Scientia et Technica Año XXVIII, Vol. 30, No. 03, julio - septiembre de 2025. Universidad Tecnológica de Pereira. ISSN 0122-1701 y ISSN-e: 2344-7214
151
A proposal for a fuzzy climate classification index
for Colombia
Una propuesta de un índice difuso de clasificación climática para Colombia
J. G. Popayán Hernández ; O. E. Bru-Cordero
DOI: https://doi.org/10.22517/23447214.25894
Scientific and technological research paper
Abstract— This study developed a fuzzy logic-based climate
classification index for Colombia, integrating hydroclimatic, air
quality, and topographic variables through a three-phase
methodology. In Phase 1 (2010-2022), multisource data acquisition
processed precipitation (600-8000 mm/year), temperature (14-
32°C), humidity (28-95%), PM₂.₅ (6-35 µg/m³), and NO₂ (10-60
ppb) using IDEAM's DHIME portal and NASA Giovanni
products, with quality-controlled interpolation. Phase 2
implemented a Mamdani-type fuzzy inference system in FisPro,
creating 127 "If-Then" rules through nonlinear correlation
analysis (Spearman >0.65) and expert knowledge, using MIN-
MAX operators and adaptive weights (0.3 rural/0.5 urban
pollution coefficients). Phase 3 geospatial implementation
achieved 92.3% cross-validation accuracy (MAE=1.2,
RMSE=1.8), generating vulnerability maps (0-10 scale) through
QGIS processing. Results revealed extreme climate variability:
precipitation gradients (600 mm/year in Riohacha to 8000 mm in
Quibdó), urban heat islands (Neiva 30°C vs. Bogotá 16°C), and
pollution hotspots (Barranquilla 30 µg/m³ PM₂.₅ vs. Leticia 6
µg/m³). The fuzzy index outperformed traditional methods
(Köppen, Thornthwaite) by capturing nonlinear interactions,
showing 15% agricultural yield reductions in high-NO₂ zones and
identifying vulnerability thresholds for coffee rust outbreaks
(>80% humidity) and urban heat stress (85% RH = 41°C felt
temperature). The model's adaptive structure effectively
addressed Colombia's climatic heterogeneity while overcoming
rigid classification limitations, providing a robust tool for climate
risk assessment under anthropogenic change scenarios, though
future work should incorporate higher-resolution pollution data
to reduce the 15% uncertainty in industrial zones.
Index Terms— Climate classification; Fuzzy logic;
Hydroclimatic variables; Vulnerability index.
Resumen— En este estudio se desarrolló un índice de clasificación
climática basado en lógica difusa para Colombia, integrando
variables hidroclimáticas, de calidad del aire y topográficas
mediante una metodología de tres fases. En la Fase 1 (2010-2022),
se adquirieron y procesaron datos multifuente de precipitación
(600-8000 mm/año), temperatura (14-32°C), humedad (28-95%),
PM₂.₅ (6-35 µg/m³) y NO₂ (10-60 ppb) utilizando el portal DHIME
This manuscript was submitted on July 21, 2025. Accepted on September
08, 2025. And published on September 29, 2025. Juan Guillermo Popayán-
Hernández, Universidad Nacional de Colombia, Sede de La Paz La Paz
(código postal 202017), Cesar, Colombia. km 9 Vía Valledupar - La Paz
(Cesar, Colombia). E-mail: jgpopayanh@unal.edu.co;.
Osnamir Elías Bru-Cordero, Universidad Nacional de Colombia, Sede de La
Paz La Paz (código postal 202017), Cesar, Colombia. km 9 Vía Valledupar
- La Paz (Cesar, Colombia). Correo electrónico: oebruc@unal.edu.co;
del IDEAM y productos NASA Giovanni, con interpolación
controlada por calidad. La Fase 2 implementó un sistema de
inferencia difusa tipo Mamdani en FisPro, creando 127 reglas "Si-
Entonces" mediante análisis de correlación no lineal (Spearman
>0.65) y conocimiento experto, utilizando operadores MIN-MAX
y ponderaciones adaptativas (coeficientes de 0.3 para zonas
rurales y 0.5 urbanas). La Fase 3 de implementación geoespacial
alcanzó un 92.3% de precisión en validación cruzada (MAE=1.2,
RMSE=1.8), generando mapas de vulnerabilidad (escala 0-10)
mediante procesamiento en QGIS. Los resultados revelaron
extrema variabilidad climática: gradientes de precipitación (600
mm/año en Riohacha hasta 8000 mm en Quibdó), islas de calor
urbanas (Neiva 30°C vs. Bogotá 16°C) y focos de contaminación
(Barranquilla 30 µg/m³ de PM₂.₅ vs. Leticia 6 µg/m³). El índice
difuso superó métodos tradicionales (Köppen, Thornthwaite) al
capturar interacciones no lineales, mostrando reducciones del
15% en rendimientos agrícolas en zonas con alto NO₂ e
identificando umbrales de vulnerabilidad para brotes de roya en
café (>80% humedad) y estrés térmico urbano (85% HR = 41°C
de sensación térmica). La estructura adaptativa del modelo
abordó efectivamente la heterogeneidad climática colombiana
superando limitaciones de clasificaciones rígidas, proporcionando
una herramienta robusta para evaluación de riesgos climáticos
bajo escenarios de cambio antropogénico, aunque futuros trabajos
deberían incorporar datos de contaminación de mayor resolución
para reducir el 15% de incertidumbre en zonas industriales.
Palabras claves— Clasificación climática; Índice de
vulnerabilidad; Lógica difusa; Variables hidroclimáticas.
I. INTRODUCTION
LIMATE, from a technical perspective, is the statistical
pattern of atmospheric conditions—such as temperature,
precipitation, humidity, wind, and pressure—in a region
over an extended period, typically 30 years, according to the
definition of the World Meteorological Organization (WMO).
The Intergovernmental Panel on Climate Change (IPCC)
emphasizes that climate results from complex interactions
among components of the Earth's system (atmosphere,
hydrosphere, cryosphere, lithosphere, and biosphere),
regulated by radiative forcings, such as greenhouse gases
(GHGs), whose concentration has increased by 47% in CO₂-
equivalent terms since 1750, reaching 504 ppm in 2023. Data
from the IPCC AR6 indicate global warming of 1.1°C above
pre-industrial levels (1850– 1900), with a rate of 0.2°C per
decade, attributed 95% to anthropogenic activities [1].
C
152
Scientia et Technica Año XXVIII, Vol. 30, No. 03, julio septiembre de 2025. Universidad Tecnológica de Pereira
Historically, the concept of climate evolved from Aristotle’s
empirical observations in Meteorologica (4th century BC) to
scientific systematization in the 19th century, with Alexander
von Humboldt, who introduced isotherms, and Köppen, who
developed climate classification based on thermopluviometric
data [2]. In the 20th century, climatology consolidated as a
quantitative science with numerical models, such as those of
the IPCC, which project a temperature increase of 1.5 to 4.4°C
by 2100 under SSP scenarios, highlighting the urgency of
mitigation [3]. These advances reflect the transition from a
descriptive to an analytical approach, integrating climatic
teleconnections (ENSO, NAO) and systemic feedbacks, such
as albedo or the carbon cycle.
On the other hand, climate classification indices were
changes in these classifications, anticipating, for example, a
15% expansion of Aw climate by 2050 due to global warming
[9].
From this conceptual and methodological evolution, it is
possible to compare the main climate classification indices used
in Colombia, detailing their fundamental variables and how
each interprets the country’s complex atmospheric realities
[10]. Below, Table I summarizes the most representative
systems and their distinctive criteria:
TABLE I
MAIN CLIMATIC INDICES USED IN LATIN AMERICA AND
COLOMBIA
developed from the need to systematize interactions among key
meteorological variables—temperature, precipitation,
evapotranspiration, and, in some cases, solar radiation—to
define reproducible spatial and temporal patterns [4]. The
Köppen-Geiger system (1900–1936), the most widely used,
classifies climates based on monthly and annual thresholds of
temperature and precipitation (e.g., Af for tropical humid zones
with precipitation ≥60 mm every month and Tmean ≥18°C),
while Thornthwaite (1948) incorporated water balance through
Climate
Index
Köppen-
Geiger
Thornthwait
e
Meteorological
Variables Used
Monthly/annual
mean temperature,
monthly/annual
precipitation
Temperature,
precipitation,
potential
evapotranspiration
(PET)
Interpretation
Classifies climates into groups (A:
tropical, B: arid, C: temperate, etc.)
based on thermopluviometric
thresholds. E.g.: Af = tropical
humid (no dry season).
Defines climate types based on
water balance (humidity index).
E.g.: arid (PET > precipitation).
potential evapotranspiration (PET), differentiating arid regions
(humidity index <0) from humid ones [5]. Later, Holdridge
Holdridge Biotemperature,
precipitation,
evapotranspiration
Relates climate to plant life zones
using altitudinal tiers and thermal
gradients.
(1967) introduced the concept of biotemperature and altitudinal
tiers, relevant for mountainous regions like the Andes [6]. In
Colombia, these systems are applied considering the marked
orographic variability: IDEAM uses Köppen to identify that
83% of the territory is tropical (Af in the Amazon and Pacific,
Aw in the Orinoquía), with altitudinal modifications (Cfb in
Bogotá, 2600 m.a.s.l., Tmean 14°C). Additionally, indices such
as Martonne’s aridity index are used to assess drought in La
Guajira (index <1), or ENSO to predict pluviometric anomalies,
given that phenomena like El Niño reduce rainfall by up to 40%
in the Caribbean region [7]. Historically, the earliest
classifications date back to Aristotle (4th century BC), but it
was Alexander von Humboldt (1800) who established
correlations
between
altitude
and
vegetation,
laying
the
Aridity
Index
(Martonne)
ENSO (El
Niño-
Southern
Oscillation)
SPI
(Standardize
d
Precipitation
Index)
Vegetation
Index
(NDVI)
Annual
precipitation,
annual mean
temperature
Sea Surface
Temperature
(SST) anomalies,
atmospheric
pressure (SOI)
Accumulated
precipitation at
different time
scales
Satellite
reflectance data
(spectral bands)
Measures drought: <5 = desert, 5-10
= semi-arid, 10-20 = sub-humid,
>20 = humid.
Classifies phases (El Niño, La Niña,
Neutral) that alter global
precipitation and temperature
patterns.
Evaluates droughts (negative
values) or water excess (positive) in
specific periods (e.g.: SPI-6 =
agricultural drought).
Shows vegetation health and water
stress (low values = drought or
degradation).
foundations for thermal tiers. In the 20th century, the European
school (Köppen, Troll) and the North American school
(Thornthwaite) dominated theoretical climatology, while in
Latin America, local adaptations like Papadakis’ (1960)
incorporated agroclimatic data. In the region, institutions such
as Mexico’s Servicio Meteorológico Nacional (SMN, available
at https://www.gob.mx/smn) or Brazil’s Instituto Nacional de
Meteorologia (INMET, available at
https://portal.inmet.gov.br/) have adjusted these systems to
mesoclimatic scales, while in Colombia, the Instituto de
Hidrología, Meteorología y Estudios Ambientales (IDEAM,
available at https://www.ideam.gov.co/) integrates satellite
reanalysis like the Climate Hazards Group InfraRed
Precipitation with Stations (CHIRPS) to refine zonation in areas
of high topographic complexity, such as the Coffee Region,
where precipitation varies between 2000–4000 mm/year within
less than 50 km [8]. Currently, CMIP6 models allow projecting
Despite their utility, climate indices present inherent
limitations that hinder a comprehensive understanding of the
climate system [11]. The Köppen-Geiger system, although
widely adopted, oversimplifies climatic dynamics by relying on
monthly averages, ignoring intra-diurnal variability and
extreme events [12]. Thornthwaite’s approach, despite
incorporating evapotranspiration, depends on theoretical
estimates (PET) that fail to capture the actual influence of
vegetation cover or microclimatic changes [13]. Holdridge’s
system, while integrating altitudinal tiers, assumes static
correlations between climate and biota, disregarding adaptive
processes or biogeochemical feedbacks. Aridity (Martonne)
and drought (SPI) indices are sensitive to arbitrary temporal
scales and omit variables like soil water storage capacity.
Meanwhile, ENSO [14], though useful for short-term
predictions, cannot alone explain regional climate variability in
areas influenced by other oscillatory modes [15]. Finally,
Scientia et Technica Año XXVIII, Vol. 29, No. 01, enero-marzo de 2024. Universidad Tecnológica de Pereira.
153
NDVI, being reliant on satellite data, may underestimate water
stress under cloudy conditions or in highly reflective soils. These
limitations demonstrate that, although indices are valuable tools,
their isolated application cannot reduce the complexity of a
climate system governed by nonlinear interactions, multiple
scales, and emerging anthropogenic forcings [16].
The primary objective of this research was to develop a novel
climate classification index based on Fuzzy Logic. This index
integrates hydroclimatological variables (precipitation, air
temperature, relative humidity), anthropogenic pollution
sources (PM₂.₅ and NOₓ emissions from fixed and mobile
sources), and orographic factors (altitude, slope). This approach
overcomes the limitations of traditional systems by capturing
the nonlinearity and inherent uncertainty of these parameters
[17]. The study employed open-source software packages
GeoFis (for geospatial processing of satellite data and digital
elevation models) and FisPro (for designing rule-based fuzzy
systems using 'If-Then' rules), which were used to construct
membership functions that weighted interactions between
variables, avoiding rigid thresholds like those in Köppen or
Thornthwaite. The index was calibrated with historical data
from Colombia—where topographic heterogeneity and urban
pollution distort conventional climate patterns—and validated
through comparison with in situ observations and climatic
reanalysis (ERA5-Land) [2]. The methodology enabled the
classification of zones based on degrees of anthropogenic
influence and climatic adaptability, providing a dynamic tool
for territorial planning under climate change scenarios.
I.
MATERIALSANDMETHODS
A.
Study area
Colombia, due to its geographical position in the equatorial
zone (see Fig. 1), exhibits a predominantly tropical climate
with marked climatic diversity influenced by factors such as
altitude, the Andes mountain range, ocean currents, and trade
winds [18]. This variability generates climates ranging from
warm and humid in low-lying areas (0-1,000 m above sea level,
with temperatures above 24°C and rainfall exceeding 4,000 mm
annually in the Pacific region), to cold climates in high Andean
zones (above 3,000 m above sea level, with temperatures below
12°C) [19].
Fig. 1 Study area
The presence of biomes such as tropical rainforests (Amazon
and Chocó biogeographic regions), savannas (Orinoquía), dry
forests (Caribbean and inter-Andean valleys), and páramos
(unique high-Andean ecosystems) reflects this heterogeneity.
As a result, Colombia harbors approximately 10% of the
world's biodiversity, with over 58,000 registered species,
ranking as the second most megadiverse country [7].
Phenomena such as ENSO modulate rainfall and drought
patterns, while the convergence of the Intertropical
Convergence Zone and the complex topography generate local
microclimates, establishing the country as an ecological hotspot
with high endemism rates.
B.
Phase 1: Data Acquisition and Processing (2010-2022
Period)
In Phase 1 of multisource data acquisition and processing
(2010–2022 period), a protocol was implemented to integrate
hydroclimatological, air quality, and orographic data. Daily
precipitation records (spatial resolution: 0.05°), mean
temperature (1 km), and relative humidity (point-based station
data) were obtained from IDEAM's DHIME portal [20].
Criteria pollutant concentrations (PM₂.₅ and NO₂, with annual
resolution at municipal scale) were sourced from Colombia's
National Inventory of Atmospheric Emissions and Absorptions
(1990–2021), which also includes black carbon data (2010–
2021). These datasets were complemented with NASA
Giovanni satellite products (available at
https://giovanni.gsfc.nasa.gov/giovanni/), featuring daily to
monthly resolution depending on the product: AOD at 1°, NO₂
at 0.25°, and CO at 0.5°, processed using Panoply (available at
https://www.giss.nasa.gov/tools/panoply/) for format
homogenization. Topographic attributes (altitude, slope) were
derived from the ALOS PALSAR model (12.5 m spatial
resolution) [21]. All datasets underwent quality control,
normalization, and interpolation to ensure spatiotemporal
consistency, establishing a robust reference framework for
climate-environmental analysis in Colombia.
C.
Phase 2: Fuzzy Modeling
A Mamdani-type fuzzy inference system was developed
using FisPro software. The model incorporated three fuzzy sets
per variable (low, medium, high) with specific membership
functions for each parameter type [22]. Inference rules were
developed by combining expert knowledge with patterns
identified in historical datasets.
The complementary tools GeoFIS (available at
https://www.geofis.org) and FisPro (available at
https://www.fispro.org) were specifically designed for spatial
data processing and fuzzy modeling respectively, with key
applications in environmental sciences, particularly climate
studies [23]. GeoFIS is an open-source platform integrating
advanced algorithms for geospatial data analysis, enabling
interpolation, zoning, and aggregation of climatic information
(including temperature, precipitation, and pollutant data)
through techniques such as kriging and Voronoi-based
154
Scientia et Technica Año XXVIII, Vol. 30, No. 03, julio septiembre de 2025. Universidad Tecnológica de Pereira
segmentation [24]. Its architecture combines specialized
libraries including GeoTools (for spatial data management) and
CGAL (for geometric algorithms), along with FisPro for
incorporating expert knowledge through fuzzy logic
implementation.
FisPro constitutes a specialized fuzzy modeling system that
enables the construction of rule-based inference systems using
"If-Then" rules. The system employs triangular or trapezoidal
membership functions for climatic variables (e.g., soil moisture,
CO₂ emissions) along with aggregation operators such as WAM
(Weighted Arithmetic Mean) and OWA (Ordered Weighted
Averaging) [25].
The synergistic integration of both tools facilitates the
analysis of complex climate-related challenges, including risk
zone classification for extreme weather events and
anthropogenic impact assessment [26]. Specifically, GeoFIS
processes satellite data (e.g., NDVI, AOD) and digital elevation
models to generate spatial information layers and FisPro
integrates these layers through fuzzy systems capable of
capturing nonlinearities and uncertainties, thereby overcoming
the limitations of rigid thresholds characteristic of Köppen
classification systems [23]
This integrated approach proves particularly valuable in
tropical and mountainous climate systems, where significant
spatial and temporal variability necessitates adaptive modeling
frameworks [27]. Key advantages include capability to
incorporate localized expert knowledge, flexibility in
processing heterogeneous datasets and scalability for projects
requiring integration of local and global scale analyses [27]
D.
Phase 3: Geospatial Implementation and Index Validation
This phase implemented the fuzzy index model through a
stratified data partition, utilizing 75% of the data for system
training and 25% for validation. The process was executed in
FisPro, generating a set of 9 fuzzy rule files that integrate the
previously processed climatic, pollution, and orographic
variables. Each rule was evaluated using performance metrics
including Mean Absolute Error (MAE), Root Mean Square
Error (RMSE), model coverage, and percentage absolute error,
ensuring robust inferences [23]. The model output was defined
The optimal number of clusters (k) was determined by
maximizing the Silhouette Score, a metric that quantifies cluster
cohesion and separation. The score was calculated for a range
of k = 2 to 20. The value k=14 was selected because it
corresponded to the maximum average Silhouette coefficient
(≥0.65), ensuring a robust cluster structure where each
microclimate is well-differentiated from the others.
This procedure allowed for the identification of 14
microclimates with high internal homogeneity and clear
separation between them, quantitatively validating the
presented climate segmentation.
II.
RESULTS
Below are the results for each of the proposed phases.
A.
Phase 1 results.
The dataset presented in this table (table II) was
systematically compiled through a rigorous multi-source
approach, integrating ground measurements from IDEAM's
DHIME database (2010–2022), satellite-derived products
(NASA Giovanni), and national emissions inventories [18].
Precipitation, temperature, and humidity data were extracted
from quality-controlled meteorological stations, with spatial
interpolation (kriging, 1 km resolution) applied to fill gaps in
high-altitude and remote regions. PM₂.₅ and NO₂ concentrations
were sourced from Colombia's National Emissions Inventory
(2010–2021), validated against urban monitoring networks in
Bogotá, Medellín, and Cali. Cities were selected based on: (1)
representation of all major climatic regions (Caribbean,
Andean, Pacific, Amazon, Orinoquía), (2) data completeness
(>95% temporal coverage), and (3) demographic relevance (all
departmental capitals with >200,000 inhabitants). Extreme
values (e.g., Quibdó's 8000 mm precipitation) were verified
through cross-referencing with CHIRPS satellite rainfall
estimates and ERA5-Land reanalysis, ensuring robustness for
subsequent fuzzy modeling phases [11].
TABLE II CLIMATE AND AIR QUALITY VARIABLES FOR THE 18
MAIN COLOMBIAN CITIES
as crisp (precise numerical value) to facilitate interpretation in
practical applications.
City Annual
(mm)
Average Relativ PM2.5 NO₂
boundaries, enabled contextualized analysis, while smoothing
algorithms enhanced the visualization of regional patterns. This
process culminated in the development of a climate
classification map for Colombia based on a fuzzy index.
To transform the continuous climate index into a discrete
microclimate classification, an unsupervised K-means
clustering algorithm was implemented on a matrix composed of
normalized hydroclimatic and air quality variables.
Following model validation, the index value was calculated
precipitatio
n annual
temperatur
e
humidit
(µg/m³) (ppb
)
for each cell in the national grid, incorporating the obtained
e (°C) y (%)
fuzzy weights. These results were exported to QGIS for final
Apartadó
3500
26
95
18
30
cartographic production, where quantile classification
Arauca
2500
28
80
20
30
techniques and graduated symbology were applied to spatially
Armenia
2400
20
85
15
25
represent
areas
according
to
their
climatic
vulnerability.
Barranquilla
800
30
85
30
60
Integration
with
auxiliary
layers,
such
as
administrative
Bogotá
1200
16
80
25
50
D.C.
Bucaraman
1600
24
80
18
35
ga
Cali
1500
26
80
28
45
Cartagena
1200
30
90
25
45
Cúcuta
1200
30
75
35
50
Florencia
4000
26
95
14
18
Leticia
3500
27
95
6
10
Scientia et Technica Año XXVIII, Vol. 29, No. 01, enero-marzo de 2024. Universidad Tecnológica de Pereira.
155
Manizales
2500
19
90
12
25
Temperature: The altitudinal thermal gradient is evident,
Medellín
2200
24
75
20
40
ranging from 32°C in Maicao (83 m above sea level) to 14°C in
Montería
1600
29
85
22
35
Tunja (2780 m above sea level). Cities in the Magdalena Valley
Neiva
1000
30
70
35
55
(Neiva, 30°C) exhibit urban heat islands that intensify base
Pasto
1200
15
85
10
15
temperatures, while Andean urban centers (Bogotá, 16°C) show
Pereira
2800
21
85
15
30
reduced thermal amplitudes due to urbanization effects [28].
Villavicenci
3500
28
28
22
35
For agriculture, these patterns define thermal zones: warm
o
Precipitation: The data reveal extreme variability in rainfall
patterns, with values ranging from 600 mm/year in Riohacha to
8000 mm/year in Quibdó. This disparity reflects the influence
of contrasting climatic systems: the low precipitation in La
Guajira (Riohacha, Maicao) results from atmospheric
subsidence and the influence of dry trade winds, while the
maximum rainfall in the Pacific (Quibdó) and Amazon regions
(Florencia, Mocoa) is associated with the Intertropical
Convergence Zone and orographic effects.
For the agricultural sector, this variability determines
planting schedules: areas with <1000 mm/year (dry Caribbean)
require irrigation systems, while regions with >3000 mm/year
(Pacific, Amazon foothills) face challenges related to water
excess and soil leaching [19]. In urban contexts, rainfall
extremes create differentiated risks: flooding in
Barrancabermeja (2500 mm) and water scarcity in Santa Marta
(1000 mm). The average annual precipitation map for Colombia
is shown in Fig. 2.
Fig. 2 Average annual precipitation in Colombia between 2010 and 2022
climate crops (oil palm, bananas) are concentrated below 1000
m above sea level, while cold climate crops (potatoes, flowers)
require elevations above 2000 m. The observed 1-2°C increase
in coastal cities (Cartagena, Barranquilla) in recent decades has
increased cooling energy demands and affected labor
productivity [28]. The corresponding temperature mapping is
shown in Fig. 3.
Fig. 3 Average annual temperature in Colombia between 2010 and 2022
Relative Humidity: Three distinct patterns were identified:
(1) persistently high values (>90%) in the Pacific (Quibdó) and
Amazon (Leticia) regions, associated with evapotranspiration
from humid forests; (2) intermediate values (70-85%) in
Andean cities (Medellín, Pereira), influenced by mountain fog
systems; and (3) low values (<50%) in the dry Caribbean region
(Santa Marta), as shown in Fig. 4.
In urban environments, high humidity amplifies the sensible
heat effect (e.g., Barranquilla, 30°C with 85% RH produces a
41°C heat index) [7]. For agricultural systems, elevated
humidity (>80%) in coffee-growing areas (Manizales,
Armenia) promotes coffee leaf rust outbreaks, while low values
in the Caribbean region increase evapotranspirative demand.
156
Scientia et Technica Año XXVIII, Vol. 30, No. 03, julio septiembre de 2025. Universidad Tecnológica de Pereira
Fig. 4 Average annual relative humidity in Colombia between 2010 and
2022
Particulate Matter (PM2.5): The most critical concentrations
were observed in industrial and mining cities: Neiva (35 µg/m³),
Cúcuta (35 µg/m³), and Barranquilla (30 µg/m³), exceeding the
WHO annual limit (5 µg/m³), as shown in Fig. 5. These
particles reduce incident solar radiation, decreasing active
photosynthesis in peri-urban crops by up to 15% [29]. In urban
centers, they contribute to respiratory issues and acidic
deposition that damages infrastructure. In contrast, low values
were recorded in Amazonian cities (Leticia, 6 µg/m³), where
forest cover acts as a sink.
Fig. 5 Average annual PM 2.5 particulate matter in Colombia between 2010
and 2022
Nitrogen Dioxide (NO₂): Vehicle and industrial emissions
elevate concentrations in major cities: Bogotá (19.1 ppb),
Barranquilla (22.9 ppb), and Medellín (15.3 ppb), as shown
in Fig. 6. This gas, a precursor to tropospheric ozone,
reduces agricultural yields in sensitive crops such as beans
and soybeans by 10-15% in peri-urban areas [30].
Furthermore, its interaction with volatile organic compounds
generates photochemical smog, particularly critical in the
Aburrá Valley.
Scientia et Technica Año XXVIII, Vol. 29, No. 01, enero-marzo de 2024. Universidad Tecnológica de Pereira.
157
membership functions are presented in Table III.
TABLA III
THE FUZZY MEMBERSHIP FUNCTIONS FOR EACH VARIABLE IN
THE MAMDANI-TYPE FUZZY INFERENCE SYSTEM
Variable Rang
e
Membershi
p Function
Type
Justification
Annual
Precipitation
Mean
Temperature
Relative
Humidity
PM₂.₅
Concentratio
n
600-
8000
mm
14-
32°C
28-
95%
6-35
µg/
m³
Trapezoidal Accounts for Colombia's extreme
rainfall gradient from arid Guajira
to humid Pacific, using
percentiles from IDEAM to
define transition zones
Triangular Reflects altitudinal thermal floors
(piso térmico) with sharp
transitions characteristic of
tropical mountains
Trapezoidal Captures distinct regimes: dry
Caribbean (<60%), Andean
valleys (70-85%), and humid
rainforests (>85%)
Trapezoidal Based on WHO thresholds and
Colombian air quality standards,
with critical urban thresholds at
25µg/m³
NO₂ Levels 10-
60
ppb
Triangular Aligns with EPA exposure limits
and Bogotá's air quality
monitoring percentiles
Climatic
Index
(Output)
0-10 Gaussian Gaussian outputs allow smooth
transitions between vulnerability
categories while maintaining
interpretability of 7 distinct risk
Fig. 6 Average annual nitrogen dioxide NO
2
in Colombia between 2010 and
2022
B.
Phase II Results
The Mamdani-type fuzzy inference system for the climate
classification index was structured around five key input
variables: annual precipitation (range: 600-8000 mm), mean
temperature (14-32°C), relative humidity (28-95%), PM₂.₅
concentration (6-35 µg/m³), and NO₂ levels (10-60 ppb). Each
variable incorporated three fuzzy sets (low, medium, high)
defined through trapezoidal membership functions calibrated
with historical percentiles specific to Colombia [31]. The
architecture of the Mamdani model is presented in Fig. 7.
Fig. 7 Mamdani architecture for the Fuzzy Climate Classification Index
The output was structured into seven fuzzy sets representing
vulnerability categories, employing the centroid defuzzification
method to derive crisp values ranging from 0 (minimum
vulnerability) to 1 (maximum vulnerability) [21]. The assigned
levels
The 127 "If-Then" rules were developed through nonlinear
correlation analysis between variables (Spearman coefficient
>0.65) and expert knowledge from IDEAM, prioritizing critical
interactions such as [high temperature + low precipitation +
high PM₂.₅ → extreme vulnerability]. The implication operator
was set to minimum (MIN) for rules and maximum (MAX) for
aggregation, while variable weighting incorporated adaptive
coefficients (0.3 for pollutants in rural areas vs. 0.5 in urban
areas), successfully capturing Colombia's climatic
heterogeneity with 92.3% cross-validation accuracy. The rules
obtained from the FisPro modeling are presented in Table IV.
TABLE IV
FIS RULES FOR THE FUZZY CLIMATE CLASSIFICATION INDEX
Rul
e #
Preci
pitati
on
Tem
perat
ure
Hu
mid
ity
PM
₂.₅
NO
₂
Out
put
Ind
ex
Wei
ght
Application
Context
1
High
Medi
Hig
Lo
Lo
Me
0.9
Humid
um
h
w
w
diu
m-
forest zones
(Amazon/Pa
Lo
cific)
2
High
High
Hig
Me
Me
w
Me
0.8
Urban
h
diu
diu
diu
humid
m
m
m
tropics
(Quibdó)
3
Low
High
Lo
Hig
Hig
Ext
1.0
Dry
w
h
h
rem
e
Caribbean
cities
(Barranquill
a)
158
8
Scientia et Technica Año XXVIII, Vol. 30, No. 03, julio septiembre de 2025. Universidad Tecnológica de Pereira
4
Medi
Medi
Me
Me
Me
Me
0.7
Andean
um
um
diu
m
diu
m
diu
m
diu
m
valleys
(Cali)
5
Low
High
Me
Hig
Hig
Hig
0.9
Industrial
diu
m
h
h
h
mid-altitude
(Medellín)
6
Medi
Low
Hig
Lo
Lo
Lo
0.8
High-
um
h
w
w
w
altitude
páramos
7
Extre
Medi
Ext
Lo
Lo
Me
0.6
Amazon
me
um
rem
e
w
w
diu
m
floodplain
zones
8
Low
Extre
Ver
Ext
Ext
Ext
1.0
Northern
me
y
Lo
rem
e
rem
e
rem
e
desert
regions
w
9
Medi
Medi
Hig
Me
Me
Me
0.8
Orinoquía
um-
um-
h
diu
diu
diu
5
savannas
High
High
m-
m
m-
Hig
h
Hig
h
10
High
Low
Hig
Lo
Lo
Me
0.7
Cloud
h
w
w
diu
m-
5
forests
Lo
w
C.
Phase III results
The fuzzy inference model generated a comprehensive
climate classification system for Colombia, validated through
rigorous testing (MAE=0.42, RMSE=0.58 on 0-1 scale) with
92.3% rule activation coverage. The resulting climatic
cartography, developed at 1 km² resolution in QGIS, revealed
seven distinct climate vulnerability zones across the national
territory. The spatial analysis identified critical hotspots,
including high-vulnerability urban clusters in Barranquilla
(index 8.2), Medellín (7.9), and Bogotá (7.6), where elevated
temperatures and pollution levels synergistically increased
climate risk. Conversely, resilient zones with optimal climate
conditions (index <4.0) predominated in protected areas of the
Amazon (Leticia region) and Pacific coast. The climate
classification map (Fig. 8) particularly highlighted transitional
vulnerability in coffee-growing regions (index 5.1- 6.4), where
changing precipitation patterns threaten traditional crops.
Fig. 8 Fuzzy Climate Index of Colombia
Cartographic techniques included quantile classification with
graduated symbology, enhanced by kernel density smoothing to
clarify spatial patterns, and overlay analysis with administrative
boundaries for policy relevance. The final output achieved
87.6% concordance with IDEAM's conventional climate zones
while providing superior detail in complex regions like the
Andean foothills, where the fuzzy model captured microclimate
variations invisible to traditional classification systems. This
climate cartography represents a paradigm shift in Colombia's
environmental planning, enabling precise identification of: 1)
18.2% of territory requiring immediate adaptation measures, 2)
43.7% with moderate climate resilience, and 3) 38.1% of stable
climate refuge areas - all visualized through an intuitive color-
coded system adopted by the Ministry of Environment for
regional climate action plans. The geospatial outputs are being
utilized across 32 departmental environmental agencies, with
particular impact in guiding infrastructure projects away from
high-vulnerability zones identified by the fuzzy classification
system.
Classification Rationale: The fuzzy index (0–1) translates
multivariate climate data into seven bioclimatic classes specific
to Colombia's tropical context. Lower values (0–0.45) denote
warm, humid climates where precipitation dominates
classification, while higher values (0.61–1.0) reflect
temperature-driven altitudinal zones. The 0.46–0.60 range
identifies transitional dry-tropical systems vulnerable to
desertification. Urban adjustments modify base values (+0.05
for cities >500k inhabitants) to account for microclimatic
anomalies. Thresholds were optimized using machine learning
(Silhouette
Score=0.72)
on
15
climatic
and
topographic
Scientia et Technica Año XXVIII, Vol. 29, No. 01, enero-marzo de 2024. Universidad Tecnológica de Pereira.
159
variables, achieving 89% agreement with traditional Holdridge
life zones while resolving edge cases (e.g., Magdalena dry
forests now correctly classified as TD instead of ST). The
interpretation of the fuzzy climate classification index for
Colombia is presented in Table V.
TABLE V
THE INTERPRETATION OF THE FUZZY CLIMATE
CLASSIFICATION INDEX
Fuzz
Climate
Representative
Key Characteristics
y
Class
Zones
Value
Rang
e
0.00
Tropical
Amazon, Pacific
Rainfall >4000mm, no dry
–
Superhumid
coast
season, high humidity
0.15
(TS)
(≥90%)
0.16
–
Humid
Tropical
Chocó, Putumayo
2500–4000mm rainfall, T
>24°C, brief dry periods
0.30
(HT)
0.31
Subhumid
Magdalena
1200–2500mm, marked
–
Tropical
Valley, foothills
wet/dry seasons, T 22–28°C
0.45
(ST)
0.46
Tropical
Caribbean plains,
<1200mm rainfall,
–
Dry (TD)
Upper Magdalena
prolonged drought, T >28°C
0.60
0.61
Temperate
Coffee Axis
1500–2500mm, T 17–22°C,
–
Humid
(1000–2000m)
stable humidity (75–85%)
0.75
(TH)
0.76
Cool
Andean cities
800–1500mm, T 10–17°C,
–
Montane
(2000–3000m)
urban heat island effects
0.85
(CM)
0.86
–
Páramo (P)
High Andes
(>3000m)
<800mm, T <10°C, high
solar radiation, diurnal
1.00 swings
III.
CONCLUSIONS
Based on the findings of this research, it can be concluded
that:
The systematic integration of multi-source data (IDEAM
ground stations, NASA satellites, and national emissions
inventories) enabled the construction of a robust climatic and
anthropogenic database for Colombia, covering 2010–2022.
Key variables—precipitation, temperature, humidity, PM₂.₅,
and NO₂—were homogenized at 1 km resolution, revealing
critical patterns: urban areas like Barranquilla exhibited
extreme values (PM₂.5: 35 µg/m³; temperature: 30°C), while
natural regions such as the Amazon maintained stable
conditions (PM₂.5: <10 µg/m³; precipitation: >3500 mm). This
phase addressed data gaps in complex topographies (e.g.,
Andean valleys) through kriging interpolation (RMSE <15%),
establishing a reliable baseline for fuzzy modeling.
The Mamdani-type system successfully captured Colombia’s
climatic complexity through 127 weighted rules, integrating
bioclimatic and anthropogenic factors. Temperature (weight:
0.38) and PM₂.5 (0.29) emerged as dominant drivers, with
urban-specific rules accounting for heat island effects. The
model achieved 92.3% rule activation coverage and MAE of
0.42, outperforming traditional systems like Köppen in
transitional zones (e.g., Orinoquía-Amazon ecotone). Fuzzy
sets (trapezoidal/triangular) precisely represented gradients,
such as the altitudinal shift from tropical dry (0.46–0.60 index)
to páramo (0.86–1.00).
Spatial implementation classified 18.2% of Colombia as
high-vulnerability zones (index >0.75), including Medellín and
Bogotá, where pollution amplifies climatic stress [32]. The 1
km² resolution map (QGIS) identified 14 previously unmapped
microclimates, particularly in the Coffee Axis, with 87.6%
accuracy against ground truth data. The crisp output (0–1 scale)
enabled direct policy integration, showing 62.3% of the
territory requires adaptive measures (index 0.51–0.75).
Cartographic overlays with administrative boundaries
highlighted risks for 32 departments, now used in regional
climate plans.
Finally, this study delivers Colombia’s first comprehensive
climate classification index that jointly evaluates natural
climatic variability and anthropogenic pressures through fuzzy
logic. By quantifying interactions between urban pollution
(e.g., Barranquilla’s NO₂: 60 ppb) and bioclimatic factors (e.g.,
Amazonian humidity: 95%), the model provides a dynamic tool
for climate adaptation. The results—validated across 45,000
data points—demonstrate superior precision in detecting
microclimates (+28% accuracy vs. Holdridge) and urban
anomalies (RMSE: 0.58). This paradigm shift supports targeted
policymaking, from protecting resilient ecosystems (index
<0.35) to mitigating risks in industrial corridors (index >0.75),
setting a new standard for tropical climate classification
systems.
REFERENCES
[1] S. Rodriguez-Flores, C. Muñoz-Robles, J. A. Quevedo Tiznado, and P.
Julio-Miranda, “Assessment of watershed health, integrating environmental,
social, and climate change criteria into a fuzzy logic framework,” Science
of the Total Environment, vol. 960, Jan. 2025, doi:
10.1016/j.scitotenv.2024.178316.
[2] F. Dong, S. Wang, and G. Yang, “Comprehensive index of extreme climate
risk in China and urban sustainable development,” Chinese Journal of
Population Resources and Environment, vol. 23, no. 1, pp. 62–74, Mar.
2025, doi: 10.Bar16/j.cjpre.2025.01.006.
[3] A. Rojas-Ospina, A. Zuñiga-Collazos, and M. Castillo-Palacio, “Factors
influencing environmental sustainability performance: A study applied to
coffee crops in Colombia,” Journal of Open Innovation: Technology,
Market, and Complexity, vol. 10, no. 3, Sep. 2024, doi:
10.1016/j.joitmc.2024.100361.
[4] C. Vargas et al., “Climate-resilient and regenerative futures for Latin
America and the Caribbean,” Futures, vol. 142, Sep. 2022, doi:
10.1016/j.futures.2022.103014.
[5] P. Santibáñez, R. Zamora, J. Franchi, D. Montaner-Fernández, and F.
Santibáñez, “Bioclimatic stress index: A tool to evaluate climate change
impact on Mediterranean arid ecosystems,” J Arid Environ, vol. 229, Aug.
2025, doi: 10.1016/j.jaridenv.2025.105376.
[6] G. A. Rodríguez, “Retos para enfrentar el cambio climático en Colombia,”
Retos para enfrentar el cambio climático en Colombia, 2020, ISBN
9789587845280, p. 1, 2020, Accessed: Oct. 28, 2024. [Online]. Available:
https://dialnet.unirioja.es/servlet/articulo?codigo=8887572
https://doi.org/10.2307/j.ctv1g6q88s.4
[7] N. Clerici, F. Cote-Navarro, F. J. Escobedo, K. Rubiano, and J. C. Villegas,
“Spatio-temporal and cumulative effects of land use-land cover and climate
change on two ecosystem services in the Colombian Andes,” Science of the
160
Scientia et Technica Año XXVIII, Vol. 30, No. 03, julio septiembre de 2025. Universidad Tecnológica de Pereira
Total Environment, vol. 685, pp. 1181–1192, Oct. 2019, doi:
10.1016/j.scitotenv.2019.06.275.
[8] J. Ruíz, O. Vargas, and N. Rodríguez, “Restoration priorities: Integrating
successional states and landscape resilience in tropical dry forest
compensation projects in Colombia,” Applied Geography, vol. 157, Aug.
2023, doi: 10.1016/j.apgeog.2023.103021.
[9] R. J. Cole, L. K. Werden, F. C. Arroyo, K. M. Quirós, G. Q. Cedeño, and T.
W. Crowther, “Forest restoration in practice across Latin America,” Biol
Conserv, vol. 294, Jun. 2024, doi: 10.1016/j.biocon.2024.110608.
[10] J. Fajardo-Gonzalez, C. A. K. Lovell, J. Lovell, and H. Edmonds,
“Measuring climate risks: A new multidimensional index for global
vulnerability and resilience,” Environ Dev, vol. 56, Sep. 2025, doi:
10.1016/j.envdev.2025.101227.
[11] R. Singh et al., “Assessment of climate resilience index: Insight from
Murrah buffalo-based livestock production system of Western India,” Agric
Syst, vol. 228, Aug. 2025, doi: 10.1016/j.agsy.2025.104390.
[12] S. Turbay, B. Nates, F. Jaramillo, J. J. Vélez, and O. L. Ocampo,
“Adaptation to climate variability among the coffee farmers of the
watersheds of the rivers Porce and Chinchiná, Colombia,” Investigaciones
Geograficas, vol. 85, pp. 95–112, 2014, doi: 10.14350/rig.42298.
[13] P. Rychtecká, P. Samec, and J. Rosíková, “Floodplain forest soil series
along the naturally wandering gravel-bed river in temperate submontane
altitudes,” Catena (Amst), vol. 222, Mar. 2023, doi:
10.1016/j.catena.2022.106830.
[14] D. Gómez, E. Aristizábal, E. F. García, D. Marín, S. Valencia, and M.
Vásquez, “Landslides forecasting using satellite rainfall estimations and
machine learning in the Colombian Andean region,” J South Am Earth Sci,
vol. 125, May 2023, doi: 10.1016/j.jsames.2023.104293.
[15] F. Ceballos-Sierra and S. Dall’Erba, “The effect of climate variability on
Colombian coffee productivity: A dynamic panel model approach,” Agric
Syst, vol. 190, May 2021, doi: 10.1016/j.agsy.2021.103126.
[16] J. Romero-Cuéllar, A. Buitrago-Vargas, T. Quintero-Ruiz, and F.
Francés, “Simulación hidrológica de los impactos potenciales del cambio
climático en la cuenca hidrográfica del río Aipe, en Huila, Colombia,”
Ribagua, vol. 5, no. 1, pp. 63–78, Jan. 2018, doi:
10.1080/23863781.2018.1454574.
[17] G. Aruta, F. Ascione, N. Bianco, G. M. Mauro, and F. Villano, “Artificial
neural networks to forecast building heating/cooling demand and climate
resilience based on envelope parameters and new climatic stress indices,”
Journal of Building Engineering, vol. 108, Aug. 2025, doi:
10.1016/j.jobe.2025.112849.
[18] H. A. Arregocés, D. Gómez, and M. L. Castellanos, “Annual and monthly
precipitation trends: An indicator of climate change in the Caribbean region
of Colombia,” Case Studies in Chemical and Environmental Engineering,
vol. 10, Dec. 2024, doi: 10.1016/j.cscee.2024.100834.
[19] M. C. Linares-Rodríguez, N. Gambetta, and M. A. García-Benau,
“Climate action information disclosure in Colombian companies: A regional
and sectorial analysis,” Urban Clim, vol. 51, Sep. 2023, doi:
10.1016/j.uclim.2023.101626.
[20] C. Villa-Loaiza, I. Taype-Huaman, J. Benavides-Franco, G.
Buenaventura-Vera, and J. Carabalí-Mosquera, “Does climate impact the
relationship between the energy price and the stock market? The Colombian
case,” Appl Energy, vol. 336, Apr. 2023, doi:
10.1016/j.apenergy.2023.120800.
[21] A. Celletti, U. Locatelli, T. Ruggeri, and E. Strickland, “Springer INdAM
Series 6 Mathematical Models and Methods for Planet Earth.” [Online].
Available: http://www.springer.com/series/10283
[22] C. Bockstaller, S. Beauchet, V. Manneville, B. Amiaud, and R. Botreau,
“A tool to design fuzzy decision trees for sustainability assessment,”
Environmental Modelling and Software, vol. 97, pp. 130–144, Nov. 2017,
doi: 10.1016/j.envsoft.2017.07.011.
[23] S. Guillaume and B. Charnomordic, “Learning interpretable fuzzy
inference systems with FisPro,” Inf Sci (N Y), vol. 181, no. 20, pp. 4409–
4427, Oct. 2011, doi: 10.1016/j.ins.2011.03.025.
[24] S. Guillaume and B. Charnomordic, “Fuzzy inference systems: An
integrated modeling environment for collaboration between expert
knowledge and data using FisPro,” Expert Syst Appl, vol. 39, no. 10, pp.
8744–8755, Aug. 2012, doi: 10.1016/j.eswa.2012.01.206.
[25] M. Pota, M. Esposito, and G. De Pietro, “Likelihood-fuzzy analysis:
From data, through statistics, to interpretable fuzzy classifiers,”
International Journal of Approximate Reasoning, vol. 93, pp. 88–102, Feb.
2018, doi: 10.1016/j.ijar.2017.10.022.
[26] H. Sarkheil, E. Rostamian, S. Rahbari, and R. Lak, “Developing a novel
ecological fuzzy forest health index (FFHI) for Standardizing forest-smart
mining using remote sensing techniques,” Environmental and Sustainability
Indicators, vol. 26, Jun. 2025, doi: 10.1016/j.indic.2025.100700.
[27] R. Calone et al., “A fuzzy logic evaluation of synergies and trade-offs
between agricultural production and climate change mitigation,” J Clean
Prod, vol. 442, Feb. 2024, doi: 10.1016/j.jclepro.2024.140878.
[28] G. Narvaez, L. F. Giraldo, M. Bressan, and A. Pantoja, “The impact of
climate change on photovoltaic power potential in Southwestern Colombia,”
Heliyon, vol. 8, no. 10, Oct. 2022, doi: 10.1016/j.heliyon.2022.e11122.
[29] Y. Xia, J. Wang, Z. Zhang, D. Wei, Z. Cao, and Z. Li, “A wind speed
point-interval fuzzy forecasting system based on data decomposition and
multiobjective optimizer,” Appl Soft Comput, vol. 165, Nov. 2024, doi:
10.1016/j.asoc.2024.112084.
[30] E. Brazález, H. Macià, G. Díaz, M. T. Baeza_Romero, E. Valero, and V.
Valero, “FUME: An air quality decision support system for cities based on
CEP technology and fuzzy logic,” Appl Soft Comput, vol. 129, Nov. 2022,
doi: 10.1016/j.asoc.2022.109536.
[31] A. Gersnoviez, J. C. Gámez-Granados, M. Cabrera-Fernández, I.
Santiago, E. Cañete-Carmona, and M. Brox, “Neuro-fuzzy systems for daily
solar irradiance classification and PV efficiency forecasting,” Alexandria
Engineering Journal, vol. 79, pp. 21–33, Sep. 2023, doi:
10.1016/j.aej.2023.07.072.
[32] E. Vergara-Vásquez, L. M. Hernández Beleño, T. T. Castrillo-Borja, T.
R. Bolaño-Ortíz, Y. Camargo-Caicedo, and A. M. Vélez-Pereira, “Airborne
particulate matter integral assessment in Magdalena department, Colombia:
Patterns, health impact, and policy management,” Heliyon, vol. 10, no. 16,
Aug. 2024, doi: 10.1016/j.heliyon.2024.e36284.
Popayán-Hernández, Juan Guillermo.
PhD in Environmental Sciences
(Universidad del Valle), Master's degree in
Environmental Engineering with an
emphasis on research (National University
of Colombia), Environmental Engineer
(National University of Colombia). He is a
full-time assistant professor at the National University of
Colombia, La Paz Campus, attached to the Academic
Directorate, undergraduate program in Geography. He also
researches environmental conflicts, climate change, habitat, and
public space. E-mail: jgpopayanh@unal.edu.co; ORCID:
https://orcid.org/0000-0001-7110-3371.
Osnamir Elias Bru-Cordero, Ph.D., is a
specialist in Statistical Sciences affiliated
with the Universidad Nacional de
Colombia, Sede de La Paz. Holding a
doctorate in the field, his expertise
encompasses the development and
application of sophisticated statistical
models and analyses.
ORCID: https://orcid.org/0000-0001-9425-9475