Time Series Analysis of Land Surface Temperature and Drivers of Urban Heat Island Effect Based on Remotely Sensed Data to Develop a Prediction Model

Figures & data

Figure 1. Spatial variation of observed (a) and predicted (b) LST for May.

Figure 2. Study area map.

Figure 3. CNN network structure.

Figure 4. (a) and (b) CNN model methodology.

Figure 5. EVI images of the Lahore district of January (a) and May (b) of 2019.

Figure 6. Road network map (a) and road density (b) of the Lahore district.

Figure 7. Downloaded ASTER DEM (a) and resampled DEM (b) of the Lahore district.

Figure 8. Spatial variation of observed (a) and predicted (b) LST for January.

Figure 9. Difference among observed and model-predicted LST for January (009–016).

Figure 10. Difference among observed and model-predicted LST for May (137–144).

Figure 11. Spatial variation of error between the observed and predicted LST for January (a) and May (b).

Table 1. Used remote sensing data for this study

Remote Sensing Product	Short Name	Sensor	Platform	Temporal Resolution	Spatial Resolution (m)
Land Surface Temperature and Emissivity	MYD11A2	MODIS	Aqua	8-day	926.626
Vegetation Indices	MYD13A2	MODIS	Aqua	16-day	926.626
Land Cover Type	MCD12Q1	MODIS	Combined Aqua and Tera	Yearly	463.3
Digital Elevation Model	ASTGTM	ASTER	Terra	_	24.8

Table 2. Accuracy of different model types

Model Type	Accuracy
ResNet	0.81
GoogLeNet	0.85
VGGNet	0.84
CNN	0.87

Table 3. Comparison of maximum, minimum, and mean actual and predicted LST values for 2020

Month (Day No.)		Status	Maximum	Minimum	Mean	Standard Deviation
January (009–016)	Actual		293.18	286.86	291.15	1.371
	Predicted		293.74	287.25	290.31	1.361
May (137–144)	Actual		317.38	307.14	312.26	1.375
	Predicted		315.98	306.59	311.51	1.428

Table 4. MAE, MAPE, and MSE for both periods

Data Type	Training		Validation
Month (Day No.)	January (009–016)	May (137–144)	January (009–016)	May (137–144)
MAE (K)	0.29	0.28	0.27	0.25
MAPE (%)	0.14	0.15	0.13	0.14
MSE	0.24	0.25	0.22	0.24

Table 5. Number of pixels (% of pixels) of the Lahore district in different error ranges

Month (Day No.)	Between −3 and −2 K	Between −2 and −1 K	Between −1 and 0 K	Between 0 and1 K	Between 1 to 2 K
January (009–016)	186 (9%)	247 (12%)	906 (44%)	639 (31%)	83 (4%)
	Between −2 and −1 K	Between −1 and 0 K	Between 0 and 1 K	Between 1 and 2 K	Between 2 and 3 K
May (137–144)	21 (1%)	123 (6%)	1051 (51%)	597 (29%)	268 (13%)