Spatio-Temporal Behavior of Land Surface Temperatures (LSTs) in Central Chile, Using Terra MODIS Images

Abstract

Land surface temperature (LST) is one of the most important variables in the physical processes of surface energy and water balance. The temporal behavior of LST was analyzed between the latitudes 32°00′ S and 34°24′ S (Valparaíso and Metropolitana regions of Chile) for three summer months (December, January, and February) in the 2000–2017 period, using the Terra MODIS image information and applying the Mann–Kendall test. The results show an increase in LST in the study area, particularly in the Andes mountain range in January (5240 km²), which mainly comprises areas devoid of vegetation and eternal snow and glaciers, and are zones that act as water reserves for the capital city of Santiago. Similarly, vegetated areas such as forests, grasslands, and shrublands also show increasing trends in LST but over smaller surfaces. Because this study is regional, it is recommended to improve the spatial and temporal resolutions of the images to obtain conclusions on more local scales.

1. Introduction

The Earth’s surface can be considered from a geographical point of view as a layer formed by different components, related to land use and cover. This approach allows an overview to study the implications of land surface temperature (LST) and its possible impact in the context of climate change. LST is an important variable [1,2], basically because it is part of the energy balance in the Earth’s climate system [3]. From a physical point of view, LST can be defined as the Earth’s surface temperature (skin temperature) or radiometric temperature of the Earth’s surface [4,5]. The dynamics of LST, obtained mainly by satellites that have one or more far-infrared bands, have been used to study the geophysical processes that occur on the Earth’s surface and impact ecosystems at a regional level. However, studies at this spatial scale are few [3].

The study of LST provides information related to temporal spatial variations in surface temperatures and is fundamental for many applications [6,7,8], including evapotranspiration, climate change, hydrological cycle, vegetation monitoring, urban climate and environmental studies [8,9,10,11,12,13], and has been recognized as one of the parameters with the highest priority in the International Geosphere and Biosphere Program (IGBP) [14].

Remote sensors are used to determine LST, recording land use emission temperature values and allowing researchers to determine its spatial distribution [15,16,17,18,19]. The most commonly used sensors for these measurements are Landsat series, AVHRR (Advanced Very High-Resolution Radiometer), Terra ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer), and Terra MODIS (Moderate Resolution Imaging Spectroradiometer). Terra MODIS sensors provide a series of temporarily consistent LST data as opposed to AVHRR time series, which are defective due to the orbital drift of their platform [4,20,21]. In addition, Terra MODIS produces specific by-products corrected for the study of daily surface temperature and its frequency [22,23]. Cheval and Dumitrescu [24] found that satellite temperature measurements provide better results than those obtained from interpolated ground stations.

According to Alshaikh [25] and Khandelwal et al. [26], studying LST’s spatio-temporal behavior is important because changes in land use, especially the growth of cities, can cause an increase in LST that can alter the flow of energy and nutrients from the surrounding ecosystems. Another topic of analysis is the measurement of the thermal impact due to the development of new urbanizations and to evaluate the mitigation strategies of urban heat islands through the management of land uses, which You et al. [27] address with Landsat 8 images on urban infrastructure.

Other studies correlated LST with MODIS standardized vegetation index (NDVI) [4,28,29,30], land use, rainfall, elevation [31], and air temperature data collected through stations. According to Jimenes-Munoz et al. [32] and Li et al. [33], this correlation is positive especially during stable atmospheric conditions and diurnal periods, being negatively affected by factors such as soil moisture, cloud cover, and the presence of wind. They always concluded that LSTs are more correlated with NDVI than with rainfall, showing an inverse correlation with elevation, which is maintained throughout all seasons of the year [34,35,36]. Fan et al. [37] suggested that climatic variation is mostly determined by the different types of vegetation cover on the Earth’s surface (not by the rapid climate change attributable to atmospheric sources). The authors also concluded that because LST is a transient and highly non-uniform parameter, it is very important to correlate it with stable environmental features such as topography and land-use types. Therefore, it is necessary to conduct studies on the spatio-temporal variation in LST in central Chile, provided that the economic growth of the area has resulted in a change in land-use configuration [38,39]. Likewise, the effects of climate change impact vegetation and water availability, causing accelerated degradation in areas declared as global biodiversity hotspots [40].

The objective of this study was to determine the spatio-temporal behavior of LSTs in central Chile, through trend analyses over a period of 18 years, also including the relationship between different types of use and geomorphology and LST tendencies. Our hypothesis is that LSTs tend to increase and that land use and the geomorphology of the territory have an effect on these trends.

2. Materials and Methods

2.1. Study Area

The study took place in central Chile (Valparaíso and Metropolitana administrative regions) between 32° and 34° south latitudes (Figure 1), covering a total area of 31,367 km². Geomorphologically speaking, this territory is characterized by four macro units from east-to-west [41]: the Andean mountain range (with maximum elevation of 6500 m.a.s.l.), the intermediate depression (central valley, whose average elevation is 550 m.a.s.l.), the coastal mountain range (with elevations ranging between 700 and 2200 m.a.s.l.), and the coastal plains. These units are developed within an average distance of 200 km, producing an abrupt landscape with steep slopes and deep river canyons [42]. Alternating thermal belts can be described from sea-to-inland, giving a variety of climates with moderate temperatures and coastal humidity such as coastal slopes with higher humidity, greater thermal amplitude in the coastal mountain range, strong daily thermal difference with low rainfall amounts in the central valleys, and thermal decrease and increased rainfall and snow on the Andean slopes. In general, four types of climates are distinguished in this zone: Mediterranean winter rainfall climate (Csb), Mediterranean climate with winter rainfall, highland variant (Csb [h]), Tundra climate with dry summers (Et [s]), and Cold semi-arid climate with dry summers (Bsk [s]) [43].

During 2016, the weather stations recorded the highest temperatures in 56 years (1961–2016), and both maximum and minimum temperatures gradually increased during that period [44].

2.2. Satellite Information

MODIS atmospheric- and surface emissivity-corrected images were used in this study, obtained from the US Geological Survey’s Global Visualization Viewer image server. The thermal sensors of the satellites produce daily LST maps, with global coverage in spatial resolution of 1 km × 1 km, coded as MOD11A1 LST [22]. This product was used in its 8-day version, where each pixel value in MOD11A2 is a simple average of all the MOD11A1 LST pixels collected within that time period [45]. The data format was HDF-EOS (Hierarchical Data Format—Earth Observing System, applied to the Earth Observation System) [46].

As supplementary information for the space–time LST analysis, the following imaging systems were used: (1) “Chile’s native plant resources inventory, for the Valparaíso, Metropolitana, and O’Higgins regions” [38], at a scale of 1:35,000, which classifies the current land use into eight main categories [47] and (2) Albers’s [48] study at a scale of 1:250,000, called “Geomorphological Units of Chile”, which classifies the study area into nine geomorphological types. To fulfill the study’s objectives, the methodological scheme of Figure 2 was developed, illustrating the sequence of each stage.

2.3. Downloading and Image Processing

Two hundred four (204) images from the MOD11A2 LST MODIS product were downloaded, corresponding to LST averages every 8 days during summer periods (December, January, and February) from 2000 to 2017. If there was high cloudiness in the images, they were discarded and therefore not considered in the analysis. These were cut according to the limit of the study area, leaving each one with 31,308 pixels. The unit of measurement of LST was converted from Kelvin (the original unit) to Celsius degrees, multiplying each satellite image by the factor reported by MODIS equal to 0.02 and subtracting 273.15 from each pixel unit. Subsequently, from the 8-day images, the arithmetic mean was calculated to obtain one image per month, which generated three images per year with a total of 51 images for the entire 2000–2017 time interval.

Calculation of LST Anomalies

From the temperatures calculated in the previous step, the anomalies for land use and geomorphology in the studied period were estimated based on Equation (1) as follows:

$$z_i=\frac{y_i-{\overline{x}}_y}{s_y}$$

(1)

where $y_i$ is the LST from pixel $i$ for a specified month and year, ${\overline{x}}_y$ and $s_y$ are the mean and standard deviation, respectively, along months and years for the studied time.

2.4. Estimation of the Trend and Its Significance

The non-parametric Mann–Kendall (MK) test [49] was applied for the summer months, obtaining positive or negative LST trends, at 95% confidence (α = 0.05). This test has been widely applied to hydrometeorological time series for trend detection [50,51,52]. The Mann–Kendall test is a robust technique; it does not require the data to come from a normal distribution, and it is not affected by outliers [53,54,55,56]. In other words, this test allows for the analysis of the statistical significance of the trend in a time series, by comparing the increments with the decreases between consecutive pairs. For this, Equations (2) and (3) were used.

$$\mathrm{S}=\sum _{k=1}^{n-1}\sum _{j=k+1}^{n}sgn(x_j-x_k)$$

(2)

$$sgn\left(x_j-x_k\right)=\left\{\begin{array}{c}1 if x_j-x_k>0\\ 0 if x_j-x_k=0\\ -1 if x_j-x_k<0\end{array}\right.$$

(3)

where $\mathrm{S}$ represents the trend of the analyzed series; $x_j \mathrm{a}\mathrm{n}\mathrm{d} x_k$ are consecutive pairs; and sgn() extracts the sign of the expression displayed in Equation (3). Thus, a positive $\mathrm{S}$ value indicates a growing trend, whereas a negative value is an indication of a negative trend. If $\mathrm{S}$ is calculated with more than eight values, it represents a relatively normal distribution [57], allowing for it to be used for hypotheses comparison. For this, the Z statistic must be calculated using Equations (4) and (5). This test was evaluated considering a significance level of 0.05 (Equation (5)).

$$VAR\left(\mathrm{S}\right)=\frac{1}{18}⌈n\left(n-1\right)\left(2n+5\right)-\sum _{p=1}^q{t}_p\left(t_p-1\right)(2t_p+5)⌉$$

(4)

$$\mathrm{Z}=\left\{\begin{array}{ll}\frac{\mathrm{S}-1}{\sqrt{VAR\left(\mathrm{S}\right)}}& ; if \mathrm{S}>0\\ 0& ; if \mathrm{S}=0\\ \frac{\mathrm{S}+1}{\sqrt{VAR\left(\mathrm{S}\right)}}& ; if \mathrm{S}<0\end{array}\right.$$

(5)

The Mann–Kendall analysis was performed on the normalized anomalies of the LST [58,59] given by Equation (1). This procedure generated two rasters: the “p-value”, which indicates whether the trend is significant, and the “score”, which indicates the sign of the trend.

2.5. Significance Threshold Calculation

The “p-value” file was used to generate a binary type of raster (or matrix), where values “1” correspond to pixels that have a significant trend (p < 0.05) and the values “0” to pixels without a significant trend.

2.6. Regression’s Slope Calculation

A routine was created in R 4.1 [60], which calculates linear regression over the three time series of LST. A matrix was obtained with the linear regression’s slope values, corresponding to the trend values of annual increases or decreases in temperatures (in degrees Celsius), for each month of the considered summer period. To obtain monthly slope values, only pixels with a significant trend (MK) were considered. To accomplish this, the binary matrix of significance thresholds was multiplied by the matrix with the trend values.

2.7. Rasterization of Land Uses and Geomorphology Maps

To evaluate the relationship between the significant trend values with the types of land uses [38] and with the geomorphological units [48] (being both vector files), a numerical code was assigned to the different typologies in a new base field of data. For land uses: (1) Urban and industrial areas, (2) Agricultural land, (3) Meadows and bushes, (4) Forests, (5) Wetlands, (6) Areas devoid of vegetation, (7) Eternal snow and glaciers, and (8) Water bodies. Similarly, for geomorphological units: (1) Active volcanic mountain range, (2) Semi-arid transitional basins, (3) Marine or fluviomarine plain, (4) Andean foothills, (5) Fluvial or alluvial sedimentation plains, (6) Central plain of the Santiago basin (7) oastal range (8) Transverse mountain ranges, and (9) Andean mountains. Then, polygons were converted from vector format to raster format, transferring the new codes as pixel values, matching the output matrix of the raster for the MK analysis. From this result, 17 binary matrices were generated, one for each type of land use and geomorphology.

2.8. Obtaining Trend Values for Each Use and Geomorphological Type

Monthly significant trend matrices were multiplied by each of the binary matrices of land uses and geomorphological typologies. The product was a matrix of mean values with positive trends, mean values of negative trends, and the surface of each category.

3. Results

When analyzing the MODIS LST 17-year series, it was possible to know the spatial and temporal distributions of the trends that meet the requirement of the MK test to have statistical significance. Due to the variability of the data series, the territorial coverage of places in the study area that meet this requirement represents 12% of the total area for December (3730 km²), 28% coverage for January (8773 km²), and 17.5% coverage for February (5500 km²) (Figure 3).

Figure 4, Figure 5 and Figure 6 show the spatial variation in the slopes of the linear regression line for the significant pixels according to the Mann–Kendall test. From these results, the areas with positive trends (in red in the figures) were separated from those with negative trends (blue).

Figure 7 and Figure 8 illustrate the intersection of land use and geomorphology layers, respectively, with the layer of significant pixels according to the MK test. This provides the spatial variation in the slope for each land use and geomorphology analyzed.

Table 1. Average values of the positive trends in °C per year and total surface area and MK surface (Mann–Kendall) in Km² by type of current land use of the LST in the summer months for the areas with a significant trend. Years 2000–2017. Source: Own elaboration.

Land Use	Total Area (km2)	December		January		February
		Trend (°C year−1)	Area (km2)	Trend (°C year−1)	Area (km2)	Trend (°C year−1)	Area (km2)
Urban and industrial areas	1911	0.176	368	0.178	292	0.158	650
Agricultural land	3899	0.196	710	0.192	686	0.170	1326
Meadows and bushes	9475	0.234	878	0.193	2259	0.188	1348
Forests	9281	0.223	767	0.181	1237	0.186	1842
Wetlands	221	0.320	19	0.222	113	0.202	10
Areas devoid of vegetation	4922	0.339	289	0.246	2759	0.198	106
Eternal snow and glaciers	1530	0.382	113	0.298	1018	0.231	23
Water bodies	128	0.166	20	0.175	33	0.157	35
Total	31,367		3163		8398		5339

and

Table 2. Average values of the positive trends in °C per year and surface area in km² by geomorphological type of the LST in the summer months for the areas with a significant trend. Years 2000–2017. Source: Own elaboration.

Geomorphology	December		January		February
	Trend (°C year−1)	Area (km2)	Trend (°C year−1)	Area (km2)	Trend (°C year−1)	Area (km2)
Andean mountains	0.361	492	0.237	5240	0.183	117
Coastal range	0.199	213	0.195	344	0.187	1094
Active volcanic mountain ranges	0.251	7	0.291	151	0.297	1
Transverse mountain ranges	0.184	535	0.158	732	0.166	723
Semi-arid transitional basins	0.283	438	0.263	430	0.247	400
Central plain of the Santiago basin	0.192	957	0.196	698	0.169	1588
Fluvial or alluvial sedimentation plains	0.161	232	0.160	298	0.163	831
Marine or fluviomarine plain	0.192	91	0.170	162	0.186	277
Andean foothills	0.241	196	0.183	334	0.182	309
Total area		3162		8389		5339

present the average slope value for each land use and geomorphology, respectively.

At a macro level, LST trends are positive (i.e., there is a sustained increase in the LST variable) for all land uses studied. Likewise, it can be seen that the month of December generally presents higher LST values for all coverages except for wetlands in February.

4. Discussion

The factors that explain these changes in LST can be natural or anthropogenic: (a) Changes in land use, as outlined by Xiao and Weng [61], where the conversion of natural areas into urbanized or agricultural areas can cause an increase in surface temperature due to an alteration in soil reflectivity and heat retention capacity; (b) In densely populated areas where increased construction of buildings, roads, and other infrastructure can generate a “heat island effect”, where temperatures are significantly higher than in surrounding, less-urbanized areas [61]; (c) Climate change may contribute to rising surface temperatures through changes in weather patterns, such as an increase in average temperatures and increased frequency and intensity of heat waves [62]; (d) Alterations in vegetation cover such as deforestation and ecosystem degradation can reduce the capacity of natural areas to regulate temperature through processes such as transpiration and evapotranspiration, which can lead to an increase in surface temperature [63]; (e) Local modifications of the microclimate due to changes in the distribution of vegetation, soil cover, or the geometry of the terrain at the local level, which can influence the amount of solar radiation absorbed and air circulation, affecting the temperature of the surface [64].

It is worth mentioning that the results of the analysis of LST trends were compared with the different types of current land uses. We agreed that studying changes over time in land uses in synchrony with changes in LST trends is the subject of another study.

The results indicate that a sector to the north of the city of Santiago stands out with significant changes in the three months analyzed, which means that it experiences a sustained increase in LST measurements throughout the entire summer period. In this area, there is a degraded sclerophyllous forest and degraded shrubland. This could result in a greater penetration of solar radiation to the ground, thereby defining the detected temperature increases. This sector has a warm steppe climate [65] and corresponds geomorphologically to transverse mountain ranges, semi-arid transitional basins, and a sector of the central valley within the Santiago basin [48]. The calculated trends show an increase of 0.26 °C per year in December and 0.22 °C per year in both January and February.

The Andean mountain range (Figure 8) shows significant trends only in the January series. Concurrently, there has been an observed decrease in snow cover due to an increase in altitude of the zero-degree isotherm [66]. This phenomenon defines an increase in areas of exposed rocks to solar radiation. These two factors could explain the greater presence of significant positive temperature trends in this Andean zone. On the other hand, the results evidence a larger area with positive trends in the three summer months considered (Figure 4, Figure 5 and Figure 6). This predominance of positive trends in LST coincides with the general trend detected in air temperature by DMC [44] in the same area.

By crossing the information of land use and areas with significant tendency, Figure 7 and Table 1 were obtained, where positive trends predominate in the area during the months of December and February for Meadows and bushes and Forests and agricultural land uses, while for January, the largest area with positive trends corresponds to areas with no vegetation. Analyzing the net value of km² per land use and per month, it is concluded that, due to its greater coverage, January is the most representative month to evaluate the trends of Meadows and bushes (0.19 °C year⁻¹); Areas with no vegetation (0.24 °C year⁻¹); Eternal snow and glaciers (0.29 °C year⁻¹); and Wetlands (0.22 °C year⁻¹). February, on the other hand, is the best month to study the trends of Urban and industrial areas (0.15 °C year⁻¹); Agricultural land (0.17 °C year⁻¹); Forests (0.18 °C year⁻¹); and Bodies of water (0.15 °C year⁻¹). Considering only trend values, there was 113.2 km² (7% of the total) with the highest average LST increase, recorded in Eternal snow and glaciers during December (0.38 °C year⁻¹). However, 66% of the total area of Eternal snow and glaciers recorded increases of 0.29 °C year⁻¹. These increases can strongly affect the supply of summer water in the Santiago basin.

The largest area with a negative trend in December was on the land use “Forest”, but it only represents 2.3% of the total forested area. During the months of January and February, “Agricultural land” predominates. The land use “Urban and industrial areas” represents an area of 1911 km², and, according to the MK test, only 19.2% of the total area during December, 15.2% in January, and 34% in February shows significant positive trends.

Urban areas with a negative trend represent 1.4, 0.7, and 0.3% for December, January, and February, respectively. From this analysis, the MK test leaves out the city of Valparaíso by 90%, due to the strong presence of morning clouds, which characterizes the mild Mediterranean coastal climate. The average LST increments for this type of land use, recorded for the December, January, and February series, were 0.176, 0.178, and 0.158 °C year⁻¹, respectively.

Separating the city of Santiago from the rest of the urban and industrial areas, the territorial coverage with significant trends was irregular (Figure 8), which does not allow for an LST trend analysis for the city in the morning schedule of the MODIS sensor (11:00 a.m.). Sarricolea and Martin-Vide [23] used the LST MODIS measured in the evening schedule (21:00 p.m.) to study Santiago’s urban heat island effect. The areas that do have significant trends during the three months analyzed were the industrial areas north of the city and, promptly during February, areas appear in the southern sector of Santiago. Average LST increases in these sectors for December, January, and February, were 0.177, 0.183 and 0.157 °C year⁻¹, respectively. Those values are very similar to those documented for the rest of the urban and industrial areas evaluated.

After crossing information between geomorphological types and the areas with significant LST tendencies, the results can be seen in Figure 9. During the three summer months analyzed, it is possible to say that the largest areas with positive trends occur in the “Andean mountain range” unit (5240 km²) during January, with trends of 0.23 °C year⁻¹, but there was another 491 km² (0.36 °C year⁻¹). The largest areas with negative trends were manifested in the unit “Plain river or alluvial sedimentation”, during the month of December (−0.16 °C year⁻¹). When analyzing the relationship between the surface amount with the values with significant trends, the most representative summer month to study LST trends behavior corresponds to the geomorphological units in January: Andean Mountains, Active volcanic mountain range, Transversal cords, and Andean foothill. However, February is the month that allows for a better analysis of the trends of the geomorphological types Coastal range, Central plain of the Santiago basin, Plains of fluvial or alluvial sedimentation, and Marine plain or fluviomarine. In the case of semi-arid transitional basins, the series of the three months analyzed have equivalent surfaces.

Veblen et al. [67] suggest that temperature increases can lead to diverse ecological impacts depending on the geography and type of biome. By correlating trends in LST in significant areas with different land uses, it has been possible to analyze the behavior of trends in a specific manner, detecting increases above the average within certain types of coverage such as “Eternal snow and glaciers”, trends that can alter the rate of glacier melting [68]. In the relationship between LST and geomorphological types, the “Semi-arid transitional basins” show the highest increases in LST (0.26 °C per year on average during the summer period).

Finally, an improved spatial resolution will allow these results to be adjusted to the local conditions of specific land uses, which is why it is necessary to adapt the current sensor offering. An example of this is applying new methodologies such as those developed by Yi Yu et al. [69], who evaluated the results of fusing different remote sensors to improve the special resolution of these analyses. This is especially important because the MODIS sensor, which is the most commonly used (52% of the investigations), will be discontinued [3], which will force it to be replaced by LST measurements from other sensors, to complete the significant 30-year time series.

5. Conclusions

At a macro level, the results from this investigation show a clear increase in LST within the study area, particularly in the Andean mountain range in January. This part of the territory mainly consists of areas devoid of vegetation and eternal snow and glaciers. It is worth noting that areas covered with vegetation such as forests, meadows, and bushes also show increasing trends in LST but in a smaller surface area. In other words, these results indicate a sustained increase in temperatures within the study area, impacting the vegetation and even more so the Andean mountain range, where solid water reservoirs of the city of Santiago are located.

These results provide a broader perspective to the organizations in charge of planning land uses and water resources of the analyzed basins, which supply almost nine million inhabitants, representing nearly half of the country’s total population. Likewise, they will allow for the improvement and evaluation of the application of mitigation measures such as urban design that allow us to face urban heat islands.

This study was based on a regional scale, but with images of better spatial and temporal resolutions for LST estimation, it will be possible to analyze variations in trends at a more local scale.

Finally, the methodology used is especially useful for monitoring large surface areas, so it is recommended to replicate this analysis across the entire continental territory of Chile.