Study area

The city of Durango, Durango, is located between 24°01'26" north and 104°40'13" west at an altitude of 1 890 m (^{Inegi, 2022}), and is the capital of the state of Durango, Mexico (Figure 1). Durango city covers around 10 041 km² and has a population of 688 697 inhabitants with significant urban growth in recent years (^{Inegi, 2022}). Average temperature of the coldest month is 1.7 °C and that of the hottest month 31 °C and 500 mm annual average rainfall.

a) Polygon of Durango city; b) Location of Durango city on the map of Mexico; c) Altitude differentiation in the polygon of Durango city.

Figure 1 Location of the study area. 

Satellite information

Images were obtained from the Landsat 8 OLI/TIRS satellite of Durango city (Table 1). These images correspond to the winter 2021 and spring 2022 seasons and are available from the United States Geological Survey (USGS) (^{USGS, 2017}). In this process, the Rstudio getSpatialData package (^{Kwok, 2018}) was used.

Image preprocessing

In order to eliminate or reduce factors caused by atmospheric and topographic effects that could alter the results, the satellite images were preprocessed, which consisted of a radiometric and atmospheric correction with the help of the Semi Automatic Classification plug-in tool within the QGIS geospatial software version 3.14 (^{QGIS Development Team, 2020}), using the Apparent Reflectance (TOA) technique, which allows transforming the digital levels (DL) to a normalized spectral scale (^{Chávez, 1988}; ^{López-Serrano et al., 2016}).

Coverage classification

A supervised classification was carried out to differentiate public and private green areas, based on the green areas of the city (cadastre) and areas such as parks, medians and gardens or squares were identified through remote sensors. The classification included green areas, urban (concrete, construction and asphalt areas), bare soil and bodies of water.

In addition, the surface area was calculated based on the area extension of each class. Thus, the following spectral index were estimated: Normalized Difference Vegetation Index (NDVI), which was mathematically developed by ^{Rouse et al. (1974)} to improve the visualization of the amount and plant photosynthetic activity on the surface of the Earth.

Where:

NIR = Spectral band in the near infrared region

R = Band in the red region

The Normalized Difference Water Index (NDWI) was estimated using the technique proposed by ^{McFeeters (1996)} where the green portion band (GREEN) and the thermal infrared band (NIR) are used.

Where:

NIR = Near infrared band reflectance values

G = Green portion band reflectance values

The Water Stress Index (MSI) allows identifying land surfaces with higher and lower moisture content. It takes values ranging from 0.4 to 2, and higher values are related to vegetation with water stress (^{Doraiswamy and Thompson, 1982}).

Where:

SWIR = Shortwave infrared band reflectance values

NIR = Near infrared band reflectance values

Object-based image analysis (OBIA) was performed on the images, through segmentation with the Orfeo Tool Box (OTB) algorithm, which consists of grouping contiguous pixels that have similar characteristics and results in a vector layer with as many polygons as the classification process obtain (^{Hossain and Chen, 2019}). The training polygons represented at least 2 % of the total surface of the study area (^{Congalton and Green, 2008}). Finally, the Random Forest classification algorithm was used using the Random Forest library (^{Liaw and Wiener, 2002}) in Rstudio. In order to evaluate the accuracy of the supervised area classification, a confusion matrix was generated to estimate the Kappa statistic, and the user accuracy and classification product were estimated (^{Liu et al., 2007}). 20 % of the training pixels were used for the validation process.

Surface temperature

Surface temperature (ST) was estimated from band 10 for Landsat OLI/TIRS. Thus, the formula that provides the brightness temperature (BT) was applied through the Planck´s law equation (^{Callejas et al., 2011}), with which the pixel value of the thermal infrared band is transformed to values of temperature in Kelvin degrees.

Where:

TB = Brightness temperature

K ₁ and K ₂ = Calibration constants present in the metadata file of each image

L _λ = Spectral radiance obtained from the radiometric correction (TOA), followed by the calculation of surface temperature (Equation 2)

λ = Wavelength not converted to radiance values

ρ = 14 380 value

ε = Earth's surface emissivity

Earth's surface emissivity (^{Becker, 1987}).

Where:

0.986 and 0.0004 = Constant values defined for each sensor

Pv = Vegetative portion estimated with the NDVI values per pixel and the maximum and minimum NDVI values present in the study area

Analysis of thermal data with respect to coverage

Once the supervised classification was carried out, the minimum and maximum temperatures were extracted for each of the classes. Of the types of soil (i. e., green areas, bodies of water, urban soil, bare soil) and the main urban parks, temperature was associated with the type of green area (park, squares and gardens, and medians). Finally, the city polygon was divided according to the cardinal points by zone in the polygon, in order to classify the city for the analysis of the ST according to the distribution of its green areas, starting from the Catedral Basílica Menor de la Inmaculada Concepción (Immaculate Conception Minor Basilica Cathedral), located in the center of the city, as a central point.

Public green areas, temperature and surface

The overall accuracy of the supervised classification was 97 % (Kappa=0.95 %), a result similar to that of ^{Amini et al. (2022)} in a study of the analysis of changes in urban land use and land cover with Landsat images and the Random Forest algorithm (Kappa=93-97 %). Table 2 shows the precision matrix, in which the producer precision is the specific probability of each class correctly classified, while the user precision is the possibility that a specific sample of each class represents the category in reality.

The results of the surface of green area per inhabitant indicate that for public green areas in the city, according to the Municipal Directorate of Public Services for the year 2018 (^{Inegi, 2018}), it was 2.49 m² per inhabitant, while for green areas obtained from ^{INEGI (2005)} was 2.03 m². With this supervised classification methodology, a value of 10.56 m² per inhabitant was calculated. In this case, the green areas of public and private schools, sports and health institutions are considered, which were not included in the other alternatives. These results show that the green area per inhabitant in Durango city is below the recommendation for GEO Cities that establishes 12 m² per inhabitant as a reference value as an indicator of the quality of life in an urban area (^{PNUMA, 2011}). Some studies carried out in the city record similar results, such as those of ^{Blancarte (2016)} who determined an average of 3.67 m² per inhabitant and, if private green areas are included, it is 5.06 m² per inhabitant. On the other hand, ^{Inevap (2021)} mentions 6.64 m² per inhabitant as an area coverage for the year 2021.

In Latin America, ^{Romero (2017)} obtained an average of 2.71 m² per inhabitant in the city of Lima, Peru, a reduced surface area that could have negative effects related to the degradation of ecosystems and detriment to the thermal comfort of the urban population (^{Ullah et al., 2019}). Frequently, the estimate of the square meters per inhabitant ratio is based on information on green areas recorded by the public administration, however, it is important to consider that such spaces with restricted access also offer environmental services to the population (^{Cilliers et al., 2013}).

Surface temperature distribution

Figure 2 illustrates the distribution of surface temperature (ST) in Durango city during winter for the different green areas in which a maximum temperature of 36 °C and a minimum of 13 °C is observed. Table 3 shows the average ST for winter, which is lower in larger green areas (parks) than in squares and gardens, followed by medians with a 0.93 °C difference greater than parks. The maximum temperature was 27.17 °C and the minimum 20.88 °C in squares and gardens. The results coincide with what has been reported by some authors about the negative relationship between the surface of the green area and the decrease in temperature, while the ridges, although small green areas, they should be considered as urban planning tools that achieve positive effects in lowering ST (^{Meyers et al., 2020}).

Figure 2 Map of the distribution of ST in Durango city, Durango, Mexico, during winter 2021. 

Likewise, Figure 3 shows the distribution of the ST in the city for the spring season, when the maximum temperature was 50 °C and the minimum was 24 °C. Table 4 shows that the average ST of Durango city for spring is lower (2.65 °C) in the largest green areas (parks) than in squares and gardens, followed by medians with a difference of 2.14 °C higher than in the parks. In the present study, a lower average ST was recorded in the median areas compared to squares and gardens. Although these results are contrasting with what was expected, an explanation could be related to the fact that the ridges receive constant irrigation from municipal services being easily accessible green areas.

Figure 3 Map of the distribution of ST in Durango city, Durango, Mexico, during spring 2022. 

It should be noted that the maximum temperature in April was 42.7 °C in squares and gardens, and the minimum was 32.17 °C in parks. When comparing both tables for the two seasons (winter and spring), the greatest difference on average was verified in squares and gardens with 7.5 °C between winter and spring; likewise, a difference between the two times of the year of up to 15.58 °C is recognized in the maximum temperature of squares and gardens.

The results show a considerable increase in the ST in spring, but highlights the thermal dynamics in the type of green areas by showing that in this season the ridges seem to have a lower temperature compared to the parks and gardens, which suggests a greater regulatory effect in this season. These data allow to infer that the ridges, despite their small surface area, are important thermal regulation mechanisms for the Durango city. It is recommended to allocate a greater amount of resources (economic and human) to the provision of municipal services for the maintenance of parks and gardens in spring, in order to increase the benefits of these places for its citizens.

Thermal effect by city section

When dividing the city polygon in order to analyze the thermal effect by zone (Figure 4), a constant temperature was observed that kept both periods (winter and spring) cooler with a temperature of 22.80 and 36.60 °C, respectively (Table 5). This is explained by the fact that the center of the city is where the largest extension of green areas is concentrated with 250.95 hectares, followed by the southern area with a similar trend in terms of temperature and 149.79 hectares of green areas (Figures 5 and 6). The southeast zone has the minimum extension of green areas and at the same time, a cooler temperature, since in this area there is a large coverage of crop lands, which can serve as thermal regulation areas while providing food provision for the nearby population (^{FAO, 2023}).

Figure 4 Temperature distribution map by area in Durango city, Durango, Mexico. 

Figure 5 Temperature and green areas by zone in the winter period. 

Figure 6 Temperature and green areas by zone in the spring period. 

Finally, the areas that show a higher temperature are the northwest and the north, where the first has 79.17 ha and temperatures of 24.38 and 38.47 °C in each period, and the north part has only 47.11 ha of green areas and a temperature of 24.67 °C in winter (Figure 5) and 38.61 °C in spring (Figure 6).

The extension of green areas in a city has the ability to reduce the size and intensity of the urban heat island (UHI) through the evapotranspiration process considered as a natural cooling mechanism (^{Qiu et al., 2013}). With these results it is observed that the greater the extension of the green area, the lower the temperature; this agrees with ^{Kurbán et al. (2007)} who indicate that a minimum surface with vegetation will produce an increase in relative humidity, but its scope will be limited to the area of the green space (GS) itself. Only by increasing this surface area to 2 500 m² will the scope of the effect cover the surrounding urban area, consequently, the ST of the surrounding urban areas will be influenced by the green areas: the greater their extension, the greater their regulatory effect.

On the other hand, the trend of apparent ST is higher in urban land than in green areas. ^{Soto-Soto et al. (2020)} obtained results that demonstrate this in continuous and discontinuous urban soils which present higher ST with average values of 23.2 °C and 21.6 °C respectively, unlike low vegetation with 20 °C and forest vegetation with 18.8 °C.

^{López et al. (2021)} reached similar conclusions for the arid city of Hermosillo, Sonora, where they identified urban heat islands and urban oases using Landsat 8 image processing; the highest ST points were on bare soil with 44.2 °C and the urban oases varied in green areas and grasslands such as golf courses, which were estimated using the Moran index.