Data on time series analysis of land surface temperature variation in response to vegetation indices in twelve Wereda of Ethiopia using mono window, split window algorithm and spectral radiance model.

In the past, decadal time-series analysis has been done traditionally using meteorological data. In particular, decadal analysis of land surface temperature has been a major issue due to the unavailability of remote sensing techniques. But, nowadays, with the recent advances in remote sensing techniques and modern software Land Surface Temperature (LST) can be calculated through the thermal bands. LST can be estimated through many algorithms such as Split-window, Mono-Window (SW), Single-Channel (SH), among others. LST was estimated using Mono-Window algorithm on Landsat-5 TM, Landsat-7 ETM+ and split window algorithm on Landsat-8 OLI/TIRS Thermal Infrared (TIR) bands. Vegetation index was obtained by using Normalized Difference Vegetation Index (NDVI) from red and Near-Infrared (NIR) bands. NDVI has been effectively used in vegetation monitoring and to analyze the vegetation in responses to climate change such as surface temperature variation. The twelve Weredas (third-level administrative divisions) of Ethiopia which are highly prone to drought were selected to investigate decadal land surface temperature variations and its impact on the surrounding environment, especially on vegetation cover. Ten Landsat images of three different sensors from 1999 to 2018 were used as the basic data source. The processed data of surface temperature and vegetation indices showed a strong correlation. The higher LST values indicate the smaller NDVI and vice versa and it is also identified the areas with high temperature being barren regions and areas with low temperature covered with more vegetation.

Data

The data discussed in this article were used to calculate the spatiotemporal variation of LST using mono, split window algorithm and land cover types based on NDVI over two decades from 1999 to 2018 for twelve Weredas (Fig. 1, Table 1). Ten Landsat images were spatially analysed using their respective thermal infrared bands and the computed LST are illustrated in Fig. 2a–j using the temperature range. The variation of minimum and maximum surface temperature is depicted in Fig. 3. The area coverage of land covers such as water bodies, sparse vegetation and dense vegetation are shown in Fig. 4a–j using their NDVI values and range used to identify the different land cover types are depicted in Table 2 with their respective area coverage in Table 3.The minimum and maximum NDVI values for two decades are specified in Table 4. The comparison of area coverage of three land cover types and its spatial correlation with temperature are analysed in Table 5. Fig. 5 exhibits the relationship between land cover feature, temperature and respective area. The raw data provided in supplementary file.

Fig. 1

Drought-prone Weredas in Ethiopia.

Table 1

The spatial extent of Weredas.

S.No	Wereda	Region	Zone	Elevation	Area (Km2)
1	Adaa Chukala	Oromia	East Shoa	1953	1668.9
2	Adama	Oromia	East Shoa	1675	999.3
3	Akaki	Oromia	East Shoa	2173	629.6
4	Berehet	Amhara	Semien Shoa	1442	996.12
5	Berehna Aletu	Oromia	North Shoa	2707	1297.1
6	Boset	Oromia	East Shoa	1378	1378.5
7	Dodotana Sire	Oromia	Arsi	1742	1005.4
8	Gimbichu	Oromia	East Shoa	2305	737.1
9	Hagere Mariamna Kesem	Amhara	Semien Shoa	2323	861.3
10	Jeju	Oromia	Arsi	1952	846.1
11	Lomie	Oromia	East Shoa	1981	685.9
12	Shenkorana Minjar	Amhara	Semien Shoa	1570	1534.7

Fig. 2

LST scenario of twelve Weredas from 1999 (a) to 2018 (j).

Fig. 3

Minimum and maximum surface temperature variation.

Fig. 4

Scenario of Water bodies, dense and sparse vegetation from 1999 (a) to 2018 (j).

Table 2

Land cover classification range based on NDVI.

Table 3

Area coverage of Land cover types.

S.No	Year	Water bodies (hectare)	Sparse Vegetation (hectare)	Dense Vegetation (hectare)
1	1999	30,827.25	284,304	13,230.99
2	2000	34,471.17	266,304.7	17,204.94
3	2003	11,017	604,279.1	40,497.48
4	2010	36,625	436,734.9	74,115.81
5	2011	90,989	252,815.5	46,334.34
6	2013	24,748	343,214.6	115,047.1
7	2014	29,823	403,135.4	55,068.84
8	2015	32,927	274,678.1	23,976.81
9	2016	32,810	355,501.7	51,117.39
10	2018	63,275	182,237.8	18,660.96

Table 4

Minimum and maximum of NDVI values for two decades.

S·NO	NDVI image of (year)	Minimum value	Maximum value
1	1999	−0.68067	0.58427
2	2000	−0.99225	0.60656
3	2003	−0.99225	0.71605
4	2010	−0.44528	0.70213
5	2011	−0.45882	0.71429
6	2013	−0.25203	0.59245
7	2014	−0.23899	0.57231
8	2015	−0.22938	0.52663
9	2016	−0.26238	0.57817
10	2018	−0.24386	0.5475

Table 5

Comparison matrix between Land covers and LST.

Land cover	1999			2000			Decadal change
	Area (Ha)	%	Avg.T	Area (Ha)	%	Avg.T	Area (Ha)	Avg.T
S. veg	2,84,304	86.58		2,66,044.70	83.74		↓
D. veg	13,230.90	4.03	25.02	17,204.9	5.42	29.77	↑	↑
WB	30,827.20	9.39		34,471.10	10.84		↑
Total	3,28,362.10			3,17,720.70
	2000			2003			Decadal change
S. veg	2,66,044.70	83.74		6,04,279.10	92.14		↑
D. veg	17,204.9	5.42	29.77	40,497.48	6.18	27.99	↑	↓
WB	34,471.10	10.84		11,017	1.68		↓
Total	3,17,720.70			6,55,793.58
	2003			2010			Decadal change
S. veg	6,04,279.10	92.14		4,36,734.90	79.77		↓
D. veg	40,497.48	6.18	27.99	74,115.81	13.54	28.86	↑	↑
WB	11,017	1.68		36,625	6.69		↑
Total	6,55,793.58			5,47,475.71
	2010			2011			Decadal change
S. veg	4,36,734.90	79.77		2,52,815.50	64.8		↓
D. veg	74,115.81	13.54	28.86	46,334.34	11.88	29.07	↓	↑
WB	36,625	6.69		90,989	23.32		↑
Total	5,47,475.71			3,90,138.84
	2011			2013			Decadal change
S. veg	2,52,815.50	64.8		3,43,214.60	71.06		↑
D. veg	46,334.34	11.88	29.07	1,15,047.10	23.82	29.24	↑	↑
WB	90,989	23.32		24,748	5.12		↓
Total	3,90,138.84			4,83,009.70
	2013			2014			Decadal change
S. veg	3,43,214.60	71.06		4,03,135.40	82.61		↑
D. veg	1,15,047.10	23.82	29.24	55,068.84	11.28	29.19	↓	↓
WB	24,748	5.12		29,823	6.11		↑
Total	4,83,009.70			4,88,027.24
	2014			2015			Decadal change
S. veg	4,03,135.40	82.61		2,74,678.10	82.84		↓
D. veg	55,068.84	11.28	29.19	23,976.81	7.23	28.72	↓	↓
WB	29,823	6.11		32,927	9.93		↑
Total	4,88,027.24			3,31,581.91
	2015			2016			Decadal change
S. veg	2,74,678.10	82.84		3,55,501.70	80.9		↓
D. veg	23,976.81	7.23	28.72	51,117.39	11.63	30.21	↑	↑
WB	32,927	9.93		32,810	7.47		↓
Total	3,31,581.91			4,39,429.09
	2016			2018			Decadal change
S. veg	3,55,501.70	80.9		1,82,237.80	68.98		↓
D. veg	51,117.39	11.63	30.21	18,660.96	7.06	31.41	↓	↑
WB	32,810	7.47		63,275	23.96		↑
Total	4,39,429.09			2,64,173.96

Note: S.Veg – sparse vegetation; D.Veg – Dense vegetation; WB – Water bodies; Avg.T- Average temperature; Ha-hectare; ↓ - decrease; ↑- increase.

Fig. 5

Comparisons of land cover feature, area and temperature.

Experimental design, materials, and methods

Micro and global Climate change have been a significant issue over the last several decades. Time series cloud free remotely sensed data are acquired in different spectral bands [2] provides a powerful tool to learn from past, monitoring current scenario and also to predict the feature change. Land surface temperature and land cover have a vital role in influencing the environment and landscape attributes [3]. TM (2010, 2011), ETM+ (1999, 2000, 2003) and OLI-TIRS (2013, 2014, 2015, 2016, 2018) of overall 30 m spatial resolution and 60 m spatial resolution for TIR respective to band six in TM, ETM+ and 100 m spatial resolution for TIR1 and TIR2 respective band ten and eleven with path and row of 168/54 were the secondary data employed in this article. The primary data of air temperature was collected from automatic weather station by Ethiopian meteorological agency were utilised to validate the LST images. To begin with NIR and the red band were radiometrically corrected to remove the error due to noise from sensor and atmosphere by converting the digital number to radiance and further to reflectance using spectral radiance model.

To estimate the radiance of Landsat-5 TM (Eq. (1)) and Landsat-7 ETM+ (Eq. (2)) from digital number (DN) the following equation was used [4].where, Lλ is the spectral radiance, LMin is the spectral radiance of DN value 1, LMax is the spectral radiance of DN value at 255, and DN is the digital Number [4].where, Lλ is the spectral radiance at sensors aperture or the calculated radiance associated with the ground area enclosed in the pixel and referred to as the wavelength range of specific band, LMax is spectral at the Sensor Radiance scaled to QCal Max, LMin is Spectral at the Sensor Radiance scaled to QCal Min, QCal Max is the maximum quantized calibrated pixel values corresponding to LMax, QCal, Min is the minimum quantized calibrated pixel value (DN) corresponding to LMin and QCal is the quantized calibration pixel value [4].

Further the obtained radiance value from Landsat −5 and 7 were converted reflectance using Eq. (3).where unit less planetary reflectance, represents the spectral radiance at sensors aperture, d is the earth-sun distance in astronomical units calculated from the meta data, obtained from mean solar exo-atmospheric irradiance from table and represents the solar zenith angle.

Whereas the DN values are directly converted to reflectance in Landsat-8, OLI/TIRS, TIR band using the following Eq. (4).where p λ′ is the Spectral reflectance, Mp is the multiplicative reflectance value, Ap is the additive reflectance value and QCal is the pixel values in DN. The obtained spectral reflectance is not true reflectance it is needed to carry out solar elevation angle correction to obtained true top of atmosphere (TOA) Eq. (5).where p λ is the true planetary reflectance and ɵ is the solar elevation angle.

Then the radiometrically corrected NIR and red bands were used to calculate the NDVI values from 1999 to 2018 using Eq. (6).

Later the land cover features such as Water Bodies, Barren Area, Sparse Vegetation and Dense Vegetation [[5], [6], [7], [8], [9]] were obtained from the processed NDVI data using the ranges depicted in Table 3. Finally the LST from band six was obtained by converting the radiance value to surface temperature in Kelvin for Landsat-5 and 7 using mono window algorithm [[10], [11], [12]] as expressed in Eq. (7).where, TB is the effective satellite brightness temperature in Kelvin, K1 is the calibration constant 1, K2 is the calibration constants 2, Ln is the natural logarithm, Lλ is the spectral radiance in watts [4].

The radiance value for band ten and eleven of Landsat-8 from the raw DN pixels was calculated using Eq. (8).where L λ is the Spectral Radiance, ML is the multiplicative radiance value, AL is the additive radiance value and QCal is the pixel values in DN.

The surface temperature using band ten and eleven from Landstat-8 was derived using the split window algorithm [13,14].where LST = Land surface temperature, C0–C6 = Split window coefficient values, TB10 and TB11 = Brightness temperature of band 10 and band 11, Ɛ = M band 10 and band 11, Ɛ = Mean LSE of TIR bands, W = Atmospheric water vapor content, ΔƐ = Difference in LSE based on the methods provided by Ref. [15] in Eq. (10).where A L = Radiance add, M L = Radiance multiplier, DN = Digital number.

The Brightness temperature (TB) for both TIR bands was calculated by adapting the following Eq. (11).where K1 and K2 = Thermal constant for TIR bands, TB = Brightness temperature, TOAr = Atmospheric spectral radiance.

The data processed to obtain NDVI can be interpreted based on their class range Table 3. Feature such as water bodies, dense vegetation, sparse vegetation is found commonly in the input data. Before classifying the NDVI values it must be sure that values must fall between the range −1 and 1 [4]. Negative values of NDVI approaching −1 corresponds to water bodies, values close to (−0.1 to 0.1) generally indicates barren land, rock, sand or snow. Whereas positive values between 0.2 and 0.4 represent shrub and grassland. While, a high value approaching 1 indicates dense, temperate or tropical rainforest. It may also require additionally performing the manual adjustment in the NDVI range due to the difference in sun angle and acquisition date and year. The spatially analyzed Landsat images using their thermal infrared bands clearly exhibits the surface temperature variation in 20years from 1999 to 2018. It is identified that the minimum and maximum temperature value was least in 1999 with an average temperature of 25.02 0C and surged to the highest of 31.41 0C in 2018. Interestingly it is found that NDVI and LST show a trend of strong correlation. In many Weredas the higher NDVI implies lower LST value and vice versa. The area of land cover types also decreased in case of increasing LST. One of the drought-prone Wereda such as Adama showed a steady rise in land surface temperature with a decrease in the area of water bodies and vegetation. Hence this condition can be reversed by increasing the plantation as a part of microclimate change.

Specification Table

Subject Area	Land Surface Temperature and Environmental Studies
More specific subject Area	Land Surface Temperature and Vegetation Indices
Type of data	Table, figure and text file
How data was acquired	Data were extracted from various Landsat sensors such as ETM+, TM and OLI TIRS with path/row numbers168/54 and primary data of air temperature was obtained from Ethiopian meteorological agency.
Data format	Raw and Analyzed
Experimental factors	We make use of data from USGS and Ethiopian meteorological agency for mapping change in surface temperature and its impact on vegetation and vice versa.
Experimental features	The data were radiometrically corrected using spectral radiance model. The land surface temperature and vegetation indices were calculated employing Mono window, split window and NDVI algorithm, using Thermal bands NIR and Red band respectively in R Studio.
Data source location	Landsat ETM+, TM and OLI TIRS, twelve Wereda (8°10′–9°25′N, 38°40′–40°00′E)
Data accessibility	Data are available in this article and supplementary file
Related research article	A.S.M. Abdul Athick, K. Shankar, Data on Land Use and Land Cover Changes in Adama Wereda, Ethiopia, on ETM+, TM and OLI- TIRS landsat sensor using PCC and CDM techniques, Data in Brief, 24 (2019) [1].

Value of the Data• The Presented datasets speculate the LST and its spatial correlation with NDVI across twelve Weredas of Ethiopia for two decades. • The data provides information on the variation of surface temperature on drought-prone Weredas. • The data can aid to analyze the influence of land surface temperature variation over large and as well as a small area. • The generated data can be utilized for statistical analysis of LST, NDVI and other variables for twelve Weredas in Ethiopia. • The data can be useful for further research in various aspects of environmental monitoring.