Evaluating the Spatiotemporal Dynamics of Land Surface Temperature in Relation to the Land Use/Land Cover changes in Nag-Hammadi District, Egypt, using Remote Sensing and GIS

Mostafa K. Mosleh

Department of Geography, Faculty of Arts, South Valley University, Qena, 83523, Egypt

E-mail: dr.mostafak.mosleh@gmail.com

Abstract

In this study, three multi-temporal remotely sensed data acquired from Landsat-5 Thematic Mapper (TM) and Landsat -8 Operational Land Imager/Thermal Infrared Sensor (OLI/TIRS) in 1990, 2005, and 2020 were used. The maximum likelihood classifier (MLC) was opted to classify land use and land cover (LULC). Land surface temperature (LST) and LULC spectral indices i.e., Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), Normalized Difference Latent Heat Index (NDLI) and Bare Soil Index (BSI) have been computed and their relationships were examined. The overall accuracy of LULC was more than 93%. The analyses showed a notable transformation in LULC over the study period. For instance, built-up areas increased 103.7% with a rate of 45.5 ha/year and agriculture land increased by 28.9% with a rate of 186.4 ha/year. Whereas, bare soil was sharply decreased by 36.4% at a rate of 227.7ha/year. The minimum and maximum LST values increased by 2.9°C and 4.9°C, respectively, from 1990 to 2020. Furthermore, LST has a negative relationship with NDVI and NDLI (NDVI: 1990: r² = 0.62; 2005: r² = 0.62; 2020: r² = 0.65. NDLI: 1990: r² = 0.79; 2005: r² = 0.78; 2020: r² = 0.61) and a positive relationship with NDBI and BSI (NDBI: 1990: r² = 0.68; 2005: r² = 0.73; 2020: r² = 0.44. BSI: 1990: r² = 0.77; 2005: r² = 0.78; 2020: r² = 0.53). These results provided useful information about LULC changes and its impact on LST, which are necessary for experts and land-use planners to formulate sustainable LST mitigation policies, create an environmental comfort in Nag-Hammadi district, and other geographical locations with similar conditions.

Keywords: Geospatial Mapping, GIS Analysis, LST, LULC, Spectral Indices

1. Introduction

The land use and land cover (LULC) rapidly changes due to various natural and anthropogenic activities. This promotes serious worldwide environmental and human concerns regarding energy balance, near-surface temperature over the cities and its surroundings, human comfort, physical health, and quality of life [1] [2] and [3]. In this context, land surface temperature (LST) is sensitive to LULC changes which, in turn, influences its spatial distribution and, as such, needs accurate monitoring [4] and [5]. Traditional ground field and in-situ methods for monitoring LULC and LST are costly, time-consuming, and involve a tedious process making the spatial analysis of LULC and LST difficult [6] and [7].

In this regard, geospatial technologies i.e., remote sensing and geographic information system (GIS) are effective for detecting the changes of LULC and LST due to several reasons such as the low cost, availability, and large spatial coverage [8] and [9]. Several studies explored the impact of LULC changes on LST using GIS techniques and various satellite data over different geographical locations. Zhang et al., [10] used IKONOS data to obtain urban vegetation types and Landsat 7 Enhanced Thematic Mapper Plus (ETM+) thermal bands to retrieve LST over Nanjing City, China. They found that higher LST values are negatively correlated with the vegetation cover (r= - 0.59). The results also revealed greater cooling effect of large-vegetated areas over smaller ones. Mallick [11] applied ASTER image at 90m for mapping LULC and LST over semi-arid region of Abha, Kingdom of Saudi Arabia and found that the spatial distribution of LST was affected by LULC and negatively correlated with Normalized Difference Vegetation Index (NDVI) with r² value of 0.62. El-Zeiny and Effat [12] used Landsat Thematic Mapper(TM), ETM and Landsat 8 Operational Land Imager/Thermal Infrared Sensor (OLI/TIRS) images over El-Fayoum governorate, Egypt, and found that desert bare lands exhibited mean LST of ˃42℃ followed by urban areas, vegetation, and water bodies. Dhar et al., [13] utilized Landsat-5 TM, Landsat-8 OLI/TIRS, and Sentinel-2A for mapping LULC and estimating the LST over Rajarhat Block, West Bengal, India. The study showed an increase in LST to the tune of <1°C. Furthermore, the results revealed negative correlation between NDVI and LST, whereas positive correlation occurred between Normalized Difference Built-up Index (NDBI) and LST of urban areas. Aik et al., [14] utilized Landsat 7 ETM+, Landsat 8 OLI/TIRS, and Moderate Resolution Imaging Spectroradiometer (MODIS) images over Cameron Highlands district, Malaysia. The results demonstrated that the measured LST showed a rise of about 2-3°C with an extreme of 5°C on average.

The current study utilized Landsat satellite images to evaluate the LULC changes and their impact on LST in Nag-Hammadi district over the past 30 years (1990 and 2020) as this region is the economic, commercial, agriculture and industrial hub in Upper Egypt. The specific objectives of this study are to (i) analyze the spatiotemporal changes of LULC over the last three decades from 1990 to 2020; (ii) analyze the spatiotemporal variation of LST, and (iii) evaluate the relationship between LST and selected LULC spectral indices.

2. Study Area

Geographically, Nag-Hammadi district is located in Upper Egypt between 25°26ʹ 18ʺ to 26°08ʹ58ʺ N latitude and 32°09ʹ47ʺ to 32°27ʹ03ʺ E longitude (Figure 1). The district belongs to Qena governorate with a total area of 40703.58 hectare (ha). The population increased from 373504 in 1996 to 451891 in 2006 and reached 578237 in 2017 [15]. Climatologically, it is characterized by a desert climate i.e., hot, and dry summer and warm winter.

Figure 1: Geographical location of the study area (Nag-Hammadi district)

In the summer, the average maximum temperature is 44°C, while it is 24°C in the winter. The maximum relative humidity is 53% in the winter and 29% in the summer. It rarely rains [16]. Agriculture is the prime source of livelihood for the population, and it’s cultivated with various fruits, vegetables, and field crops depending on the irrigation from the Nile River and underground sources. In addition, the district is home to the largest factory of aluminium production in the Middle East, and the largest factory for sugar production in Egypt.

3. Materials and Methods

Figure 2 shows the conceptual diagram of the major procedures adopted in this study.

3.1 Data Collection and Pre-Processing

Remotely sensed surface reflectance and surface temperature imageries from Landsat 5 and Landsat 8 were used for the LULC classification and the LST estimation of Nag-Hammadi district. Three satellite images from the Landsat series i.e., Landsat-5 for 1990, 2005, and Landsat 8 OLI/TIRS for 2020 were freely acquired from the United States Geological Survey (USGS) (https://earthexplorer.usgs.gov). These images were selected in June and July to avoid the seasonal variation of climatic conditions with cloud-free status peering in mind that the LST and its relation to LULC is maximize in the summer. Table 1 provides a descriptive summary of the Landsat imageries that have been used in this study. Only band 10 from Landsat 8 was used for retrieving LST because band 11 is affected by stray light and ghosting issues and faced calibration uncertainty [14]. All the used spectral bands were stacked and clipped to the boundary of the study area. Furthermore, other auxiliary datasets such as topographic maps at scale (1: 50,000), land use maps and high spatial resolution images from Google Earth were used to support the analysis. In this context, several band combinations of the Landsat images were tested to generate different composite effects and increase the interpretation of LULC. A combination of NIR, red, and green spectral bands were used in displaying the RGB color composite image for Landsat 5 TM and Landsat 8. Then, four LULC classes were visually identified namely: built-up area, agricultural land, water bodies, and bare soil and verified from the field investigation.

Table 1: Description of the satellite imagery used for LST and LULC in this study

* Resampled to 30 m pixels by USGS from the original 120 and 100m resolution.

Figure 2: Conceptual diagram of the adopted methodology, (a) generation of LULC maps and derived-based indices, (b) derivation of LST and its relation with LULC and spectral indices

Figure 3: Geographical distributions of spatio-temporal LULC changes in Naj Hammadi district for the years (a) 1990, (b) 2005, and (c) 2020

Table 2: Area coverage for each land cover class (Hectares) and percentage of LULC classes

4. Results and Discussion

4.1 LULC changes from 1990 to 2020

The spatiotemporal analysis of LULC changes is presented inFigure 3(a) to (c) and Table 2. Results showed that the agricultural land was dominant during the period of interest and experienced

gradual increase from 47.4% to 61.1% of the study area. The built-up area increased from 3.2% to 6.6%, while the bare soil and water bodies were declined from 46.2 to 29.4% and from 3.2 to 2.9% from 1990 to 2020, respectively.

Table 3 shows the MC of the LULC classes between the years. During period one, between 1990 - 2005, the study shows a noticeable decrease of 2558.11 ha with a change difference of -6.3% for the bare soil class. Water bodies observed an MC of -53.01 ha with a change difference of -0.1%. Whereas, the agricultural land witnessed an increase in MC of +1790.37 ha, with a change difference of 4.4%. The built-up area demonstrated an increase in MC of +820.75 ha with differences change of 2%. In the second period, between 2005 - 2020, bare soil class had witnessed a decline of -4273.23 ha with a change difference of -10.5%. Water bodies observed an MC of -70.92 ha with a change difference of -0.2%. On the contrary, agricultural land witnessed an increase of +3800.34 ha with a change difference of + 9.3%. Furthermore, the built-up area witnessed an increase of +543.81 ha with a change difference of + 1.4%. Additionally, the land cover distribution demonstrated various change between 1990 - 2020. The bare soil and water bodies witnessed a loss of -6831.34 ha (-16.8%) and -123.93 ha (-0.3%), respectively. Land coverage by agricultural land and built-up area witnessed the highest increase with +5590.71 ha (+13.7%) and +1364.56 ha (+3.4%), respectively. The analysis of the MC and PC remarkably show that the increase of built-up areas was at expense of bare soil and agricultural lands. This increase would be related to the population growth, which has contributed to massive human activities and development. Moreover, it can be observed that agricultural lands were increased in line with the decrease of bare soil and maintain its domination over the entire study period. Such increasing trend in agriculture lands exists mainly towards the southern side of the study area.

This would be due to the continuous investment in agricultural activities (i.e., new land reclamation) over the past thirty years in the study area which significantly contributed to the alteration of LULC types. Figure 4 shows the annual rate of change (ARC) of the LULC in three different periods (1990 – 2005; 2005 – 2020 and 1990 – 2020). In the first period, a significant annual increase in the built-up area (54.7 ha/year) was mostly occurred at the expense of agricultural and bare soil (-170.5). Slight annual increase and decrease were observed in the agricultural land and water bodies; 119.4 ha/year and -3.5 ha/year, respectively. During the second period the annual increase in built-up areas with rate of 36.3 ha/year, and agricultural land increased with a rate of 253.4 ha/year, which occurred on expense of the bare soil (-284.9 ha/year). During the entire study period (i.e., 1990-2020) agricultural land was the largest land cover in the region with annual increase of 186.4 ha/year. The built-up area witnessed an increase of 45.5 ha/year. While, both water bodies and bare soil decreased by annual rate of -4.1 and 227.7 ha/year, respectively. It was observed that this increase was synchronized with a decrease in the bare soil during the whole period of study which might be a reflection of the governmental developmental projects (i.e., urban development and fringe desert reclamation).

Table 3: The magnitude of the change (MC) in land cover classes between years

Figure 4: Graph showing annual rate of change for each LULC classes (ha/year) during the period of study

4.3 Land Surface Temperature Change (LSTC)

Figure 5 shows the spatial distributions of LST retrieved from satellite images for the three study periods (1990, 2005 and 2020). It shows the variation of the minimum and maximum LST across Nag-Hammadi district which ranges between 21.94°C and 49.57°C in 1990. In 2005, the LST values varied between 24.12°C and 45.57°C, while in 2020, the LST values were within the range of 24.84°C to 54.56 which revealed an increase in the minimum and maximum LST over the last three decades by approximately of 2.9°C and 4.98°C, respectively. This increase in the LST values could be attributed to the increasing built-up areas and other impervious surface in the study area from 1990 to 2020.

4.4 Variation of LST for Different LULC Types

Figure 6 summarizes the mean values of LST (°C) of each LULC type during the three years of interest 1990, 2005 and 2020. Zonal statistics analysis was performed to summarize the relationship between LULC and LST. The comparison between the LST and LULC map indicated low mean values of about 23°C, 25°C, and 26 °C in water bodies in 1990, 2005, and 2020, respectively.

While high mean values of 38°C, 40°C, and 43°C in built-up area and 39°C, 42°C, and 45°C in bare soil in 1990, 2005, and 2020, respectively. This indicates that urban areas contribute on rising LST on the account of other LULC as the urban areas are mainly constructed from impervious surfaces such as stone, metal, and concrete which exhibit low evaporation. Meanwhile, the bare soils exhibit higher LST values according to the diurnal cycle variation of LST.

The agricultural lands show moderate LST values in all years due to the evapotranspiration and vegetation density. To deepen the understanding of the impact of LULC changes on LST, three arbitrary cross sections were created across the study area to represent the LULC types wise LST. These included two cross-sections (i.e., A-B, C-D) from north to south and E-F from west to east (Figure 7).

Figure 7: LST versus LULC profiles for the spatial profiles shown in Figure 5 for the years 1990, 2005, and 2020

It is seen from the cross profiles that bare soil exhibits highest LST with ~45°C followed by built-up area ~ 40 °C. In three different cross-sections, the water bodies and agricultural land showed the lower LST of ~ 22ºC and ~ 34ºC, respectively, in 1990. The LST values for bare soil, built-up area, water bodies, and agricultural land were ~ 40°C, ~ 42°C, ~ 23ºC and ~ 34ºC, respectively in 2005. In 2020, these values were ~ 46°C, ~ 44°C, ~ 25ºC and ~ 37ºC, respectively. From the analysis, it can be inferred that agricultural lands shows low LST in all years due to the heat reduction within vegetation cover due to its density and transpiration. These outcomes conform to the findings of different studies [30] and [31].

4.5 Relationship between LST with LULC Spectral Indices

Figure 8 shows the spatial distribution of NDVI values derived using Landsat images of 1990, 2005 and 2020. The results showed that the NDVI values ranged between -0.50 and 0.72 in 1990, while it was -0.41 and 0.69 in 2005, and -0.19 and 0.57 in 2020.These results demonstrate the highest NDVI values in the east and west banks along the Nile River within Nag-Hammadi district, which mainly is covered by densely vegetated areas (i.e., most productive agricultural land) leading to lower LST. The distribution of NDLI values of Nag-Hammadi district in 1990, 2005, and 2020 is presented in Figure 9. The results demonstrated that the NDLI values ranged between -0.09 to 0.17 in 1990. This was -0.12 and 0.09 in 2005 as well as -0.08 and 0.08 in 2020. The NDLI values were low in built-up areas and bare soils and high in the water bodies and vegetation cover which is consistent with other studies such as [25].

The NDBI(Figure 10) values ranged between -0.51 to 0.36 in 1990. In 2005, these values were between -0.49 and 0.44 as well as -0.39 and 0.28 in 2020. The NDBI values were very high in the built-up areas and bare soils as they are exposed to the incoming solar radiation. Finally, the spatial distribution of BSI values in 1990, 2005, and 2020 is presented in Figure 11. The result demonstrates that the BSI values ranged between -0.54 and 0.13 in 1990, -0.46 and 0.14 in 2005, and -0.30 and 0.19 in 2020. Among them, the bare soils have the highest BSI value. Figures 12 and 13 illustrate the results of the relationship between LST and LULC spectral indices. The linear regression analysis demonstrates that higher LST were corresponding to areas of low vegetation cover and vice-versa. In this study, the statistical analysis and scatter plots clearly shows that the LST is negatively correlated to NDVI; having a coefficient of determination R²value of 0.62 in 1990, 0.62 in 2005, and 0.65 in 2020. The result of the correlation and regression analysis discloses a positive correlation between NDBI-derived built-up areas and LST with R² values of 0.68 in 1990, 0.73 in 2005, and 0.44 in 2020 (Figure 12).

Figure 8 : Spatial distribution of NDVI in Naj-Hammadi district in (a) 1991; (b) 2005; and (c) 2020

Figure 9: Spatial distribution of NDLI in Naj-Hammadi district in (a) 1990; (b) 2005; and (c) 2020

Figure 10 : Spatial distribution of NDBI in Naj-Hammadi district in (a) 1990; (b) 2005; and (c) 2020

Figure 11: Spatial distribution of BSI in Naj-Hammadi district in (a) 1990; (b) 2005; and (c) 2020

Figure 12 : Graph showing scatter plot for (a) LST and NDVI; (b) LST and NDBI

Figure 13: Graph showing scatter plot for (a) LST and NDLI; (b) LST and BSI

Figure 14: (a) Spatial distribution of LST in Naj-Hammadi district in 2020, (b) true colour image of the Aluminium factory area, and (c) spatial profile of LST detected for the Aluminium factory area