Description of the study area

The Abaminus watershed is part of the upper Blue Nile Basin in the northwest highlands of Ethiopia. The watershed lies between 10°38′6.161″N and 11°8′48.328″N latitude and 38°5′12.436″E and 38°31′49.06″E longitude.

The total area of the study watershed covers about 172,142.2 ha, and it is characterized by diverse topographic conditions. Gently sloping to slightly undulating plains, rugged and dissected hilly mountains and relatively deep river gorges characterize the entire relief. The local climate is dominantly humid sub-tropical. The watershed in general falls within three agro-climatic zones (cool-moist, tepid-moist, and warm highlands) that are equivalent to the Ethiopian traditional agro-ecological zones of Dega, Woina-Dega, and Kola, respectively with the elevation ranges from 1,145 to 3,534 m.a.s.l. (Figure 1). Rainfall varies spatially from 1,326.5 mm over the cool-moist agroecology to 917.9 mm in the warm highlands agroecology. More than three-fourths of the total rainfall occurs during the summer season (from June to September).

Figure 1.

Map of the study watershed.

In all agro-ecological zones of the watershed, subsistence primitive crop-livestock mixed-farming agriculture is the predominant livelihood activity. Conventional tilling through a hand press with ox-pulling traditional technique is the most common one of all crop productions. In the Dega agro-ecological zone (⩾2,800 masl cool-moist highlands) barley (Hordeum vulgare) and potato (Solanum tuberosum) are chiefly produced. Degradation and soil acidity are most likely serious problems over the steeper hill slopes in this agroecology. In the Woina-Dega agro-ecological zone, tef (Eragrostis tef), barley (Hordeum vulgare), wheat (Triticum vulgare), and horse beans (Vicia faba) are common crops. Tef (Eragrostis tef) and Wheat (Triticum vulgare) are predominantly grown crops among the others. In the Kola agro-ecological zone, sorghum is largely produced. Crop production in the agro-ecological zone is suffering perhaps due to erratic rainfall and severe ecological degradation.

Communal and privately owned lands and livestock are the most important livelihood assets of the local community in all of the agroecological zones. The common-pool resources (forests, shrublands, wetlands, and grasslands) are usually managed by the local community and administration units. The hilly and steep slope communal lands of the watershed are highly threatened by vegetation degradation and dissected by gullies due to uncontrolled overgrazing.

Cattle, sheep, goat, donkey, and horse are reared under the cut-and-carry and open-grazing systems. Local labor and firewood/charcoal/sales are alternative means of farmers’ livelihoods to meet household food demands, particularly in kola agroecology.

In the study watershed, five major LULC classifications, such as forests, shrublands, grasslands, no vegetation, and wetlands are identified (Table 1). Privately owned individual farmer-managed settlements and croplands are the largest land-use systems in the area.

Table 1.

Descriptions of Land Use/Cover Classes.

Data sources and collection techniques

Remote sensing data are important primarily to understand the degree of interaction between the social and ecological systems. Space-based observations of Landsat TM (Thematic Mapper) image of 1987, 1999, and 2010, and Landsat-8 Operational Land Imager (OLI) image of 2018 were acquired from the US Geological Survey (USGS;  http://glovis.usgs.gov/). Clear-sky and similarity of the ambient temperature were considered for the selection of 4-year satellite imagery data. For each year, cloud-free from January to February, and one frame with Path 169 and Row-052 satellite imagery data were acquired. The Landsat TM and OLI are usually capable of mapping vegetation, moisture, water bodies, and LST dynamics analysis at the watershed level (Bai et al., 2008). Based on the results of LUCCs detection, three Kebeles (small administrative units) were purposely selected across three traditional agro-ecologies. The high rate of conversion from vegetation and wetland into no-vegetation was the main criterion to select Geses from Kola, Yebuchir-Yewoya from Woina-Dega, and Dequat-Goshera from Dega agroecology. From the selected Kebeles, about 164 household heads were selected using a systematic sampling technique. A household survey was conducted with the selected household heads using a semi-structured questionnaire.

Calculation of spectral indices

When the land surface reflectance is measured, an atmospheric correction should perform to correct for atmospheric effects on satellite scenes (Zhang et al., 2010).

The spectral indices’ raster of six bands of Landsat-5 and 8 with a 30 m spatial resolution was used to classify the land use/covers and to quantify LST (Table 2). Spectral indices are the combinations of spectral reflectance from two or more wavelengths that indicate the domination of certain characteristics of land cover (Gašparović et al., 2019). Normalized difference vegetation index (NDVI), modified normalized difference water index (MNDWI), normalized difference soil index (NDSI), and normalized difference bareness index (NDBaI) were applied to develop spectral indices.

Table 2.

Bands Micrometers and Resolution of Thematic Mapper (Landsat-5) and Operational Land Imager (Landsat-8).

The main principle of detecting vegetation using NDVI is vegetation highly reflective in the near-infrared and high absorptive vegetation pigments (chlorophyll) in the visible red. The contrast between these channels can be used as an indicator of the status of the vegetation.

(1)

Where Near-infrared represents band 4 in Landsat-5 and band-5 in Landsat-8. Red represents band-3 for Landsat-5 and band-4 for Landsat-8.

The NDVI is a biophysical parameter that relates to photosynthetic vegetation to offer valuable information about the dynamics of vegetation covers over time (Xie et al., 2008). From other various available vegetation indices, the NDVI is widely employed since it normalizes the effect of changes in illumination conditions and surface topography (Reddy & JangaReddy, 2013). Genesis et al. (2015) recommend that a linear model with NDVI time-series analysis, perhaps use to best fit the cyclic vegetation difference into a line. Values of NDVI ranges from −1 to 1, where vegetated areas generally result in high values because of their relatively high near-infrared reflectance and low visible reflectances. Determination of the threshold value is the main problem of the automatic extraction of individual land cover classification from a single index raster. It is a difficult and lengthy process (Xian et al., 2009). Value ranges of the index for different land cover/use types can be found in certain ranges and are specific to each satellite imagery scene (Lee et al., 2011). The K-means unsupervised classification method can be used to extract a specific single index raster (Li et. al, 2017). It was used as a popular algorithm for classifying NDVI data into k-clusters, such as grasslands, shrublands, and forests. Vegetation class was classified into sparse for grasslands, moderate for shrublands, and dense to forests.

Modified normalized difference water index (MNDWI) is one of the methods of remote sensing to measure surface water characteristics where field investigations are difficult to carry out (Soti et al., 2009). It is mostly computed by using (Xu, 2006):

(2)

The indices of MNDWI range between 1 and −1. This requires reclassification to classify surface water from other land use land covers. Indices greater than zero, commonly represent standard thresholds to detect the distribution and extent of surface water (Xu, 2006). The MNDWI is one the most widely used water indices for surface water mapping as the built-up land, soil, and vegetation all have negative values and thus is notably suppressed and even removed (Campos et al., 2012; Duan & Bastiaanssen, 2013; Hadeel et al., 2011; Poulin et al., 2010).

The combination of the normalized difference bareness index (NDBaI) with the normalized difference soil index (NDSI) is the proper method to derive the accuracy class of no vegetation (Zhao & Chen, 2005).

(3)

Where, MIR is band-5 for the reflectance of Landsat-5 and band-7 for Landsat-8 and the Green represents band-2 for Landsat-5 and band-3 for Landsat-8 (Deng et al., 2015). This is attributable to only soil more reflective in band-5 of Landsat-5 and band-7 of Landsat-8 (Rogers & Kearney, 2004).

Normalized difference bareness index (NDBaI) was proposed to recognize different types of bare areas that formed due to physiographic and anthropogenic factors (Li et al., 2017; Zhao & Chen, 2005). The index is based on significant differences in the spectral signature in the near-infrared between bare soil and background. The NDBaI is calculated as follows (Zhao & Chen, 2005):

(4)

Where, SWIR1 is a short-wavelength infrared band, such as Landsat-5 band-5 and Landsat-8 band-6 and TIR is the thermal infrared band, such as band-6 for Landsat-5 and band-10 for Landsat-8. Based on the k-mean unsupervised classification, the class with the highest mean value represents bare land and the second class represents the built-up class (Li et al., 2017).

All spectral indices were calculated using the raster based on the respective index equation. The maximum likelihood supervised classification was used to compare the results K-means unsupervised method. It calculates the relative class membership likelihoods incorporating all training sets for each pixel in an image. All rasters with an extracted certain land cover were merged to generate land use and cover classes in one raster. According to the mean value of the classified indices, the automatic algorithm extracts the final land cover classes. The LUCCs were calculated using a land-use transition matrix. It indicates the amount of different land uses/covers that remain unchanged and quantifies the gains and losses during the study periods. The method is employed to describe the conversion size land-use types in three periods. The matrices of land-use transition were developed in the first period from 1987 to 1999, the second period from 1999 to 2010, and the third period from 2010 to 2018. Factors that drive the conversion of one LUC into another LUC were described in percentage (Figure 2).

Figure 2.

Workflow of LUCs analysis method.

LST extraction from thermal bands

Developing methodologies to measure land surface temperature (LST) from space is being increasingly recognized (Li et al., 2013). The use of satellites in the TIR is important to estimate reliable LST over large spatial and temporal scales as it is practically difficult to obtain such information from ground-based measurements (Ermida et al., 2017). LST is detected by infrared thermal sensors and provides finer spatial resolution only under clear-sky conditions (Liang et al., 2019). The atmospheric effects on top of atmosphere (TOA) spectral radiance are usually administered by atmospheric transmittance (T), atmospheric path radiance (Lu), and sky radiance (Ld) (Chatterjee et al., 2017). The spectral radiance model was employed to retrieve LST from Landsat-5 TM and Landsat-8 OLI (Hua & Ping, 2018). Thermal infrared data (band-6 from Landsat-5 and band-10 from Landsat-8) were used. From Landsat-8, thermal band-10 was used because thermal band-11 data are significantly more contaminated by thermal energy from outside the thermal field of view (stray light) than band 10 (USGS, 2016). The thermal infrared bands were used to estimate LSTs according to the reference values, calibration data, and empirical models widely used in land surface climate studies.

Conversion of digital number into spectral radiance

Based on the reference values in the sensor handbook, the digital number (DN) values convert into spectral radiance (Lʎ) (USGS, 2016). The conversion of DN values of the thermal bands into absolute radiance values is the standard method for retrieving LST from the raw Landsat dataset (Chander et al., 2009; USGS, 2016; Weng, 2009). The DN is converted into space reaching radiance or top-of-atmospheric (TOA) radiance that is directly measured by remote sensing instruments (Chander & Markham, 2003). The TOA reflectance is a unitless measurement that provides the ratio of radiation reflected by the incident solar radiation on a given surface. It compensates for different values of solar irradiance arising from spectral band differences.

In Landsat-5, TOA or spectral radiance (Lʎ) was computed as:

(5)

where; Lʎ is TOA radiance at the sensor’s aperture in watts/(m² × ster × µm); L_maxʎ is radiance maximum constant value from the metadata file of the band 6, L_minʎ is radiance minimum constant value from the metadata file of band 6, Q_calmax is the Quantize–Cal-Max constant values from the metadata file of band 6, and Q_calmin is Quantize–Cal-Min constant values from the metadata file of the band 6.

(6)

where; Lʎ represents the spectral radiance, ML represents the band-specific multiplicative rescaling factor, and AL represents the band-specific additive rescaling factor.

Conversion of spectral radiance (Lʎ) to brightness temperature (T)

After converting a digital number to reflection, the TOA radiance was then converted to the surface leaving radiance or brightness temperature by removing the effects of the atmosphere in the thermal region (Barsi et al., 2005). It is calculated with an assumption of unity emissivity and using pre-launch calibration constants. The following equation was used in the tool’s algorithm to convert reflectance to brightness temperature (T).

(7)

Where T is the temperature in Kelvin (K), K1 is the pre-launch calibration constant 1 in W/(m² srµm), K2 is the pre-launch calibration constant 2 in Kelvin (K) and Lʎ is spectral radiance. For obtaining the results in degree Celsius (°C), the radiant temperature is revised by adding the absolute zero (approximately −273.15°C), implying that C = T − 273.15.

The emissivity correction

The calculated radiant temperature was corrected for emissivity by using the NDVI values. After converting the brightness temperature values into degree Celsius (°C), the emissivity corrected LST was calculated as (Sobrino et al., 2004):

(8)

The last step of the LST or emissivity-corrected LST was retrieved (Grover & Singh, 2015) as

(9)

Where, LST is the land surface temperature in Celsius, T is at sensor brightness temperature (°C) in equation seven, ʎ is The wavelength emitted radiance for which peak response and the average of limiting wavelength of band-10 or band-6 (Table 2), Eʎ is the calculated emissivity value, and p = h × c/α = 1.438 × 10⁻² mk, where α is the Boltzmann constant (1.38 × 10⁻²³ JK⁻¹), h is Planck’s constant (6.626 × 10⁻³⁴ Js) and c is the velocity of light (3.0 × 108 ms⁻¹).

Statistical analysis

Descriptive statistics, such as percentage, range, and mean were used to analyze the changes of land use/covers and land surface temperature. Data generated from household surveys were presented using percentage values to identify the driving forces of land use/cover changes. The mean, range, and standard deviation of each LUCs and LST change were calculated using ArcGIS software. Simple linear regression was used to show the LST change over three decades. Data generated from household surveys were presented using percentage values to identify the driving forces of land use/cover changes.

Dynamics of land surface temperature (LST)

Figure 3 indicates land surface temperature (LST) distributions were varied from 4.99°C to 44.1°C in 1987, from 7.5°C to 45.93°C in 1999, from 9.5°C to 43.1°C in 2010, and from 10.1°C to 43.4°C in 2018. In all years, the high LST was distributed in the lower elevation of the study area (Figures 1 and 3). The influence of elevation also significantly affects LST (Peng et al., 2017). LST significantly raised with decreased altitude, implying that in the area with a relatively high elevation difference the LST noticeable vertical variation of decrease with altitude increase (Deng et al., 2018). This may be due to LST being affected by air temperature decrease with elevation and the surface vegetation in a high elevation area is relatively good. In high elevation areas, therefore, the solar radiation received by the ground surface mostly spread in the latent heat form and the LST is low.

Figure 3.

LST distribution of the study watershed.

The average LST was 27.9°C in 1987, 30.8°C in 1999, 30.83°C in 2010, and 31.65°C in 2018 (Figure 4). Results from curve fitting regression indicate that consistent increment of mean LST was explained by time at R² = .824 over three decades. The LST was increased by 2.88°C, 0.01°C, 0.79°C in the first period (from 1987 to 1999), in the second period, and the third period of the study in that order (Figure 5). In the first period, the highest LST change was observed when compared to other periods of the study. This may be due to the occurrence of metrological drought across the study area in 1999, which in turn contributes to the increment of LST on vegetation biomass. Over three decades, the mean LST was increased by 3.67°C. Based on the dataset from 1850 to 2015, the mean land surface air temperature has increased by 1.53°C, while global mean surface temperature has increased by 0.87°C (Jia et al., 2019). Inconsistent with the global warming trend, the mean LST was increased by 3.88°C in the last 20 years (Tan et al., 2020) and by 0.10°C year⁻¹ (Zhao et al., 2019) in China. The land surface temperature of most of the areas of all land cover features have significantly increased overtimes (Majumder et al., 2020; Zhao et al., 2019). Land degradation due to land use/cover changes contributes to about 23% of anthropogenic emissions of carbon dioxide (CO₂), methane (CH₄), and nitrous oxide (N₂O) (Jia et al., 2019).

Figure 4.

Mean LST in different years at watershed level.

Figure 5.

Mean LST difference over different years at watershed level.

Land use/covers (LUCs) and land surface temperature (LST) distribution

Of the total area, the forest cover was 8%, 7.2%, 4.5%, and 3.2% in 1987, 1999, 2010, and 2018, respectively (Figure 3). In the forest area, the average value of LST was 25.27°C over 31 years (Figure 4).

The forest cover declined by 8,220.5 ha over 31 years, implying that on an average 265.2 ha year⁻¹ forest degradation (Figure 6). Deforestation was carried out on 1,386.1, 4,660.8, and 2,173.6 ha lands in the first period (from 1987 to 1999), in the second period (from 1999 to 2010), and the third period (from 2010 to 2018), in that order (Figure 6). This implies that on average, 115.51, 423.71, and 271.7 ha year⁻¹ forest areas were cleared in the first, second, and third periods of the area, respectively. In the second period, deforestation was higher by 308.2 and 152.01 ha year⁻¹ compared to the first and third periods, respectively. Concerning deforestation and other global factors, the average LST was varied from 23.75°C in 1987 to 26.54°C in 2018, implying that it was increased by 2.79°C over 31 years in the forest area (Figure 7).

Figure 6.

Proportion of LUCs from 1987 to 2018 (in ha).

Figure 7.

LST (in °C) differently in response to LUCs in different years.

From the total, shrubland contained 16.7% in 1987, 17.1% in 1999, 14.2% in 2010, and 9.6% in 2018. From 1987 to 2018, 12141.4 ha shrublands were lost, with an average of 391.66 ha year⁻¹ (Figure 6). Over the three decades, the average LST was 28.2°C in the shrublands (Figure 7). Shrubland coverage was increased by 771 ha in the first period, while it was declined by 5,080.2 ha in the second period (Figure 6). In the third period, shrubland was decreased also by 7,832.2 ha. The average clearing rate of shrubland was increased by 517.2 ha year⁻¹ in the third period compared to the second period. In the shrublands, LST was increasing from 26.57°C in 1987 to 28.98°C in 2018, indicating that it was increasing by 2.4°C over the 31 years (Figure 7).

The analysis of LUC dynamic trends over the 31 years showed grassland covered 12.5% in 1987, 10.4% in 1999, 8.4% in 2010, and 7.6% in 2018. The coverage proportion was consistently decreasing by 8,380.3 ha over the three decades, with an average of 270.33 ha year⁻¹ (Figure 6). Grassland area decreased by 3,654.6, 3,485, and 1,240.7 ha in the first, second, and third periods in that order (Figures 6 and 8). This further implies that an average grassland area was decreasing by 304.55, 316.82, and 155.1 ha year⁻¹ in the first, second, and third periods, respectively. During the second period, the mean lost grassland was higher by 12.27 and 161.72 ha year⁻¹ when relatively seen for the first and third periods, respectively. In the grassland area, LST increased from 27.82°C in 1987 to 30.54°C in 2018, signifying that it was increasing by 2.72°C over the three decades (Figure 7).

Figure 8.

LUCs of the study watershed.

Areas with no vegetation were covered by 61.7% in 1987, 64.9% in 1999, 72.7% in 2010, and 79.4% in 2018 (Figure 6). In the area with no vegetation, the mean LST was 32.3°C over the three decades (Figure 7). The highest average LST was retrieved from the area with no vegetation cover when compared to other LUCs (Figure 7). The average value of LST over the 31 years in the area with no vegetation cover was higher by 7.58°C from wetlands, 7.03°C from forests, 4.1°C from shrublands, and 2.57°C from grasslands. Barren land showed the highest mean value of LST, while shrubs and sparse vegetation cover classes exhibited the lowest mean LSTs (Haylemariyam, 2018). The LSTs in areas with high human activity intensity like cultivated land were high but low in areas with low intensity of human activity, such as in water bodies, forests, shrublands, and grasslands (Deng et al., 2018; Ullah et al., 2019). Haylemariyam (2018) noted that the LST value of barren land surface exceeds other LUC classes with a mean value of 26.59°C but vegetation covers like shrubs and forests showed the lowest LST values having a mean of 24.57°C.

The result indicates that areas with no vegetation cover were incessantly increasing by 30,499.1 ha over three decades, with a mean of 983.84 ha year⁻¹ (Figures 6 and 8). No vegetation areas were increasing by 5,555.2, 13,375.6, and 11,568.3 ha in the first, second, and third periods, respectively. This indicates that on average 462.93, 1,215.96, and 1,446 ha year⁻¹ of no vegetation covers were increasing in the first, second, and third periods, respectively. In the third period, the average increment of no vegetation areas was higher by 983.1 ha year⁻¹ for the first and 230 ha year⁻¹ for the second period. In the no vegetation area, the average LST was ranging from 30.147°C in 1987 to 33.42°C in 2018, implying that it was increased by 3.05°C over the three decades (Figure 7).

From the total area of the study watershed, wetlands were covered by 1.1% in 1987, 0.32% in 1999, 0.24% in 2010, and 0.05% in 2018. Wetlands were constantly decreasing by 1,756.9 ha over the three decades, with an average of 56.7 ha year⁻¹. They decreased by 1,285.5 ha in the first period, 149.6 ha in the second period, and 321.8 ha in the third period. On average, wetlands declined by 107.125 ha year⁻¹ in the first period, 13.6 ha year⁻¹ in the second period, and 40.225 ha year⁻¹ in the third period. In the first period, the wetlands disappeared by 93.525 ha year⁻¹ higher than for the second period and 66.9 ha year⁻¹ for the third period. Because of the declining of wetlands coverage, the mean LST of the wetlands was decreasing from 24.63°C in 1987 to 18.96°C in 2018 (Figure 7). In the wetlands, the average LST was 24.72°C over the three decades. The least LST was observed in the wetland ecosystems when compared to other LUCs. Wetland surfaces play an important regulatory role in reducing LST and mitigating thermal effects on the ground (Tan et al., 2020). The distribution of average changes in LST in the water bodies showed the lowest increase than other LUCs followed by forest lands (Hua & Ping, 2018). Despite the enhanced available energy, the average daytime surface temperature in the wetlands was cooler by 5.1°C than the nearby cropland surface temperatures (Stanley, 2018).

Landuse/cover changes and surface temperature dynamics

In the first period, 38.2% of forests changed into other LUCs. Of the total changed, the highest proportion (32%) of forests lost at the gain of shrubs. During the second period, a more transition rate (53.5%) was observed compared to other periods of the study. From this, 44.3% of the forest was changed into shrubs. In the third period, 46.6% of forest lands were lost at the gains of other LUCs. Of this, 30.1% was modified into shrublands. Over the three decades, of the total 72% losses of forests, 40.6%, 17.2%, and 14.2% were converted into shrublands, no vegetation, and grasslands, respectively (Table 4). This may be due to the encroachment of croplands at the expense of grasslands, shrublands, and forests (Assefa & Singh, 2017; Barlow et al., 2016; Liyehu et al., 2019; Twisa & Buchroithner, 2019). The conversion of forests into shrublands and grasslands was driven by cutting live vegetation (ranked by 79%), overgrazing (ranked by 87.6%), and weak institutional arrangement (ranked by 73%) (Table 3). Table 3 indicates that clearing of the forest into no vegetation was also influenced by the encroachment of croplands (rated by 81%) and weak institutional arrangement (ranked by 69%). Expansion of cultivated land and settlements and cutting of vegetation for charcoal production and construction were the most crucial drivers of deforestation (Kindu et al., 2015; Liyehu et al., 2019). Deforestation, removing the natural vegetation cover, leads to soil erosion and ecosystem disturbance, alters surface roughness and reduces evapotranspiration, and increased albedo (Yin et al., 2018). Deforestation was the second most contributing factor to the increment of LST by 3.22°C (Figure 9). The increase of LST was largely determined by deforestation (Meseret, 2019). Because of anthropogenic disturbances, deforestation has a significant warming effect in the tropical regions with little seasonality (Danneyrolles et al., 2019; Findell et al., 2017; Muro et al., 2018) since it systematically results in higher radiative fluxes leaving the surface (Duveiller et al., 2018). Deforestation-induced warmer temperatures tend to reduce land carbon, mostly because of increased heterotrophic respiration (Harper et al., 2018). From biophysical and biogeochemical points of view, conversion of forests into croplands, and open lands appears to be the worst land cover transition for climate because it constantly leads to local warming as a result of lower carbon stocks and considerable emission of nitrous oxide and methane (Duveiller et al., 2018).

Table 3.

Conversion of Forest and Shrubs into Other LUCs Like Grasslands and No Vegetation (Such as Croplands).

Figure 9.

LST dynamics (in °C) over three decades due to conversion from one LUC to others.

From the total area, a consistent increasing loss of shrubland was observed at 42.3% in the first, 55% in the second, and 56.3% in the third period (Table 4). In the first period, from 42.3% lost shrubland, 28.3% was cleared for crop production, and 11.5% was developed into the forest. In the second period, from 55% lost shrubland, 29.5%, and 21.8% were cleared to the gain of cultivated lands and grasslands, respectively. Of the total lost shrublands, 27.9% were cleared into croplands, followed by 23.5% were converted into grasslands in the third period. Over the three decades, from 77.7% lost shrublands, 53.8% were cleared into no vegetation (croplands) and 20.25% were modified into grasslands. This agrees with the expansion of cultivated lands at the expense of shrublands (Barlow et al., 2016; Liyehu et al., 2019; Twisa & Buchroithner, 2019). Cropland expansion and overgrazing due to the weak institutional arrangement were the major driving factors for the clearing of shrublands into no vegetation (Table 3). The clearing of shrublands caused an increase of LST by 2.02°C over the three decades (Figure 9). Changing shrublands alter the radiative and non-radiative properties of the land surfaces (Duveiller et al., 2018).

Table 4.

Transition Matrix of LUCs in Abaminus Watershed Between 1987 and 2018 (Unit: Hectare).

The grasslands were decreased by 78.7% in the first, 78.1% in the second, and 76.3% in the third period. The highest loss was observed in the first period. During this period, from the total 78.7% lost, 55.5% changed into croplands, and 20.9% converted into shrublands (Table 4). This further implies that 4.3% and 1.61% of the grasslands were lost to the gain of croplands and shrublands per year, respectively. Of the total lost area, 63.3% was changed into cropland and 13.9% was converted into shrubland in the second period. In the third period, 63.4% changed into cropland and 12.4% changed into shrubland. An area covered with grassland was lost by 89.4% of the gain of other LUCs over the three decades. Of the total lost, 61.8% was cleared into no vegetation cover (Table 4). The grassland can be converted into no-vegetation, including croplands and barren surfaces due to overgrazing (highly rated 77.5%) and weak institutional arrangements, highly rated by 76% (Table 3). The results of this study agree with the encroachment of croplands at the expense of grasslands (Assefa & Singh, 2017; Liyehu et al., 2019; Twisa & Buchroithner, 2019). Clearing of grasslands into barrenland and the soil surface contributed to increasing LSTs by 0.45°C (Figure 9). The conversion of grasslands into croplands has consistently reduced the amount of energy available for evapotranspiration whilst brightening the surface (Duveiller et al., 2018).

The largest area of the wetlands was converted into no-vegetation. From the total lost area of the wetlands, 78.5% (in the first period), 59.9% (in the second period), and 84.5% (in the third period) were converted into no-vegetation. The result indicates that the largest lost area of wetlands at the gain of no-vegetation was observed in the third period when compared to the first and second periods. Of the total lost area of the wetlands (96.6%), 94.5% were to the gains of no-vegetation over the three decades (Table 4) due to drainage for irrigated croplands (83.6% highly rated), sedimentation (71% highly rated), and weak institutional principles and arrangement (73 % highly rated) (Table 5). Loss of wetlands or water surface changes over time were the major environmental degradation indicators mainly identified in the arid environment (Hadeel et al., 2011). Figure 9 reveals that wetland disappearance contributed to the increment of LST by 11.5°C over 31 years, implying that on average 0.37°C year⁻¹ due to wetlands losing. Despite a decline in methane oxidation emission following wetland drainage, conversion of wetlands into cropland results in a considerable net increase in land surface temperature or heat-trapping (Muro et al., 2018). Anthropogenic disturbance and rising LST have great effects on the status of wetlands changing their hydrological and evapotranspiration regimes (Muro et al., 2018), and carbon storage (Moomaw et al., 2018). Drainage and drying of wetland soils increase microbial decomposition of stored organic carbon and cause to emit significant amounts of this stored natural carbon into the atmosphere and reduce the ability of wetlands to sequester additional carbon (Pendleton et al., 2012). Preventing further wetland loss is found to be crucial to protect carbon stores and in limiting future emissions to meet climate goals, but is seldom considered (Moomaw et al., 2018). Moomaw et al. (2018) further noted that wetlands are among the most carbon-rich sinks on the planet sequestering approximately as much carbon as do global forest ecosystems.

Table 5.

The Most Driving Factors for the Conversion of One Land Cover into Another.

The area with no vegetation was covered by 3.6% of grasslands, 2.2% of shrublands, and 0.4% of forest over the three decades (Table 4). From the total planted area, an area with no vegetation was vegetated by 8.4% grasslands and 6.1% shrublands in the first period; by 4.4% grasslands, and 2.3% shrublands in the second; and by 8% grasslands and 1.2% shrublands in the third period. This indicates that the largest area with no vegetation was consistently covered by grasslands, followed by shrublands. Table 5 indicates that from the total respondents, 76.3% reported that fallow practice was the most important factor to change cultivated and barren land surfaces into grasslands. Of the total, 89.1% reported, plantation of eucalyptus and other exotic plant species on unproductive croplands and degraded communal lands was rated a highly important factor in the conversion of no vegetation into shrublands (Table 5). A realistic representation of the Spatio-temporal dynamics of key land-use practices, such as fallowing, and abandonment of land, afforestation, and reforestation is one of the priorities of earth system modeling (Chen et al., 2019). Figure 9 reveals that the plantation of shrubs and trees has also contributed to the decrease of LST by 3.77°C, with an average rate of 0.12°C year⁻¹. Thomas et al. (2018) reported that daytime air temperatures under shrubs and trees were cooler compared with other LUCs areas.

Afforestation is being changed of other LUCs into forested areas (Hu et al., 2019). Afforestation of soil and the barren surface has contributed to the decline of LST by 5.42°C, on average, by 0.17°C year⁻¹ (Figure 9). The mean LST of green space was 3°C lower than impervious and barren soil surface, implying the important role of plantation practice in mitigating surface temperature (Estoque et al., 2017). Forests can serve as a thermal insulator compared with open areas and such buffering effect has the potential to reduce the severe impacts of climate change on the forest ecosystem (De Frenne et al., 2019). Re-/afforestation as local mitigation and adaptation measures could be an attractive strategy because of the increasing annual cooling effects of forest cover (Bright et al., 2017). Afforestation/reforestation, and avoiding deforestation encourage carbon capture and storage; can support large-scale carbon dioxide (CO₂) removal from the atmosphere to limit global warming to below 2°C by 2,100 (Harper et al., 2018). Reducing emissions from tropical deforestation and forest biomass degradation is beneficial for climate mitigation for both biogeochemical and biophysical reasons (Duveiller et al., 2018; Syktus & McAlpine, 2016). Syktus and McAlpine (2016) noted that restoration of forest and shrubland biomass triggers a positive feedback loop between the land surface and the atmosphere by increasing an evaporative fraction (proportion of latent heat flux relative to the sensible heat flux), eddy dissipation, and turbulent mixing in the boundary-layer resulting in the enhanced cloud formation and precipitation over the restored areas.

Restoration of wetlands decreased LST by 5.21°C, implying that on an average decline of 0.17°C year⁻¹. The conversion of the bare surface into water bodies reduced surface temperatures by 4.5°C in Zimbabwe (Mushore et al., 2017). Water bodies play a significant regulatory role in reducing LST and mitigating thermal effects on the ground (Tan et al., 2020). Tracts of open water between the patches of wetland vegetation could delay heat exchange with the atmosphere and keep temperature cooling that helps counteract rising temperatures on a local scale (Stanley, 2018). Preventing and restoring wetland ecosystems can regenerate their ability to remove and sequester CO₂ from the atmosphere (Moomaw et al., 2018). Wetland restoration has significant implications for an atmospheric carbon cycle (C-cycle) because a substantial portion of the soil carbon pool is stored in the wetlands (Moomaw et al., 2018).