Niche Modeling of Dengue Fever Using Remotely Sensed Environmental Factors and Boosted Regression Trees

Abstract

Dengue fever (DF), a vector-borne flavivirus, is endemic to the tropical countries of the world with nearly 400 million people becoming infected each year and roughly one-third of the world's population living in areas of risk. The main vector for DF is the Aedes aegypti mosquito, which is also the same vector of yellow fever, chikungunya, and Zika viruses. To gain an understanding of the spatial aspects that can affect the epidemiological processes across the disease's geographical range, and the spatial interactions involved, we created and compared Bernoulli and Poisson family Boosted Regression Tree (BRT) models to quantify the overall annual risk of DF incidence by municipality, using the Magdalena River watershed of Colombia as a study site during the time period between 2012 and 2014. A wide range of environmental conditions make this site ideal to develop models that, with minor adjustments, could be applied in many other geographical areas. Our results show that these BRT methods can be successfully used to identify areas at risk and presents great potential for implementation in surveillance programs.

Spatial‐Temporal Assessment of Environmental Factors Related to Dengue Outbreaks in São Paulo, Brazil

“…Since the water vapor could not migrate to the southeastern region, precipitation levels in the peak of the summer were half the average in 2014 to 2015. A suggestion to explain similar inconsistency in the watershed of the Magdalena River in Colombia has been proposed by Eastin et al (2014), Stanforth et al (2016), and Ashby et al (2017) who found dengue incidence to be importantly associated with temperature but not with precipitation in interanual trends. Yet, the normal intra-annual increase in dengue incidence tends to coincide with that of precipitation, which in Sao Paulo occurs during the first semester of the year (Figures 2 and 3).…”

“…However, Stanforth et al (2016) showed that while temperature, elevation, and vegetation cover significantly correlated with dengue outbreaks reported in Magdalena River watershed in Colombia, rainfall did not. However, since the Moderate Resolution Imaging Spectrometer products-usually used for dengue research (Ashby et al, 2017;Buczak et al, 2012;Stanforth et al, 2016)-do not model these factors spatially, they do not model the spatial distribution of dengue cases accurately in the Magdalene River watershed (Stanforth et al, 2016). However, since the Moderate Resolution Imaging Spectrometer products-usually used for dengue research (Ashby et al, 2017;Buczak et al, 2012;Stanforth et al, 2016)-do not model these factors spatially, they do not model the spatial distribution of dengue cases accurately in the Magdalene River watershed (Stanforth et al, 2016).…”

“…One of the satellite-derived environmental variables that dengue outbreaks modeling uses most is rainfall, usually from the Tropical Rainfall Measuring Mission satellite (Ashby et al, 2017;Moreno-Madriñán et al, 2014;Pinto et al, 2011). However, Stanforth et al (2016) showed that while temperature, elevation, and vegetation cover significantly correlated with dengue outbreaks reported in Magdalena River watershed in Colombia, rainfall did not.…”

Performances of machine learning algorithms for mapping fractional cover of an invasive plant species in a dryland ecosystem

“…First, shallow Neural Networks (NN) attracted a lot of attention and were widely applied to many different research problems (Zhou et al, ). In the remote sensing community, the DNN was soon followed by other MLAs: the GBM‐BRT, the SVM, and the RF, which provided better results both in regression and classification (Ashby et al, ; Pal & Mather, ). Our regression analyses in the present study indicated that the RF, the ensemble model, and the GBM outperformed the SVM.…”

“…Although this seems rather low compared to the other datasets, these day-and nighttime temperature datasets have shown to be useful in species distribution modeling and, particularly in Africa where weather stations are scarce. These datasets have shown to be more accurate than other global climate datasets (Ashby, Moreno-Madriñán, Yiannoutsos, & Stanforth, 2017;He et al, 2015). Moreover, we used variables representing topography, infrastructure as well as watercourses as these variables have shown to have an influence on Prosopis distribution (Shiferaw et al, 2019).…”