levels of
risk in this region were located in areas in early stages of irrigated
agriculture, compared with areas in a ‘mature’ state of economic and
social development19. Our model provides a mechanistic basis to
support the similarities observed in these empirical studies by link-
ing the transitional stages of high malaria risk to the antagonistic
forces of economic and ecological trends.
The numerical and sensitivity analyses of the new hybrid eco-
nomic–epidemiological models we have developed suggest there is
no single variable or parameter that by itself explains the rise and
fall of malaria risk. Instead, we were able to identify a subset of fac-
tors that together help to explain the formation of asymmetries in
the system, characterized by rapid ecological change and slower eco-
nomic development. These differences in velocities and direction
create tensions between the ecological and economic components of
our model that can lead to rapid and abrupt increases in mosquito
populations and human–mosquito contact rates. These interact syn-
ergistically to increase malaria transmission early on in the process.
Socio-economic factors that play a role in modulating malaria risk
(for example, improved dwellings24, drainage and water supply infra-
structure25, knowledge about the disease and preventing its trans-
mission26, and the accumulation of other forms of social capital) are
factors that accumulate at slower rates than those of the ecological
processes (for example, mosquito growth, parasite development). An
understanding of the differences in velocity with respect to the stage
in which a system is found has implications for control strategies. By
developing a hybrid model framework that includes both economic
and ecological/epidemiological processes, we are able to identify dif-
ferent time periods of development when different strategies will
be more effective in controlling malaria. In early stages, mosquito
control and reducing transmission are probably the most effective
measures to avoid potential long-lasting transmissions with increase
in incidence. In contrast, at later stages of development when malaria
is already established, long-lasting measures that can increase the
rate of ‘economic protection’, such as better housing, can be more
effective. Short-term strategies are more easily available at relatively
low costs, while long-term strategies are costly and more robust, and
have many side-benefits that create important trade-offs in long-
term cost–benefit analysis. Making these trade-offs explicit with
respect to the process of economic, ecological and epidemiological
maturity of a region can help tailor the most cost-effective strategies
to achieve and sustain a stable state of malaria elimination27.
We also showed that a trajectory with an initial stage of high
risk is not always followed by a decay, but that it can settle at a higher

nature ECOLOGY & EVOLUTION 1, 0108 (2017) | DOI: 10.1038/s41559-017-0108 | www.nature.com/natecolevol 3

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles Nature Ecology & Evolution
a
0.2
pcc

0.0

−0.2
% migrants with immunity

Average mosquito lifetime

Average time in infected class

Capital depreciation

Carrying capacity

Cost per area planted

Debilitation effect

Ecological differences

Effectiveness of treatment

Elasticity alpha

Elasticity beta

Factor of productivity

Gain in economic protection

Gonotrophic/feeding cycle

Human blood index

Immunity

Investment in treatment per case

Loss economic protection

Mosquito fecundity

Population density in D

Population density in U

Price of crop

Productivity differences

Rate of land transformation

Saving rate

Transmission efficiency H−M

Transmission efficiency M−H

Human mobility
Parameters

b 0.3
0.2
0.1
pcc

0.0
−0.1
% migrants with immunity

Average mosquito lifetime

Average time in infected class

Capital depreciation

Carrying capacity

Cost per area planted

Debilitation effect

Ecological differences

Effectiveness of treatment

Elasticity alpha

Elasticity beta

Factor of productivity

Gain in economic protection

Gonotrophic/feeding cycle

Human blood index

Human mobility

Immunity

Investment in treatment per case

Loss economic protection

Mosquito fecundity

Population density in D

Population density in U

Price of crop

Productivity differences

Rate of land transformation

Saving rate

Transmission efficiency H−M

Transmission efficiency M−H

Parameters

Figure 3 | Summary of statistical analyses of simulation outputs. Results from sensitivity analyses of the model’s parameters are shown. a, The analysis is
conducted using I*diff, the difference in incidence (cases per person) pre- and post-transformation of the land (respectively, I*pre and I*post), with I*diff =​ I*post −​ I*pre.
Thus, positive values indicate an increase in malaria after the transformation. b, The analysis was conducted using instead I*peak−diff, the difference
between the initial incidence level (I*pre) and the magnitude of the peak incidence after the transformation (I*peak), for the simulations classified as convex
trajectories, such that I*peak−diff =​ I*peak −​ I*pre. Thus, positive values of the partial correlation coefficients (pcc) now indicate an increase in the magnitude
of malaria’s burden in the transient state. Red bars indicate parameters that are influential to increase malaria at the equilibrium in a, or in the transient,
in b. Green bars indicate parameters that reduce incidences. Larger bars with darker tones indicate higher influence of the parameter. The names of the
most significant parameters are also, respectively, coloured in red or green, depending on their effect. a shows that economic parameters are the most
influential ones in explaining the long-term dynamics of malaria, while b identifies ecological factors and the associated differences that emerge between
the two landscapes (Δ​Env in Supplementary Table 5) as determining the magnitude of the short-term increase. Thus both contribute synergistically to the
temporal rise and fall of malaria characteristic of the most common type of transient dynamics identified in Fig. 2.

disease burden than those preceding land transformation, particu- should be allocated to alleviate and shorten the intermediate phase
larly when economic conditions do not improve for new migrants or of development when malaria transmission is at its peak. A conse-
are not maintained over time. Critically, this depends on the role the quence of ignoring this mismatch of scales is the possibility of unde-
economy plays in developing strategies to protect against malaria. sirable long-term outcomes of low economic development and high
These results have three major implications for colonization and disease burden29,30.
development strategies in frontier regions prone to malaria. First, Our model tracks only global, homogenous quantities of
in early stages of land transformation it is critical to provide effec- capital and disease, without considering the socio-economic
tive early malaria control to counteract the rapid rise in risk due inequalities that may emerge during the transformation of the
to environmental change, especially when settlements are still not land in both space and time. In particular, economic differences
well-equipped and have not acquired much capital to invest in self- among productive systems (for example, large cattle ranchers ver-
protection28. Second, it is necessary to accompany these environ- sus small farming) may generate inequalities in the capital avail-
mental control strategies with better conditions for early migrants, able for providing physical protection and treatment, helping to
such as provision of good-quality dwellings that offer physical maintain higher levels of malaria risk for longer periods of time.
and, more importantly, long-lasting protection against mosquitoes Future work should consider these potential differences in eco-
in areas of indoor transmission. Sustained investment in public nomic state by dividing the host population into different socio-
health infrastructure and health education for longer periods than economic classes, or by formulating analogues to the implicit
those typical of short mitigation programmes should help main- parameterization of unequal parasite burdens in models for the
tain protection over time. Finally, identifying the temporal scales transmission of macro-parasitic diseases.
of environmental change and socio-economic processes would Global changes in the distribution of malaria over a century31
help determine the period of time for which additional investments have revealed an overall decrease in the burden of the disease,

4 nature ECOLOGY & EVOLUTION 1, 0108 (2017) | DOI: 10.1038/s41559-017-0108 | www.nature.com/natecolevol

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Nature Ecology & Evolution Articles
a Investment in b Gain economic c Treatment
malaria protection effectiveness
0.300 0.35

0.275
0.30 0.30

0.250
0.25
0.225 0.25

0.20
0.200
Cases at equilibrium

0.0 0.5 1.0 1.5 −9 −8 −7 −6 −5 −7 −6 −5 −4 −3

d e Ecological f HBI
Carrying capacity
differences
0.26 0.30

0.24 0.25 0.24

0.22 0.20
0.20
0.20
0.15 0.16
0.18
0.10 0.12
−7.0 −6.5 −6.0 −5.5 −5.0 −4.5 0.0 0.5 1.0 1.5 −2.0 −1.5 −1.0 −0.5 0.0
Parameter value

Figure 4 | Relationship between cases per person at equilibrium and the values of significant parameters in determining convexity. a–f, The plots show
the relationship between the cases at final equilibrium post-transformation (I*post) and the values (in logarithmic scale) of the most significant parameters
in determining convexity identified by the sensitivity analyses (Supplementary Tables 9 and 10 for units of parameters). The effect of investment in
malaria medication (a; parameter τ in Supplementary Table 4) and the cost-effectiveness of the treatment (c; ε in Supplementary Table 4) are shown.
b, The relationship between cases per person and the rate at which people acquire protection against malaria, due to the overall improvement of economic
conditions (parameter ρ(C) in Supplementary Table 4). d, The effect of the carrying capacity (Ki) of adult mosquitoes per unit of area pre-transformation of
the land. e, The magnitude of the change in environmental conditions for malaria in post-transformed regions (parameter ΔEnv in Supplementary Table 5).
f, The human blood index, which represents the proportion of blood meals from humans by a mosquito (parameter HBI in Supplementary Table 2). In all
panels, the blue line represents the mean incidence calculated from the simulations, and the shaded regions correspond to the confidence intervals.

mostly in temperate zones of Europe and North America, regions
that have experienced high economic growth. The results presented
here underscore that different types of dynamics can be observed
on regional or local scales during the transition to the desired
long-term outcome, and that the path to this outcome may cross
higher disease levels, a phase that is transient but not necessarily
short. Increasingly, many of the world’s poor are trapped in this
transient phase, particularly in countries where malaria is endemic.
More generally, ‘short-term’ and ‘long-term’ dynamics are likely to
differ. Investigating and disentangling the causes of these differences
is of central importance to generate more accurate, effective inter-
ventions in countries experiencing periods of rapid development.
Identifying the critical paths and feedbacks that prolong higher
health risks is ultimately the key to reducing malaria by allowing
people to escape poverty in marginalized areas of the world.
Methods
General background for model building. We begin by contextualizing the socio-
economic, demographic and ecological changes that accompany the concurrent
transformation of the land in frontier regions, and we outline how those changes
influence what we know about malaria transmission. High rates of land-use
change are usually driven by external actors that provide basic infrastructure for
catalysing land conversion. For example, central governments create highways to
incentivize migration to undeveloped regions, or create investments in canal dams
and other forms of water distribution to bring water and irrigation to dry regions
(Fig. 1). Investment in such basic infrastructure provides the basis for subsequent
socio-ecological changes relevant for malaria transmission. The changes that have
been documented after initial perturbations of previously undeveloped areas
concern microclimatic and hydrological processes that, in turn, influence key
epidemiological parameters such as feeding and reproductive behaviour, biting
rate, carrying capacity, mosquito longevity, and parasite development10,13,32–34.
Following these ecological disruptions, demographic changes are observed in
frontier regions, with important consequences from an epidemiological perspective.
Newly transformed areas are usually more productive, at least in the short term,
due to the influx of new technologies. In most cases, the population of the newly
transformed areas begins to acquire more resources than the local people who
previously lived there: more water in the case of irrigation infrastructure, or new
technological assets (such as tractors or electric saws) in the forest. A higher
expectation of productivity attracts people hungry for land and for better prospects
of income. As a consequence, human population usually increases, with a high
proportion susceptible to the malaria parasites. Seasonal and daily human movement
is also commonly observed in frontiers, as people search for temporal jobs. This
movement can have important influences on the spatial patterns of transmission35.
The productive systems that early migrants rely on when transforming the
land hinge on labour-intensive activities such as logging, farming, or ranching.
Over longer time intervals, as settlements develop and age from villages to
towns, physical capital increases and primary infrastructure develops. As human
settlements grow and become more established, different forms of social and
human capital tend to accumulate, and the overall socio-economic state tends
to improve, thereby reducing human exposure to mosquitoes through better
housing, more access to drugs and other forms of protection such as bed-nets and
insecticide36. Examples of how economic prosperity can influence the decline in
malaria population have been documented for rural areas of the USA during the
elimination era (1900–1950s)21,37,22 and for other temperate areas of Europe23. These
studies have shown that the effect of economic prosperity can be decomposed
into long- and short-term effects on malaria dynamics. A long-term effect owing
to social and demographic changes occurs at a slower pace, such as improved
housing conditions or changes in the number of people per dwelling. A short-term
influence is usually related to investment in control and medication. The model,
described below and in the Supplementary equations, captures the long-term
effect of prosperity by defining a different epidemiological class—the economically
protected class—which effectively creates a form of ‘economic herd immunity’ and
a short-term effect from investment in malaria medication that reduces
the duration of infection.
The model. We have developed a novel hybrid economic–epidemiological
mathematical model based on the generalities described above. The model
considers the interactive dynamics of malaria and the human economy in a
region that begins as an undeveloped state (AU). We assume that before any
transformation occurs, the untransformed system is already at a steady state,

nature ECOLOGY & EVOLUTION 1, 0108 (2017) | DOI: 10.1038/s41559-017-0108 | www.nature.com/natecolevol 5

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles
with low economic productivity (CU*) and malaria prevalence (IU*) for a local
population whose economy is based purely on harvesting local forest products
(Fig. 1a) (or on limited agriculture in a dry landscape). Development proceeds as
the land is transformed at a daily rate (l) into a developed state (AD). Simultaneous
to the transformation of the land, an epidemiological human–mosquito–malaria
model and a macroeconomic model that mimics a labour-intensive productive
system are also at play, acting in a dynamic and interconnected way. The model
is written as a system of differential equations with three modules per region
that recreate the processes shown in Fig. 1 and in Supplementary Fig. 1. See the
Supplementary Information for mathematical details.
The first module describes the epidemiological changes in the human
population by dividing it into four classes, three of which are found in both
undeveloped and developed land states: susceptible (Si), infected (Ii), recovered
(Ri), with i =​ {U,D}. Although the structure of the model was motivated by malaria
transmission in a region of low to medium transmission intensity, the average
time a person spends in the immunity class (1/σ) was allowed to span periods
long enough to represent the effect of repeated exposure. A fourth category, the
economically protected class (PD) inhabits only the developed region (Fig. 1).
This category represents a subpopulation not at risk of contracting malaria due
to the overall improvement in socio-economic conditions, and captured by the
accumulation of capital. People can become economically protected as a function
of this capital (ρ), but they can also lose this protection if capital is reduced through
poverty. They also can invest in malaria medication (τ), which reduces the average
time a person stays in the infectious class, as a function of the cost-effectiveness (ε)
of the treatment.
The second module tracks the dynamics of the mosquito population,
subdividing the vector population into uninfected (Xi) and infected (Wi)
individuals. The total mosquito population can grow until reaching a carrying
capacity (Ki), proportional to the amount of land available; they become infectious
at a rate proportional to the rate at which they bite the fraction of the human
population in the infected state (Fig. 1).
The last module represents an agricultural, labour-intensive productive system
that takes into account available land, capital, labour and technology to generate
an aggregated quantity of production. This productive system is represented by
a Cobb–Douglas (C–D) equation, a standard macroeconomic model that can be
parameterized from empirical data, and allows the inclusion of different factors
that can internally influence the economic productivity of a large region, such
as land (Ai), capital (Ci) and labour (Li) 38–40. The C–D formulation also allows us
to include a total factor of productivity, which can depend of externalities and
drivers that may influence economic productivity, such as political or technological
changes. We modify the classical form of the C–D equation to account for the
possible debilitating effects of malaria on productivity (ω)41. The transitions
between classes and the coupling between sub-regions are represented in the
flow diagram in Supplementary Fig. 1. The model also captures the demographic
changes that have been documented in frontiers. Specifically, we account for
the movement (mi) of people between sub-regions (AU and AD) to mimic daily
mobility of the human population, habitat migration that includes an influx of
people to newly transformed areas (NM) and the susceptibility of the new migrant
population. This last process was represented by a parameter that controls
the percentage of new migrants that are susceptible or immune to malaria (v;
Supplementary mathematical details).
Our model assumes that the values of some of the economic and ecological
parameters are not the same in the two sub-regions AU and AD. Specifically, we
assume that the total factor of productivity (ϕi; Supplementary Table 1) can take
higher values in newly developed regions compared with untransformed regions
(Δ​ϕ; Supplementary Table 5). We also consider that the carrying capacity of the
mosquito population (Ki) and the biting rate (represented in the model by the
human blood index, HBI, and feeding rate, g; Supplementary Table 3) are higher
post-transformation (Δ​Env; Supplementary Table 5; Supplementary mathematical
details). However, when we relaxed this assumption to consider situations in
which, for example, the abundance of mosquitoes increases but the biting rate is
reduced, similar results were obtained (Supplementary Fig. 2).
Parameter selection. Instead of simulating the model for a particular set of
parameter values, we conducted 10,000 simulations over a wide range of economic,
ecological and demographic parameters by sampling combinations of values from
plausible ranges, with the Latin hypercube sampling method implemented in R.
Basic epidemiological and entomological parameters were obtained from empirical
and statistical estimations conducted in rural areas of Brazil, Africa and India;
they represent a broad range of transmission intensities resulting from variation
in average immunity duration, infectious period, transmission efficiency, biting
behaviour, vector longevity and the rate of parasitic development within the
mosquitoes. We also obtained parameter ranges for agricultural productivity and
for the prices of important crops in India using the publically available statistical
agricultural annual book42. Finally, we set the structural parameters of the C–D
function to always represent a labour-intense economy. For the set of parameters
from empirical sources or statistical inference, we constructed their range based
on reported confidence intervals. In the absence of literature information,
the ranges were chosen to be broad enough to include all plausible values.
6 nature ECOLOGY & EVOLUTION 1, 0108 (2017) | DOI: 10.1038/s41559-017-0108 | www.nature.com/natecolevol
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Nature Ecology & Evolution
Parameters without information are mostly those connecting economic factors
with malaria epidemiology. The range of parameter values, their sources and the
details of these choices are presented in Supplementary Tables 1–6.
Numerical experiments and typology of dynamics. We ran the model for each
combination of parameters without transformation (l =​ 0) for 100 years, a sufficient
time for the system to reach a stable state regardless of parameters. Once in the stable
state, we disturbed the system by setting the rate of land-use change to a constant
value of l >​ 0 (see Supplementary Table 5 for values of l), and we ran the model for
sufficient time to capture the transient response to the perturbation (100 years).
We considered only simulations that generated a proportion of infected individuals
and mosquitoes in the undisturbed landscape of greater than 0. The maximum and
minimum levels of incidence in each simulation were recorded together with the
time at which the system reached these values and the final equilibrium. Based on
this information, we classified and described the types of dynamics exemplified in
Fig. 2 (see Supplementary Table 7).
Sensitivity analyses of model outputs as a function of parameters. To evaluate
the influence of the model parameters on defining the typology of dynamics, we
conducted a set of formal sensitivity analyses, using the epidemiological outcomes
(the different trajectories) as independent variables. We first applied a multinomial
logit regression model (package mlogit in R) where the dependent (categorical)
variables are the different types of dynamics (1: monotonic decrease; 2: monotonic
increase; 3: convex decrease; 4: concave decrease; 5: convex increase; 6: concave
increase) and the independent variables are the parameters of the model for the
different simulations. The parameters that explain most of the variance between
types of dynamics were selected by conducting a backwards stepwise model
selection based on the Akaike information criterion, implemented in the MASS
package in R. To evaluate factors that differentiate the dynamics labelled convex
from the rest, we carried out a general linear model analysis with a binomial error
distribution and a probit function to link the values of these selected parameters to
a Boolean response variable y =​{0,1}, with 1 for convex dynamics and 0 for the rest.
Finally, a linear regression was used to analyse the long-term differences in malaria
risk pre-and post-land transformation for those simulations labelled as convex.
Specifically, we computed the difference between the cases of malaria per person
before and after the land transformation, at the end of the 100 years of simulations
(I*diff in Fig. 3a). The results of this regression are shown in Supplementary Table
10. To construct the bar plot depicted in Fig. 3a, we calculated the sensitivity
of I*diff to each parameter using a standardized scale (Fig. 3a,b). We conducted
a partial correlation analysis43 using the pcc function embedded in the package
‘sensitivity’ in R44. To construct the bar plot in Fig. 3b, we repeated the analysis
using the difference between the initial incidence level (I*pre) and the magnitude
of the peak incidence after the transformation (I*diff-peak in Fig. 3b).
Data availability. The system of differential equations was written in R and can
be accessed at the following link, also including the scripts to run the analyses of
simulation outputs: https://github.com/abaezacastro/Malaria-Land-Use-Model.
received 24 June 2016; accepted 3 February 2017;
published 20 March 2017
References
1. Patz, J. A. et al. Unhealthy landscapes: policy recommendations on land use
change and infectious disease emergence. Environ. Health Perspect.
112, 1092–1098 (2004).
2. Gottdenker, N. L., Streicker, D. G., Faust, C. L. & Carroll, C. R.
Anthropogenic land use change and infectious diseases: a review of the
evidence. Ecohealth 11, 619–632 (2014).
3. Livingstone, F. Malaria and human polymorphisms. Annu. Rev. Genet.
5, 33–64 (1971).
4. Escalante, A. A. & Ayala, F. J. Phylogeny of the malarial genus Plasmodium,
derived from rRNA gene sequences. Proc. Natl Acad. Sci. USA 91,
11373–11377 (1994).
5. Bryant, D., Nielsen, D. & Tangley, L. Frontier Forests: Ecosystems and
Economies on the Edge (World Resource Institute, 1997).
6. Shriar, A. J. Agricultural intensity and its measurement in frontier regions.
Agrofor. Syst. 49, 301–318 (2000).
7. Young, J. Along Ethiopia’s western frontier: Gambella and Benishangul in
transition. J. Mod. Afr. Stud. 37, 321–346 (1999).
8. Lambin, E. F. & Meyfroidt, P. Land use transitions: socio-ecological feedback
versus socio-economic change. Land Use Policy 27, 108–118 (2010).
9. Afrane, Y. A., Lawson, B. W., Githeko, A. K. & Yan, G. Effects of
microclimatic changes caused by land use and land cover on duration of
gonotrophic cycles of Anopheles gambiae (Diptera: Culicidae) in western
Kenya highlands. J. Med. Entomol. 42, 974–980 (2005).
10. Afrane, Y. A., Zhou, G., Lawson, B. W., Githeko, A. K. & Yan, G. Effects of
microclimatic changes caused by deforestation on the survivorship and
reproductive fitness of Anopheles gambiae in western Kenya highlands.
Am. J. Trop. Med. Hyg. 74, 772–778 (2006).
Nature Ecology & Evolution
11. Afrane, Y. A., Githeko, A. K. & Yan, G. The ecology of Anopheles
mosquitoes under climate change: case studies from the effects of
deforestation in East African highlands. Ann. NY Acad. Sci.
1249, 204–210 (2012).
12. Vittor, A. Y. et al. Linking deforestation to malaria in the Amazon:
characterization of the breeding habitat of the principal malaria vector,
Anopheles darlingi. Am. J. Trop. Med. Hyg. 81, 5–12 (2009).
13. Vittor, A. Y. et al. The effect of deforestation on the human-biting rate of
Anopheles darlingi, the primary vector of falciparum malaria in the Peruvian
Amazon. Am. J. Trop. Med. Hyg. 74, 3–11 (2006).
14. Moreno, J. E., Rubio-Palis, Y., Páez, E., Pérez, E. & Sánchez, V. Abundance,
biting behaviour and parous rate of anopheline mosquito species in relation
to malaria incidence in gold-mining areas of southern Venezuela. Med. Vet.
Entomol. 21, 339–349 (2007).
15. Lindblade, K. A., Walker, E. D., Onapa, A. W., Katungu, J. & Wilson, M. L.
Land use change alters malaria transmission parameters by modifying
temperature in a highland area of Uganda. Trop. Med. Int. Health 5,
263–274 (2000).
16. Olson, S. H., Gangnon, R., Silveira, G. A. & Patz, J. A. Deforestation
and malaria in Mâncio Lima County, Brazil. Emerg. Infect. Dis. 16,
1108–1115 (2010).
17. Hahn, M. B., Gangnon, R. E., Barcellos, C., Asner, G. P. & Patz, J. A.
Influence of deforestation, logging, and fire on malaria in the Brazilian
Amazon. PLoS ONE 9, e85725 (2014).
18. de Castro, M. C., Monte-Mór, R. L., Sawyer, D. O. & Singer, B. H. Malaria risk
on the Amazon frontier. Proc. Natl Acad. Sci. USA 103, 2452–2457 (2006).
19. Baeza, A. et al. Long-lasting transition toward sustainable elimination of
desert malaria under irrigation development. Proc. Natl Acad. Sci. USA
110, 15157–15162 (2013).
20. Singer, B. H. & de Castro, M. C. Agricultural colonization and malaria on the
Amazon frontier. Ann. NY Acad. Sci. 954, 184–222 (2001).
21. Andrews, J. M., Quinby, G. E. & Langmuir, A. D. Malaria eradication
in the United States. Am. J. Public Health Nations Health 40,
1405–1411 (1950).
22. ter Veen, A. M. Determinants of Malaria Transmission in the United States
Between 1900 and 1946 PhD thesis, Univ. London (2005).
23. Hulden, L. & Hulden, L. The decline of malaria in Finland—the impact of the
vector and social variables. Malar. J. 8, 94 (2009).
24. Lindsay, S. W., Emerson, P. M. & Charlwood, J. D. Reducing malaria by
mosquito-proofing houses. Trends Parasitol. 18, 510–514 (2002).
25. Sharma, R., Gautam, A., Bhatt, R., Gupta, D. & Sharma, V. The Kheda
malaria project: the case for environmental control. Health Policy Plann.
6, 262–270 (1991).
26. Yasuoka, J., Mangione, T. W., Spielman, A. & Levins, R. Impact of education
on knowledge, agricultural practices, and community actions for mosquito
control and mosquito-borne disease prevention in rice ecosystems in Sri
Lanka. Am. J. Trop. Med. Hyg. 74, 1034–1042 (2006).
27. Lee, K. N. Compass and Gyroscope: Integrating Science and Politics for the
Environment (Island, 1993).
28. Utzinger, J., Tozan, Y., Doumani, F. & Singer, B. H. The economic payoffs of
integrated malaria control in the Zambian copperbelt between 1930 and
1950. Trop. Med. Int. Health 7, 657–677 (2002).
29. Bonds, M. H., Keenan, D. C., Rohani, P. & Sachs, J. D. Poverty trap
formed by the ecology of infectious diseases. Proc. R. Soc. B 277,
1185–1192 (2010).
30. Gunderson, L. H. & Holling, C. S. Panarchy: Understanding Transformations
in Human and Natural Systems (Island, 2002).
nature ECOLOGY & EVOLUTION 1, 0108 (2017) | DOI: 10.1038/s41559-017-0108 | www.nature.com/natecolevol 7
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Articles
31. Gething, P. W. et al. Climate change and the global malaria recession. Nature
465, 342–345 (2010).
32. Tyagi, B. K. A review of the emergence of Plasmodium falciparum-
dominated malaria in irrigated areas of the Thar Desert, India. Acta Trop.
89, 227–239 (2004).
33. Vittor, A. Y. et al. Linking deforestation to malaria in the Amazon:
characterization of the breeding habitat of the principal malaria vector,
Anopheles darlingi. Am. J. Trop. Med. Hyg. 81, 5–12 (2009).
34. Jaleta, K. T. et al. Agro-ecosystems impact malaria prevalence: large-scale
irrigation drives vector population in western Ethiopia. Malar. J. 12, 350 (2013).
35. Wesolowski, A. et al. Quantifying the impact of human mobility on malaria.
Science 338, 267–270 (2012).
36. Ghebreyesus, T. A. et al. Household risk factors for malaria among children
in the Ethiopian highlands. Trans. R. Soc. Trop. Med. Hyg. 94, 17–21 (2000).
37. Andrews, J. M. What’s happening to malaria in the U.S.A.? Am. J. Public
Health Nations Health 38, 931–942 (1948).
38. Barro, R. J. Government spending in a simple model of endogenous growth.
J. Polit. Econ. 98, 103–125 (1990).
39. Tripathi, A. Total factor productivity growth in Indian agriculture. J. Global
Econ. 6, 286–298 (2010).
40. Gajja, B. L. & Parshad, R. Labour decomposition analysis under different soil
and land irrigability environments in the Kakrapar left bank canal irrigation
project in Gujarat state. Ann. Arid Zone 37, 187–194 (1998).
41. Gallup, J. L. & Sachs, J. D. The economic burden of malaria. Am. J. Trop.
Med. Hyg. 64, 85–96 (2001).
42. Statistical Year Book, India 2016 (Ministry of Statistics and Programme
Implementation, 2016).
43. Cariboni, J., Gatelli, D., Liska, R. & Saltelli, A. The role of sensitivity analysis
in ecological modelling. Ecol. Modell. 203, 167–182 (2007).
44. Saltelli, A., Chan, K. & Scott, E. Sensitivity Analysis (Wiley, 2000).
Acknowledgements
This research was supported by the National Socio-Environmental Synthesis Center
(SESYNC) (grant no. DBI-1052875) through the postdoctoral fellowship programme
to A.B. and the venture working group Land Use & Infectious Diseases jointly with the
National Center for Ecological Synthesis (NCEAS) to A.P.D. We especially thank
M. Bonds, C. Ngonghala, G. De Leo, N. Gottdenker and the rest of the working
group for their insightful comments during our meetings in Annapolis.
Author contributions
A.B., M.S.-V., A.P.D. and M.P. formulated the model. A.B. and M.S.-V. conducted
the numerical and statistical analyses, and all the authors contributed to the final
writing of the manuscript.
Additional information
Supplementary information is available for this paper.
Reprints and permissions information is available at www.nature.com/reprints.
Correspondence and requests for materials should be addressed to A.B.
How to cite this article: Baeza, A., Santos-Vega, M., Dobson, A. P. & Pascual, M.
The rise and fall of malaria under land-use change in frontier regions. Nat. Ecol. Ecol.
1, 0108 (2017).
Competing interests
The authors declare no competing financial interests.