Received 10 October 2022, accepted 9 November 2022, date of publication 24 November 2022,

date of current version 8 December 2022.

Digital Object Identifier 10.1109/ACCESS.2022.3224472

Model-Based Monitoring of Dengue Spreading

ALEXANDER SCHAUM 1 , (Member, IEEE), ROBERTO BERNAL-JAQUEZ 2,

AND G. SÁNCHEZ-GONZÁLEZ3
1 Automation and Control Group, Kiel University, 24118 Kiel, Germany
2 Department of Applied Mathematics and Systems, Universidad Autónoma Metropolitana–Cuajimalpa, Mexico City 05348, Mexico
3 Centro de Investigación Sobre Enfermedades Infecciosas, Instituto Nacional de Salud Pública, Cuernavaca, Morelos 62100, Mexico

Corresponding author: Alexander Schaum ([email protected])

ABSTRACT The problem of designing a system for monitoring the spread of dengue in a mixed population
of humans and mosquitos is addressed. For this purpose a model–based observer approach is employed
based on a model that describes the actual number of infected humans (with constant population size) and
mosquitos (with time–varying population size) and which has been previously validated with data from
different regions of the state of Morelos in Mexico using the number of new reported cases as available
measurement. This model has time–varying reproduction parameters for the mosquitos and thus enables to
account for different climatic conditions in the course of the year (in particular the rain and dry periods)
that affect the mosquito population size. An analysis of the structural observability, based on the preliminary
assumption of a continuous measurement of the accumulative number of reported cases of human infections
is carried out to motivate a reduced order model that is structurally observable. For this reduced order model
a continuous–discrete extended Kalman Filter is designed as basis for the monitoring and prediction scheme.
The approach is validated using real data from Cuernavaca, Morelos, Mexico, showing a high potential for
future developments towards automated monitoring schemes.

INDEX TERMS Epidemics, vector-borne disease modeling, forecast of dengue cases, population dynamics,
system identification, Kalman filter.

I. INTRODUCTION approaches (see also [5], [6], [7]). An interesting survey of

Dengue fever as an infectious disease caused by the dengue different mathematical models for dengue’s spreading predic-
virus and transmitted to humans by infected mosquitoes of the tion is given in [8]. From a population dynamics perspectives
genus Aedes aegypti that affects millions of people in tropical different approaches have been proposed for the modeling
and subtropical regions. According to estimates by the World going from the causal [9], [10], [11], [12], [13], [14], [15]
Health Organization (WHO) [1] about 390 millions people to the statistical modeling [16], [17].
get infected every year out of which only 96 million receive In several recent studies [18], [19] the problem of dengue
medical treatment. Due to the huge population affected by outbreak estimation and forecast was addressed using some
the virus, the WHO has classified this disease an epidemic. types of the vector-borne disease model SEIR-SEI with com-
For the effected countries, mainly in South Asia and South partments Susceptible-Exposed-Infectious-Recovered (for
America it additionally implies a considerable economical host) and Susceptible- Exposed-Infectious (for vector) in
burden [2], [3]. combination with ensemble Kalman filtering (EnKF) for data
Currently, there are several approaches for predicting assimilation. These studies highlight the great potential of
Dengue spreading in endemic regions of the world. A rather combining compartmental models with Kalman Filters for
complete literature review was performed by Sylvestre achieving high–accuracy now–casts of non–directly mea-
et al. [4], with most of them being based on machine learning sured states and on this basis being able to provide better
forecasts for the upcomming weeks and months.
The associate editor coordinating the review of this manuscript and In particular, understanding of the mechanisms involved in
approving it for publication was Angel F. García-Fernández . the spreading not only in the human population but also in the

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
126892 VOLUME 10, 2022
A. Schaum et al.: Model-Based Monitoring of Dengue Spreading

FIGURE 1. Classical SIR state transition diagram for the human

population with susceptible (S), infectious (I) and recovered (R) states FIGURE 2. Simple model for the propagation of dengue considering the
and associated transition rates τi →j . number of infected mosquitos as unknown and independent of the
number of infected humans.

mosquito population can provide means for better designing

monitoring and forecasting (i.e., early warning) systems for
prevention, containment and possible on–time control of the
spread of the disease (see, e.g., [13], [20]). FIGURE 3. State transition diagram for the mosquito population with
susceptible (U ), incubating (V) and infectious (W) states and associated
The main purpose of the present study is to further ana- transition rates τi →j .
lyze and employ the model proposed in [15] with respect
to its structural observability properties and the design of a
monitoring and prediction scheme based on a nonlinear state
observer. Based on the observability analysis a reduced order
structurally observable model is determined. For this reduced
model a continuous–discrete Extended Kalman Filter [21] is
designed and its performance is established through numeri-
cal simulations with a parameter set that has been validated
in [15] for the city of Cuernavaca, Morelos, Mexico. The
potential use of this state estimation technique for prediction
of the future spreading is shown using actual data from Cuer-
navaca from the year 2012. FIGURE 4. Feedback model for the propagation of dengue considering
the mutual influence of infected mosquitos and infected humans.
The paper is organized as follows. In Section 2 the
monitoring problem is discussed and the mathematical
model presented. The analysis of the structural observ-
important mechanisms in the spreading process consists in
ability is carried out in Section 3. The design of the
providing a model for the spreading within the mosquito
continuous–discrete Extended Kalman Filter is presented in
population. For this purpose consider a time–varying popu-
Section 4. In Section 5 the validation and testing by numerical
lation of mosquitoes which can be in either of the states U
simulations is presented. Final conclusions and the future
(healthy), V (incubating), W (infectious). In contrast to the
outlook are presented in Section 6.
human population, mosquitos do not live sufficiently long
to recover, once they are infected, but can indeed pass the
II. MATHEMATICAL MODELING infection to their offspring. This yields the following state
Consider a human population with N ∈ N inhabitants, each transition diagram shown in Figure 3 for the mosquito pop-
of which can be in either of the states S (susceptible), I ulation with the positive state transition rates τR→S , τU →V ,
(infectious), R (recovered), according to the state transition τV →W . As their are clear interactions between the mosquito
diagram shown in Figure 1. and human populations one obtains a feedback loop as shown
In this state automaton the state transition rates τS →I , in Fig. 4 for the overall spreading process dynamics.
τI →R are both positive. Given that dengue is not transmitted Denoting in the following the number of susceptibles,
between humans and people get infected only due to the bite infectious, and recovered humans by S, I , R, respectively, the
of an infected mosquito and are assumed to recover after a number of susceptible, incubating and infectious mosquitos
certain period (implying a constant population) and become by U , V , W , respectively, and considering mass–action–like
later on again susceptible, given that there is no immunity transition rates it holds that
against the dengue virus. Given that infections are triggered
W I
from outside the human population a possible approach in τS →I = α , τU →V = δ . (1)
the modeling consists in considering the mosquito population Nm N
an exogenous time–varying disturbance input affecting the In the preceding equations Nm denotes the number of
dynamics of the states S, I, R and thus the reports of total mosquitos, α represents the probability that a bite from an
accumulated infection numbers. This approach is schemati- infected mosquito renders a susceptible human infected, and
cally depicted in Fig. 2. δ the probability that when a susceptible mosquito bites an
This approach could be sufficient for the monitoring of the infected human the mosquito gets infected. The terms W /Nm
human population using, e.g., an adaptive observer. and I /N determine the probability that an arbitrarily cho-
An alternative approach, providing more physical insight sen mosquito is infectious or an arbitrarily chosen human
into the underlying dynamics by representing further is infectious, respectively. Having these transition rates the

VOLUME 10, 2022 126893

 A. Schaum et al.: Model-Based Monitoring of Dengue Spreading

total rate of change in S and U due to bites can then be For non–constant reproduction parameters rk , k =
computed by τS →I S and τU →V U , respectively. With the 0, 1, 2, 3 the mosquito population will be time–varying. Note
number of accumulated cases denoted as A, the dynamics that, as the reproduction rates will depend on the climatic
of the spreading process between both populations can be conditions [23] it makes sense to consider time–varying rates.
modeled by the equation set One simple way to take seasonal differences in reproduction
α rates into account, which mostly are present due to rain and
Ṡ = − SW + γ R (2a) dry seasons, the birth and decease rates of the mosquitos are
Nm
α considered piecewise constant according to
İ = SW − ρI (2b)
Nm (
rk1 , t ∈ [t1 , t2 ]
Ṙ = ρI − γ R (2c) rk (t) = , 0 < t1 < t2 < 12 (5)
rk0 , t ∈
/ [t1 , t2 ]
Ȧ = I (2d)
for k = 0, 1, 2, 3. The time varying rates and associated
δ
 
Nm
U̇ = − IU + (r1 U + r2 V + r3 W ) 1 − (2e) non–constant mosquito population size with Verhulst–like
N K dynamics represent important differences in comparison to
δ
V̇ = IU − κV (2f) other models that have been reported in the literature (see,
N   e.g., [12] and references therein).
Nm In compact form the model (2) is written as
Ẇ = κV + r0 W 1 − (2g)
K
Nm = U + V + W (2h) ẋ(t) = f (t, x(t)), t > 0, x(0) = x0 (6a)

for t > 0 with initial values with the state vector x(t) ∈ R7 , where
T
S(0) = S0 , I (0) = I0 , R(0) = R0 x= S I RAU V W .

(6b)
U (0) = U0 , V (0) = V0 , W (0) = W0 .
III. STRUCTURAL OBSERVABILITY ANALYSIS
In the preceding equation set the parameter γ > 0 represents In this section the structural observability [24], [25], [26]
the rate with which recovered humans loose their immunity, of the spreading model (2) is analyzed in the understanding
ρ > 0 the recovery rate, κ > 0 the rate with which incubat- that the structural observability is a necessary condition for
ing mosquitos become infectious, r0 , r1 , r2 , r3 ∈ R are the the (local) complete observability and provides an important
net birth and decease rates. The parameter K represents the basis for design of an observer–based monitoring scheme.
(constant) maximum mosquito population size. Definition 1: The system is structurally observable if it is
It is assumed that S, I , R, A, U , V , W : [0, ∞) → R+ are completely observable for at least one state–parameter pair.
continuously differentiable functions with R+ denoting the For the analysis of the structural observability first recall
positive real numbers. Given that the right–hand side of the the definition of the structure graph 0(x) = (V , E(x)) of
differential equation system (2) is locally Lipschitz continu- the system at a point x ∈ X , consisting of the vertex set
ous, a unique solution exists that is continuously depending V = {v1 , . . . , v7 } where vk is associated to the state xk ,
on the initial conditions. k = 1, . . . , 7, and the (directed) edge set E(x) =
Assuming a constant population of humans the total popu- {(vi , vj )| ∂f∂x
i (x)
j
6= 0}. Note that the edge set in principle
lation numbers are given by
depends on the value of x, given that ∂f∂x
i (x)
j
can be zero at some
N = S + I + R = const. > 0, (3a) particular states. Structural observability is a generic property
Nm = U + V + W ≥ 0. (3b) for a system that depends only on the intrinsic interconnection
structure and not on the particular values of ∂f∂x i (x)
j
∈ R,
Note that the first equation corresponds to an algebraic con- i.e., it is actually independent of the particular parameter set
straint that is naturally satisfied as long as S0 + I0 + R0 = N . and the particular value x where the partial derivatives are
This is summarized in the following lemma. evaluated. It only depends on whether the influence of xj on
Lemma 1: [15] The total human population size N = S + xi is not always zero. Indeed, if it is not zero for at least one
I + R remains constant over time. parameter vector and one state value x ∈ X , due to continuity
The next lemma shows an interesting property of the model reasons it will be nonzero for almost all values of x and
(2) putting it in direct relation to classical population balances parameters, up to those lying on a specific hyperplane of the
as, e.g., the Verhulst model [22]. state–parameter space which has measure zero (see, e.g., the
Lemma 2: [15] For constant reproduction parameters discussion in [25]).
r1 = r2 = r and r3 = r − r0 the total mosquito population The characterization of possible graphs implying struc-
size Nm = U + V + W is determined by the logistic growth tural controllability and observability can be reviewed, e.g.,
 
Ṅm = rNm 1 − NKm , t > 0, Nm (0) = Nm,0 (4) in [24], [26] and typically implies the extension of the state
graph discussed above by input and output vertexes. As in
and for r, r3 > 0 it holds that Nm → K . the present study no inputs are considered and the measured

126894 VOLUME 10, 2022

A. Schaum et al.: Model-Based Monitoring of Dengue Spreading

FIGURE 6. Structural connectivity graph for the 6–state reduced model

with measurement of the state A.

This model is identical to (2) with R in the first equation

FIGURE 5. Structual connectivity graph for the 7–state bi-species model substituted by the relation (7). The structure graph of this
with measurement of the state A. The red dotted connection from S to
R shows the lack of structural observability, given that from R one can
model is shown in Fig. 6 and the following result is an
not reach any other node without crossing I again, implying a loop. immediate consequence of the above considerations.
Lemma 4: The reduced order model (8) is structurally
observable.
output corresponds with the state A the essential information In compact form the reduced model (8) is written as
is already contained in the state graph. The condition for
structural observability then basically requires that a unique ẋr (t) = f r (t, xr (t)), t > 0, xr (0) = xr,0 (9a)
information path exists from the measured nodes toward all with the state vector xr (t) ∈ R6 , where
other nodes without loops and intersections. For the case at T
xr = S I A U V W .

hand this can be readily applied to the structure graph shown (9b)
in Fig. 5. Based on this graphical representation the following
result is obtained. IV. CONTINUOUS-DISCRETE OBSERVER DESIGN
Lemma 3: The spreading model (2) is not structurally Having the continuous–time model (8) (or equivalently (9))
observable. that is structurally observable it would be straight for-
Proof: As can be seen in Fig. 5 the signal flow cannot ward to implement a continuous–time observation scheme.
pass through the complete graph toward the measured node Anyway, as the reporting of new cases is carried out in
A without passing twice through the node S, implying that discrete, periodic1 time steps a continuous–discrete observa-
no information path without loops exists.  tion scheme is considered in the following. In this scheme
As a particular outcome of this analysis one can see that it the continuous–time model is used for predicting the state
is possible to obtain a straight information path without loops in between sampling times and the correction only takes
from the output node A to all other nodes except R, as high- place at the sampling moments. A well–established approach
lighted by different colors (and shapes) of the connections in to address this kind of problem is the continuous–discrete
Fig. 5. Recall from Lemma 1 that the value R can be obtained Extended Kalman Filter (cd-ekf) [21].
at any moment from the algebraic relation Given that different stochastic sources are influencing the
spreading process which are not modeled in (8) additional
R(t) = N − S(t) − I (t) (7) stochastic perturbation terms are included (so–called pro-
and one can reduce the model by this state without loosing cess noise) and measurement uncertainties are summarized
any information. This leads to the reduced order model in a measurement noise component. Assuming the process
and measurement noise variable w and v, respectively, to be
α
Ṡ = − SW + γ (N − S − I ) (8a) normal distributed with zero mean and covariances Q, R,
Nm respectively, this yields the stochastic model
α
İ = SW − ρI (8b)
Nm ẋr = f r (t, xr ) + w, w ∼ N (0, Q) (10)
Ȧ = I (8c) y = cT xr + v, v ∼ N (0, R). (11)
δ
 
Nm
U̇ = − IU + (r1 U + r2 V + r3 W ) 1 − (8d) The basic idea consists in using (i) the deterministic nonlinear
N K
δ reduced model (8) for predicting the state between measure-
V̇ = IU − κV (8e) ments, (ii) a linearization around the actual estimated state for
N 
Nm
 the prediction of the associated estimation error covariance,
Ẇ = κV + r0 W 1 − (8f)
K 1 The following considerations can easily be extended to irregular (i.e.,
Nm = U + V + W . (8g) aperiodic) sampling times.

VOLUME 10, 2022 126895

 A. Schaum et al.: Model-Based Monitoring of Dengue Spreading

and (iii) a correction at the discrete measurement instances

tk , k ∈ N of the state and covariance estimates. This scheme
is summarized in the following algorithm.
a) Initialization: Set k = 0, x̂r (0), P(0).
b) Prediction: Set the initial conditions x̂r,p (tk ) = x̂r (tk )
with the estimate x̂r (tk ) of the reduced state xr (tk ) in (9),
Pp (tk ) = P(tk ) with the estimate of the covariance matrix,
and solve the reduced differential equation system
x̂˙ r,p = f r (t, x̂r,p ) (12a)
∂f r
Ṗp = FPp + Pp F T + Q, F= (x̂r,p ) (12b)
∂xr
over the time horizon t ∈ (tk , tk+1 ] to obtain the predic-
tions x̂r,p (t) and Pp (t). FIGURE 7. Simulation results for the estimates for the accumulated
c) Correction: Update the predicted value with the actual number of infected A obtained with the continuous–discrete Extended
Kalman Filter (cd-ekf, red continuous lines) and the one predicted by the
measurement y(tk+1 ) = A(tk+1 ) + v in dependency of the model (blue dotted lines).
predicted estimation error covariance Pp (tk+1 ) as
Lk+1 = Pp (tk+1 )c(cT Pp (tk+1 )c + R)−1 (12c) (
1.71e − 3, t ∈ [t1 , t2 ]
x̂r (tk+1 ) = x̂r,p (t) − Lk+1 (Âp (t) − y(t)) (12d) r2 (t) = ,
0.53, t∈/ [t1 , t2 ]
P(tk+1 ) = (I − Lk+1 cT )Pp (tk+1 ). (12e) (
5.96e − 3, t ∈ [t1 , t2 ]
with the Kalman correction gain Lk+1 . r3 (t) =
−0.12, t∈/ [t1 , t2 ].
Repeat from b) with k = k + 1.

Generally speaking, the aim of the Kalman Filter is to min- For the simulation the following initial condition for the
imize the estimation error covariance P(t) by appropriately system (x0 ), Kalman Filter (x̂0 ), estimation error covariance
choosing the correction gain Lk (see, e.g., [21]). (P0 ), variance of measurement noise (R), and covariance
At any time instant t ≥ 0 the estimate of the number of matrix for the process noise (Q) have been used:
recovered humans is given by
x(0) = 369996 4 0 4 104 6.19 109
 
 
R̂(t) = N − Ŝ(t) − Î (t). (12f) x̂r (0) = 369930 54 14 20 1500 9.285
P0 = I , R = 1
It should be mentioned that besides the cd-ekf proposed
Q = diag 103 102 102 103 103 103 ,
 

here, e.g., unscented [27] or ensemble–based [28] Kalman
Filters, particle filters [29], or nonlinear observers could be
The associated time evolution of A and  are shown in Fig. 7,
employed for the task of providing an estimate x̂r .
and the resulting state evolution for S, I , R, U , V , W and their
estimates Ŝ, Î , R̂, Û , V̂ , Ŵ are shown in Fig. 8.
V. VALIDATION
It can be seen in Fig. 7 that the value of A is quickly recov-
In this section the performance of the proposed approach is ered at the measurement time instances tk , k = 2, . . . , 12,
tested for the problem of monitoring the spread of dengue in with an increasing deviation between measurement time
the state of Morelos, Mexico within the city of Cuernavaca instances due to wrong estimates in the remaining states, but
for which monthly reported new cases have been used to that after about 4 months (i.e., 3 measurements) the estimated
determine the increasing number of accumulated cases A. value  almost exactly follows the simulated value A. For
In [15] a parameter set for the spreading during the year the states, one can see in Fig. 8 that the states associated to
2012 has been presented and is given as follows: the human population (i.e., S, I ) are rather well recovered,
while the ones for the mosquitos are only qualitatively recon-
α = 3.36e − 3, γ = 0.9543, ρ = 4.85, κ = 0.13,
structed. The reason for this can be manifold and will need
δ
= 0.19, t1 = 3.6, t2 = 7.8, N = 37 · 104 , to be investigated with more detail in future studies. Given
N that the aim of the proposed monitoring scheme is to recover
U0 = 104 , V0 = 6.19, W0 = 109, K = 106 the information about the human infection process and does
(
0.25, t ∈ [t1 , t2 ] not focus on the estimation of the mosquito population, with
r0 (t) =
−1.09, t ∈ / [t1 , t2 ] the latter only being included to cover the main mechanisms
( involved to retain the underlying feedback structure, the result
0.028, t ∈ [t1 , t2 ] seems reasonably well working to be implemented as a mon-
r1 (t) = ,
−0.39, t ∈ / [t1 , t2 ] itoring mechanism.

126896 VOLUME 10, 2022

A. Schaum et al.: Model-Based Monitoring of Dengue Spreading

FIGURE 8. Simulation results for the state estimates obtained with the continuous–discrete Extended Kalman Filter (cd-ekf, red
continuous lines) and the ones predicted by the model (blue dotted lines).

t = 5). This highlights the potential of using this approach

including automated correction with new reported cases for
the monitoring and prediction of the spreading of dengue.

VI. CONCLUSION
The problem of monitoring the spread of dengue in a mixed
human–mosquito population has been discussed. A monitor-
ing scheme has been designed that is based on a recently
proposed validated mathematical model. Motivated by a
structural observability analysis a reduced order model has
been used for the design of a continuous–discrete Extended
Kalman Filter. The proposed estimation scheme has been
tested with the validated model for a case study for actual
reported cases in the city of Cuernavaca, Morelos, Mexico.
The potential use of the proposed scheme for monitoring and
prediction based of improved initial estimates provided by the
FIGURE 9. Example of prediction from the actual estimate at time
t = 5 with data (stars), uncorrected prediction starting from t = 4 (red
Kalman Filter has been discussed and illustrated.
dotted line), corrected prediction starting from t = 5 (yellow dash–dotted Future work will address further validation studies based
line) and the associated ±3σA variance band (shaded region starting from
t = 5). Time instances t < 5 belong to the past and time instances
on different data sets, as well as further analysis of the under-
t > 5 to the future. lying model in terms of detectability, non-local observability
and the design of robust observers. Given the intrinsic depen-
dencies of the reproduction of mosquitos on climatic condi-
A particular advantage of the corrections introduced at tions it is planned to combine these approaches with suitable
measurement time instances is that they improve the predic- machine–learning techniques for automated adaptation of the
tion capabilities of the model. To highlight this in Fig. 9 model parameters.
a comparison is provided between predictions based on the
initial value at time t = 5 for the interval [5, 12] without REFERENCES
correction (red dotted line) and with the correction using [1] Electronic References, World Health Organization, Geneva, Switzerland,
the measurement at t = 5 (yellow dash–dotted line). The 2021.
stars indicate the actual reported cases numbers for Cuer- [2] S. Bhatt, P. W. Gething, O. J. Brady, J. P. Messina, A. W. Farlow,
C. L. Moyes, J. M. Drake, J. S. Brownstein, A. G. Hoen, O. Sankoh, and
navaca during 2012 (starting from February). It can be clearly M. F. Myers, ‘‘The global distribution and burden of dengue,’’ Nature,
seen that an advantage is obtained by using the automated vol. 496, no. 7446, pp. 504–507, 2013.
correction for improving the estimate at t = 5 and using [3] K. Mulligan, J. Dixon, C.-L. J. Sinn, and S. J. Elliott, ‘‘Is dengue a disease
of poverty? A systematic review,’’ Pathogens Global Health, vol. 109,
the corrected estimate for future predictions. In the figure no. 1, pp. 10–18, Feb. 2015.
there is also the ±3σA interval shown around the prediction [4] E. Sylvestre, C. Joachim, E. Cécilia-Joseph, G. Bouzillé,
value, with σA being the predicted estimation error variance B. Campillo-Gimenez, M. Cuggia, and A. Cabié, ‘‘Data-driven methods
for dengue prediction and surveillance using real-world and big data: A
for the state A, comming from the approximated covariance systematic review,’’ PLOS Neglected Tropical Diseases, vol. 16, no. 1,
evolution dynamics (12b) (without further corrections after pp. 1–22, 2022.

VOLUME 10, 2022 126897

 A. Schaum et al.: Model-Based Monitoring of Dengue Spreading

[5] P. Guo, T. Liu, Q. Zhang, L. Wang, and J. Xiao, ‘‘Developing a dengue [27] S. J. Julier and J. K. Uhlmann, ‘‘Unscented filtering and nonlinear estima-
forecast model using machine learning: A case study in China,’’ PLoS Negl. tion,’’ Proc. IEEE, vol. 92, no. 3, pp. 401–422, Mar. 2004.
Trop. Dis., vol. 11, no. 10, pp. 1–22, 2017. [28] G. Evensen, Data Assimilation: The Ensemble Kalman Filter, 2nd ed.
[6] N. A. M. Salim, Y. B. Wah, C. Reeves, M. Smith, W. F. W. Yaacob, Berlin, Germany: Springer-Verlag, 2009.
R. N. Mudin, R. Dapari, N. N. F. F. Sapri, and U. Haque, ‘‘Prediction [29] P. D. Moral, ‘‘Nonlinear filtering: Interacting particle resolution,’’ Comp.
of dengue outbreak in Selangor Malaysia using machine learning tech- Rendus de l’Académie des Sci. Ser. I Math., vol. 325, no. 6, pp. 653–658,
niques,’’ Sci. Rep., vol. 11, no. 1, p. 939, Jan. 2021. Sep. 1997.
[7] C. Torres, S. Barguil, M. Melgarejo, and A. Olarte, ‘‘Fuzzy model identifi-
cation of dengue epidemic in Colombia based on multiresolution analysis,’’
Artif. Intell. Med., vol. 60, no. 1, pp. 41–51, Jan. 2014. ALEXANDER SCHAUM (Member, IEEE)
[8] I. B. J. Braselton, ‘‘A survey of mathematical models of dengue fever,’’
received the Diploma degree in technical cyber-
J. Comput. Sci. Syst. Biol., vol. 8, no. 5, p. 255, 2015.
[9] M. Andrau, N. Hens, C. Marais, and P. Beutels, ‘‘Dynamic epidemiolog-
netics from the University of Stuttgart, Germany,
ical models for dengue transmission: A systematic review of structural in 2006, and the Ph.D. degree in electrical
approaches,’’ PLoS ONE, vol. 7, no. 11, pp. 1–14, Nov. 2012. engineering-control from the National Univer-
[10] T. Sardar, S. Rana, and J. Chattopadhyay, ‘‘A mathematical model of sity of Mexico (Universidad Nacional Autónoma
dengue transmission with memory,’’ Commun. Nonlinear Sci. Numer. de México, UNAM), Mexico City, in 2009.
Simul., vol. 22, nos. 1–3, pp. 511–525, May 2015. From 2009 to 2010, he was a Postdoc at the Insti-
[11] I. Bakach, ‘‘A survey of mathematical models of Dengue fever,’’ M.S. the- tute of Engineering, UNAM, and the Chemical
sis, Dept. Math. Sci., Georgia Southern Univ., Statesboro, GA, USA, 2015. Engineering Department, Metropolitan University
[Online]. Available: https://digitalcommons.georgiasouthern.edu/etd/1236 (Universidad Autónoma Metropolitana, UAM), Iztapalapa, Mexico City,
[12] L. Esteva and C. Vargas, ‘‘Analysis of a dengue disease transmission from 2010 to 2011. From 2011 to 2014, he was a Guest Professor at the
model,’’ Math. Biosci., vol. 150, no. 2, pp. 131–151, Jun. 1998. Applied Mathematics and Systems Department, UAM Cuajimalpa, Mexico
[13] A. Fischer, K. Chudej, and H. J. Pesch, ‘‘Optimal vaccination and control City. Since 2014, he has been a Senior Researcher with the Chair of
strategies against dengue,’’ Math. Methods Appl. Sci., vol. 42, no. 10, Automatic Control, Kiel University, Germany. He is a member of the IEEE
pp. 3496–3507, Jul. 2019. CSS, IEEE CSS CEB, IEEE TCEB, IFAC, and IEEE TCs on Distributed
[14] P. Sauerteig, K. Worthmann, and K. Chudej, ‘‘Towards model predic-
Parameter Systems. His research interests include observer and control
tive control for maintaining a hard infection cap during an outbreak of
design for nonlinear finite- and infinite-dimensional systems and networked
dengue fever,’’ in Proc. MATHMOD Discuss. Contribution Volume, 2022,
pp. 672–673. and stochastic systems with application areas in process control, neuromor-
[15] A. Schaum, R. B. Jaquez, C. Torres–Sosa, and G. Sánchez-González, phic engineering, energy systems, and spreading phenomena in networks.
‘‘Modeling the spreading of dengue using a mixed population model,’’
IFAC-PapersOnLine, vol. 55, no. 20, pp. 582–587, 2022.
[16] B. L. Carrington, M. V. Armijos, L. Lambrechts, and W. T. Scott, ‘‘Fluctu- ROBERTO BERNAL-JAQUEZ received the B.S.
ations at a low mean temperature accelerate dengue virus transmission by
degree in physics from the Universidad Naciona
Aedes aegypti,’’ PLOS Neglected Tropical Diseases, vol. 7, no. 4, pp. 1–8,
Apr. 2013.
Autonoma de Mexico (UNAM) and the Ph.D.
[17] H. E. Link, ‘‘Statistical models of dengue fever,’’ in Data Mining (Com- degree in physics from the Universidad Autonoma
munications in Computer and Information Science), vol. 996, R. Islam, Ed. del Estado de Morelos-ICN (UNAM) Mexico,
Cham, Switzerland: Springer, 2019, pp. 35–51. in 2006.
[18] C. Yi, L. W. Cohnstaedt, and C. M. Scoglio, ‘‘SEIR-SEI-EnKF: A new He was a Postdoctoral Reasearcher at the Depar-
model for estimating and forecasting dengue outbreak dynamics,’’ IEEE tamento de Fisica, UAEM, from 2006 to 2007,
Access, vol. 9, pp. 156758–156767, 2021. and the Theoretical Chemistry Group, University
[19] Y. Chen, T. Liu, X. Yu, Q. Zeng, Z. Cai, H. Wu, Q. Zhang, J. Xiao, W. Ma, of Turin, Italy, from 2007 to 2009. He is an
S. Pei, and P. Guo, ‘‘An ensemble forecast system for tracking dynamics Full Professor with the Universida Autonoma Metropolitana Cuajimalpa,
of dengue outbreaks and its validation in China,’’ PLOS Comput. Biol., Mexico. His research interests include machine learning in chemical-physics,
vol. 18, no. 6, Jun. 2022, Art. no. e1010218. mathematical-physics, nonlinear dynamics, and complex networks.
[20] S. A. Carvalho, S. O. da Silva, and I. D. C. Charret, ‘‘Mathematical mod-
eling of dengue epidemic: Control methods and vaccination strategies,’’
Theory Biosciences, vol. 138, no. 2, pp. 223–239, Nov. 2019.
[21] A. Gelb, Applied Optimal Estimation. Cambridge, MA, USA: MIT Press, G. SÁNCHEZ-GONZÁLEZ received the B.S.
1978. degree in physics from the Universidad de
[22] N. Bacaër, A Short History of Mathematical Population Dynamics. Lon- Guadalajara, Jalisco, Mexico, and the Ph.D. degree
don, U.K.: Springer Verlag, 2011. in physics from the Universidad Autonoma del
[23] G. Sanchez-Gonzalez, R. Conde, R. N. Moreno, and P. C. L. Vazquez,
Estado de Morelos.
‘‘Prediction of dengue outbreaks in Mexico based on entomological,
meteorological and demographic data,’’ PLoS ONE, vol. 13, no. 8, 2018,
He is a Researcher with the Centro de Investi-
Art. no. e0196047. gación Sobre Enfermedades Infecciosas, Instituto
[24] C.-T. Lin, ‘‘Structural controllability,’’ IEEE Trans. Autom. Control, Nacional de Salud Pública, Cuernavaca, Morelos,
vol. AC-19, no. 3, pp. 201–208, Jun. 1974. Mexico, where he works in vector-borne dis-
[25] J.-M. Dion, C. Commault, and J. van der Woude, ‘‘Generic properties and eases of medical and agricultural importance.
control of linear structured systems: A survey,’’ Automatica, vol. 39, no. 7, His current research interests include disease vector biting midges and
pp. 1125–1144, Jul. 2003. mosquitoes and insect population management methods to reduce pathogen
[26] Y. Y. Liu, J. J. Slotine, and A. L. Barabási, ‘‘Observability of complex transmission.
systems,’’ Proc. Nat. Acad. Sci. USA, vol. 110, no. 7, pp. 2460–2465, 2013.

126898 VOLUME 10, 2022