Fig 1.
General timeline and methodology of this proposal.
Fig 2.
Locations included in the project by country.
Fig 3.
Urban and periurban areas identified.
Fig 4.
System Architecture proposed.
Fig 5.
Data collection flowchart: The grey boxes represent the central team tasks, while the white ones indicate the tasks of the partner country teams.
Table 1.
Number of variables to be extracted per country according to the CFEs approach.
Fig 6.
Spatial distribution of health variables and CFEs.
Variables in order of appearance: A) Prevalence of obesity in children under 5 years (Health variable), B) Percentage of houses whose income is insufficient for cover health needs (Political dimension variable), B) Percentage of houses with water pipe (Individual dimension variable), and B) Maximum temperature of the warmest month (Environmental dimension variable).
Fig 7.
Supporting evidence regarding modelling: Feature importance of the antecedent model for Argentinean localities.
Variables in orden of appearence: Percentage of people enrolled in the Potenciar Trabajo plan, quality of life index, and nighttime light radiance in the peri-urban area, mean precipitations, urban area, periurban NDBI, percentage of houses with at least one celphone, percentage of beneficiaries of Tarjeta Alimentar plan, periurban perimeter, percentage of the urban area occupied by informal settlements, infant mortality rate, nitrogen dioxide concentration, particular material concentration (PM10), fractal dimension, texture filter (mean contrast), urban concentration, texture filter (standar deviation contrast), percentage of people with social security coverage, population density (aged 4 to 5 years old).
Fig 8.
Spatial distribution of the 3 most important variables in the antecedent model.
Variables in orden of appearence: A) Percentage of people enrolled in the Potenciar Trabajo plan (political dimension), B) quality of life index (political dimension), and C) and nighttime light radiance in the peri-urban area (environmental dimension).
Fig 9.
Spatial characterization and patterns detected in the prevalence of original and simulated LBW.
In which: A) Original LBW prevalence, B) and C) Hotspot detection in LBW original and simulated prevalences respectively, and D) hotspot comparison between original and simulated LBW.
Fig 10.
Cover of the LLM agent trained.