Downscaling of thermal images over urban areas using the land surface temperature–impervious percentage relationship

doi:10.1016/j.jag.2012.12.007

International Journal of Applied Earth Observation and Geoinformation

Volume 23, August 2013, Pages 95-108

https://doi.org/10.1016/j.jag.2012.12.007 Get rights and content

Abstract

Intensive expansion and densification of urban areas decreases environmental quality and quality of urban life as exemplified by the urban heat island effect. For this reason, thermal information is becoming an increasingly important data source for integration in urban studies. It is expected that future spaceborne thermal sensors will provide data at appropriate spatial and temporal resolutions for urban studies. Until they become operational, research has to rely on downscaling algorithms increasing the spatial resolution of relatively coarse resolution thermal images albeit having a high temporal resolution. Existing downscaling algorithms, however, have been developed for sharpening images over rural and natural areas, resulting in large errors when applied to urban areas. The objective of this study is to adapt the DisTrad method for downscaling land surface temperature (LST) over urban areas using the relationship between LST and impervious percentage. The proposed approach is evaluated by sharpening aggregated LST derived from Landsat 7 ETM+ imagery collected over the city of Dublin on May 24th 2001. The new approach shows improved downscaling results over urban areas for all evaluated resolutions, especially in an environment with mixed land cover. The adapted DisTrad approach was most successful at a resolution of 480 m, resulting in a correlation of R² = 0.84 with an observed image at the same resolution. Furthermore, sharpening using the adapted DisTrad approach was able to preserve the spatial autocorrelation present in urban environments. The unmixing performance of the adapted DisTrad approach improves with decreasing resolution due to the fact that the functional relationship between LST and impervious percentage was defined at coarse resolutions.

Highlights

► DisTrad using impervious percentage performs better than NDVI for urban areas. ► The improvement is applicable for sharpening LST images at different resolutions. ► DisTrad using impervious percentage preserves the spatial autocorrelation.

Introduction

In 1950 30% of the world's population was living in urban areas, while this value will increase to 60% in 2030 (Golden, 2004). The growing population in urbanized areas causes an increase of anthropogenic heat. As a result the core of cities in temperate climate zones becomes warmer than their periphery, thus forming an urban heat island (UHI) (Voogt and Oke, 2003), which has an impact on the microscale and mesoscale climate. Heat-related morbidity and mortality are among the primary health concerns and are expected to increase as a result of climate change that increases the UHI effect (Johnson and Wilson, 2009).

The UHI intensity is influenced by local and synoptic weather, season, time of day, size of the city and its geographical location, urban morphology, and anthropogenic heat (Kim and Baik, 2005). Therefore, a range of spatial and temporal scales should be considered in UHI studies. One of the great challenges for current UHI studies is how to support this research with techniques to retrieve land surface temperature data with high temporal and spatial resolution. This data has to support the scientific and policy makers’ community in understanding impacts, both positive and negative, to set policies for mitigating those impacts or to develop indicators to address the UHI effect within a sustainable development context. Therefore land surface temperature is a physical parameter of prime importance for a wide variety of scientific urban studies. In urban climatology, land surface temperature is the key input parameter used in algorithms estimating net radiation, sensible and latent heat flux, thermal inertia, surface urban heat island intensity, perceptible water, evapotranspiration, urban-induced surface runoff and surface moisture (Stathopoulou and Cartalis, 2009). Various spaceborne thermal sensors provide useful information for urban studies. The choice of an appropriate sensor is currently a trade-off between spatial and temporal resolution: sensors with a high temporal resolution have a low spatial resolution and vice versa.

Thermal sharpening is a promising technique for enhancing the spatial resolution of mixed pixels acquired by coarse resolution sensors. During the last ten years various thermal sharpening techniques for unmixing land surface temperature have been developed based on land cover. Thermal sharpening methods in remote sensing can be grouped into two categories. The first method, known as the physical downscaling method, is based on establishing a functional relationship between land surface temperature and ancillary data, which satisfies two conditions: (1) the functional relationship is physically meaningful and holds across different resolutions; and (2) the ancillary data can be easily scaled (Zhukov et al., 1999, Kustas et al., 2001, Liu et al., 2006, Agam et al., 2007a, Agam et al., 2007b, Liu and Pu, 2008). The second method, known as the statistical downscaling method is based on the direct estimation of subpixel land surface temperature from higher spatial resolution ancillary data without referring to physical models at different resolutions (Ottlé et al., 2008, Stathopoulou and Cartalis, 2009). Testing of the latter method is still in a preliminary stage for applications in urban areas.

In a previous study Essa et al. (2012) evaluated the relationship between land surface temperature of urban pixels and 15 indices and concluded that the relationship with impervious percentage gives the best results. This conclusion is also in agreement with findings from Yuan and Bauer (2007), who suggest that impervious percentage accounts for most of the land surface temperature variation in urban areas. Furthermore, Essa et al. (2012) and Yuan and Bauer (2007) conclude that there is a strong linear relationship between land surface temperature and impervious percentage, independent of the season of image acquisition. In contrast, the relationship between land surface temperature and NDVI varies with the seasons (Kaufmann, 2003). Therefore, Essa et al. (2012) suggested to use impervious percentage instead of NDVI in the DisTrad sharpening algorithm (Kustas et al., 2001) when applied to urban areas. They demonstrated the improvement for sharpening of a Landsat 7 ETM+ image of Dublin.

Due to the low temporal resolution of Landsat images, the current study evaluates the adapted DisTrad algorithm developed by Essa et al. (2012) for sharpening images with a spatial resolution comparable with daily MODIS land surface temperature products. While in Essa et al. (2012) only 25% of the pixels with the lowest coefficient of variance (CV) within a coarse resolution pixel were used in the regression, this study evaluates different threshold values. Furthermore, a detailed analysis of the spatial autocorrelation of the unsharpened and sharpened products is presented. The images sharpened with the original and the adapted DisTrad algorithm are compared with each other and with non-sharpened images using visual, statistical, and geostatistical analysis methods.

Section snippets

Study area and data

This study uses the 60 m resolution land surface temperature image that was derived by Essa et al. (2012) as a basis for deriving coarse resolution images and reference images for evaluation of the sharpening. The image was derived from a Landsat 7 ETM+ scene from May 24th 2001 covering 2700 km², bounded by the longitudes 6°00′ and 6°48′W and latitudes 53°07′ and 53°35′N. It covers the urbanized Mid-East of the Greater Dublin region (Fig. 1). This area is characterized by heterogeneous urban

Sharpening experiment

Ideally, the adapted DisTrad approach applied to a coarse resolution land surface temperature product needs to be evaluated with high resolution land surface temperature data acquired at the same moment. Image products from different sensors, however, differ in acquisition time and date, data level, processing algorithms, temporal integration, geometric accuracy, etc. Therefore, in this study a Landsat 7 ETM+ image was aggregated to 960 m resolution and evaluated for downscaling using the

Regression analysis

The selection of a subset of the most uniform high resolution pixels within a coarse pixel using the CV, as suggested by Kustas et al. (2001), was evaluated for two objectives. The first objective was to evaluate the applicability of using 25% instead of all pixels at the five observed image resolutions (60, 90, 120, 240 and 480 m). Only pixels within the urban mask were considered. The second objective was to determine the optimal index pixel fraction to be used in the downscaling application

Conclusions

The adapted DisTrad algorithm proposed by Essa et al. (2012) was evaluated for its feasibility to sharpen images with a spatial resolution comparable with daily MODIS land surface temperature products. In addition, the appropriate threshold for the selection of pixels with the lowest CV that are used in the regression analysis was determined.

Although sharpening results were not evaluated using real images, the adapted DisTrad approach does appear to have potential for generating useful land

Acknowledgements

This research was part of a PhD study, which was supported by an ERASMUS MUNDUS Scholarship based on the bilateral agreement between Al-AQSA University (Palestine) and the EU. The research presented in this paper was supported by the Belgian Science Policy Office in the frame of the STEREO II programme – project SR/00/105. We are indebted to the MAMUD project for providing this research with the basic images, maps, and information. The authors of this research acknowledge the assistance of

References (20)

N. Agam et al.
A vegetation index based technique for spatial sharpening of thermal imagery
Remote Sensing of Environment
(2007)
A. Dominguez et al.
High-resolution urban thermal sharpener (HUTS)
Remote Sensing of Environment
(2011)
W. Essa et al.
Evaluation of DisTrad thermal sharpening for urban areas
International Journal of Applied Earth Observation and Geoinformation
(2012)
D.P. Johnson et al.
The socio-spatial dynamics of extreme urban heat events: the case of heat-related deaths in Philadelphia
Applied Geography
(2009)
Y. Liu et al.
Scaling of land surface temperature using satellite data: a case examination on ASTER and MODIS products over a heterogeneous terrain area
Remote Sensing of Environment
(2006)
M. Stathopoulou et al.
Downscaling AVHRR land surface temperatures for improved surface urban heat island intensity estimation
Remote Sensing of Environment
(2009)
J.A. Voogt et al.
Thermal remote sensing of urban climates
Remote Sensing of Environment
(2003)
F. Yuan et al.
Comparison of impervious surface area and normalized difference vegetation index as indicators of surface urban heat island effects in Landsat imagery
Remote Sensing of Environment
(2007)
N. Agam et al.
Utility of thermal sharpening over Texas high plains irrigated agricultural fields
Journal of Geophysical Research
(2007)
J. Golden
The built environment induced urban heat island effect in rapidly urbanizing arid regions – a sustainable urban engineering complexity
Environmental Sciences
(2004)

There are more references available in the full text version of this article.

Cited by (72)

A novel surface energy balance-based approach to land surface temperature downscaling
2024, Remote Sensing of Environment
Spatial downscaling satellite sensor-derived land surface temperature (LST) is of great importance for various environmental applications. However, the energy balance at the land surface is complex, especially in urban environments. As a result, the complexity of land surface thermal processes and the resulting LST cannot be accurately modeled using common downscaling methods. Here, we propose a new physical approach for LST downscaling based on surface energy balance (SEB) components which can capture this complexity and, thereby, increase the accuracy of downscaling. Experiments to demonstrate the effectiveness and accuracy of the proposed SEB approach were undertaken across 23 test sites in Europe and the United States. The research data included the Landsat 8 and Terra satellite sensor imagery products, land cover data, a digital elevation model, surface imperviousness maps and ground station data. Downscaled LST (DLST) maps at different spatial resolutions were produced using the developed SEB-based approach and compared to common benchmark methods including random forest regression (RFR), geographically weighted regression (GWR), area-to-point regression kriging (ATPRK), disaggregation of radiometric surface temperature (DisTrad), and three layers composition (TLC). Using the observed and In-situ LST, and soil temperature data in different land covers and seasonal conditions as the reference, the accuracy of the SEB and common benchmark approaches for LST downscaling were evaluated and compared. The average RMSE (bias) between the observed LST and DLST derived by SEB, RFR, GWR, ATPRK, DisTrad and TLC approaches for the different test areas were 1.15 (−0.42), 1.45 (0.91), 2.11 (1.02), 2.36 (−0.8), 2.79 (−2.31), and 2.45 (−1.48) K, respectively. Accuracy evaluation based on In-situ LST data showed that the RMSE decreased by 0.47–1.96 K (∼1.28 K on average) and the bias decreased by 0.15–0.98 K (∼0.57 K on average) when the SEB method was applied for LST downscaling in place of the other approaches, respectively. After the SEB approach, RFR approach had higher accuracy than other approaches. The mean RMSE between the observed LST product and the SEB (RFR) derived DLST at spatial resolutions of 480, 320, 240, 192, 120 and 30 m in the warm season were 0.89 (0.95), 0.93 (1.10), 0.99 (1.42), 1.07 (1.52), 1.21 (1.66), and 1.23 (1.94) K, respectively. In the cold season, these values were 0.52 (0.60), 0.58 (0.75), 0.67 (0.93), 0.78 (1.04), 0.87 (1.18), and 0.92 (1.36) K, respectively. Importantly, the effect of target spatial resolution (scale) on DLST accuracy was less acute for the SEB approach than the RFR approach. The mean RMSE between the DLST obtained from SEB (RFR) and the the observed LST for built-up land, cropland, barren land, grassland, shrubland and forest were 11.51 (2.32), 1.21 (1.49), 1.48 (2.04), 1.43 (1.62), 1.51 (1.53), and 1.12 (1.25) K, in the warm season and 0.92 (1.43), 0.76 (0.91), 0.86 (1.20), 0.76 (1.08), 0.77 (1.20), and 0.61 (0.72) K, in the cold season, respectively. The results of this research, thus, demonstrate the high efficiency of the proposed, novel SEB-based approach for downscaling satellite LST at different scales and under different climatic, geographical and environmental conditions.
Modeling the impact of urban land cover features and changes on the land surface temperature (LST): The case of Jordan
2024, Ain Shams Engineering Journal
Higher land surface temperature (LST) in cities than their surrounding areas presents a major sustainability challenge for cities. Decision-makers and planners use the LST measurements to monitor the urban environment to reduce the urban climate’s main challenges. Therefore, there is an urgent need to examine the impacts of urban features and changes on the LST. This study focused on the relationship between LST and urban land changes and the impact of these changes on LST during different periods. Although a set of studies explored the relationship between landscape features and urban LST, several aspects still need further discussion. Here, the study aims to explore the influence of landscape features and land cover patterns on land surface temperature LST in two cities (Amman and Zarqa) in Jordan, and identify which of these features (vegetation cover, built-up and population density) has the most effective and influence on LST values. Therefore, this paper is the first study about land surface temperature LST and its relations with these aspects in Jordan. This study used a mixed method approach using quantitative method (GIS) and qualitative method (comparative case studies). This study revealed that the most important features affecting the LST values were: (1) Population density; (2) Built-up; and (3) Vegetation, and in descending order from the strongest to the least effective. It is also concluded that in the city in which the population density is high, the effect of built-up areas on the values of LST is as high as possible and positive more than the cities with medium and low population density. As for the city in which the population density is medium to low, the effect of vegetation cover on the values of LST is greater, and this can be more positive than in cities with a high population density. Therefore, the study contributes to improving the planners' and policymakers’ suitable future decisions for making sustainable cities.
Heatwave vulnerability across different spatial scales: Insights from the Dutch built environment
2023, Urban Climate
Heatwaves in urbanized areas, even in temperate regions like the Netherlands, are getting serious attention. The Royal Netherlands Meteorological Institute predicts more frequent and intense heat events in the future. Studies have explored how Dutch cities contribute to heatwaves and suggested design and planning responses to mitigate their effects. However, a review of heatwave research in the Netherlands specifically focusing on the built environment has hardly been reported in the literature. This study aims to provide such a review utilizing the vulnerability framework. Following the PRISMA protocol, 57 articles are analysed based on the components of exposure, sensitivity, and adaptive capacity within the vulnerability framework. Subsequently, findings have been classified into five built environment scales - block, neighbourhood, district, city, and region - to critically reflect upon the extent to which the studies address various vulnerability components and the specific scales they primarily focus on. Results demonstrate that most of the studies concentrate on the hazard itself and its spatial distribution from a macro perspective on a city and regional scale. The review underlines the necessity of micro-level research on the phenomena, incorporating people's everyday experiences and resilience during heat events to find context-specific adaptation and mitigation strategies.
Fully automated land surface temperature downscaling based on RGB very high spatial resolution images
2023, City and Environment Interactions
Downscaling is a particularly needed process in many environmental, social and governance applications at the fine scale. The need for an automated and reliable very high spatial resolution downscaling approach is then required. In this paper, a fully-automated open-access downscaling approach was proposed, named HSR-LST. It is based on the High Spatial Resolution (HSR) Red, Green and Blue (RGB) bands collected from commercial and free-to-access satellite images, generating LST values lower than 2-m spatial resolutions. This is based on the Landsat-8 thermal datasets and while implementing a fully-automated Ordinary Least Squares (OLS) approach. HSR-LST was implemented over Beirut, Boston and Dubai between 2016 and 2018. In comparison to an airborne LST image captured over ElKhorn River in Nebraska, USA, HSR-LST showed an AME of 0.88 °C and a R-squared value of 86.33%. Main results showed the variability of LST based on the sensed land features’ type. Different LST distribution footprints (i.e., irregular in Beirut, intermitted in Boston, systematic in Dubai) were highlighted depicting a characteristic urban configuration in each city. This latter along buildings’ material, density and height appear also to show a different effect on the local and surrounding LST values. By implementing the automated HSR-LST model in cities around the Globe, urban planners, policy makers and inhabitants can acquire improved information to assess urban heat islands, to propose more adequate planning policies, but more importantly to tackle urban heat and thermal comfort at the finest scales. HST-LST will effectively address the low spatial resolution of thermal bands. As HSR-LST is both automated and dynamic, it can be portable to other urban areas with diverse climatic regions.
Quantification of landscape metrics effects on downscaled urban land surface temperature accuracy of satellite imagery
2022, Advances in Space Research
Citation Excerpt :
Considering the negative effects of these factors on the quality of human life, further investigations and policy-making are necessary to improve environmental conditions in the urban environment (Li et al., 2018a, Meng et al., 2018, Mijani et al., 2019, Nadizadeh Shorabeh et al., 2020). One major challenge in monitoring these negative consequences in the urban environment and implementing the necessary remedial policies is access to accurate data and information with high spatial and temporal resolutions (Essa et al., 2013, Bonafoni, 2016, Agathangelidis and Cartalis, 2019, Ebrahimy and Azadbakht, 2019, Firozjaei et al., 2021b). The Land Surface Temperature (LST) obtained from satellite images is a strategic parameter used by urban planners and decision-makers for managing and monitoring the environmental conditions of cities (Essa et al., 2017, Mijani et al., 2020, Nadizadeh Shorabeh et al., 2020, Chen et al., 2021, Firozjaei et al., 2021a).
Improving the spatial resolution of Land Surface Temperature (LST) obtained from satellite images is of great importance in various applications, especially in urban areas. Aiming to bridge the gap in the research background, this study attempted to quantify the effect of landscape metrics on the accuracy of urban LST downscaling. The research data included a collection of satellite images from Landsat and Terra satellites and auxiliary data of nine European cities, namely Lisbon, Madrid, Bucharest, Vienna, Prague, Paris, Copenhagen, Stockholm, and Helsinki. In the first stage, surface biophysical, topographic, and thematic property maps affecting the spatial distribution of LST were prepared. Secondly, the effect of the endogenous (influence of intra-pixel properties) and exogenous (influence of landscape metrics) variables of each surface property on LST was evaluated in warm and cold seasonal conditions. Finally, the effect of landscape metrics on the accuracy of LST downscaling from 960 m to 30 m was quantified. The results showed that the mean absolute correlation coefficient (r) values between the endogenous (exogenous) effects of surface biophysical, topographic, and thematic properties and LST in the cold season were 0.25 (0.12), 0.19 (0.06), and 0.09 (0.04), respectively. For the warm season, these values were 0.55 (0.24), 0.11 (0.03), and 0.23 (0.09), respectively. The mean r (root-mean-square error (RMSE)) between the actual LST and downscaled LST (DLST) obtained by considering only the endogenous effects in the cold and warm seasons were 0.79 (0.66) and 0.87 (1.38 °C), respectively. Considering the simultaneous effect of endogenous and exogenous variables in LST downscaling, these values reached 0.83 (0.47) and 0.91 (0.99 °C), respectively. The greatest effect of landscape metrics on the accuracy of LST downscaling was associated with built-up lands. Accounting for landscape metrics in the downscaling process significantly increases the accuracy of downscaled LST.
Heat vulnerability and extreme heat risk at the metropolitan scale: A case study of Taipei metropolitan area, Taiwan
2022, Urban Climate
Global climate change has intensified extreme climate events, such as torrential rain, heavy snow, heatwaves, and droughts. Among all extreme disaster events, extreme heat might lead to long and intensified heatwaves and seriously impact human living environments and ecological habitats. This study applies remote sensing to identify the spatial variation in extreme heat and principal component analysis (PCA) to explore the spatial distribution of the heat vulnerability index in the Taipei metropolitan area. Finally, a bivariate local indicator of spatial association (bivariate LISA) is used to explore the relationship between extreme heat and the heat vulnerability index. The results show that most vulnerable villages are clustered in downtown Taipei. In addition, the bivariate LISA results show an aggregation of 346 high-high clusters, indicating extreme heat and relatively high vulnerability clustered in particular areas. The findings of this study could provide the basis for proposing adequate resilience strategies for areas suffering from extreme heat based on the heat vulnerability composition.

View all citing articles on Scopus

View full text

Downscaling of thermal images over urban areas using the land surface temperature–impervious percentage relationship

Abstract

Highlights

Introduction

Section snippets

Study area and data

Sharpening experiment

Regression analysis

Conclusions

Acknowledgements

Remote Sensing of Environment

Remote Sensing of Environment

International Journal of Applied Earth Observation and Geoinformation

Applied Geography

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Utility of thermal sharpening over Texas high plains irrigated agricultural fields

Journal of Geophysical Research

The built environment induced urban heat island effect in rapidly urbanizing arid regions – a sustainable urban engineering complexity

Environmental Sciences