Remote sensing land surface temperature for meteorology and climatology: a review

Abstract

The last decade has seen a considerable increase in the amount and availability of remotely sensed data. This paper reviews the satellites, sensors and studies relevant to land surface temperature measurements in the context of meteorology and climatology. The focus is on using the thermal infrared part of the electromagnetic spectrum for useful measurements of land surface temperature, which can be beneficial for a number of uses, for example urban heat island measurements. Copyright © 2011 Royal Meteorological Society

1. Introduction

Remote sensing is defined as ‘the science and art of obtaining information about an object, area, or phenomenon through the analysis of data acquired by a device that is not in contact with the object, area or phenomenon under investigation’ (Lillesand et al., 2004). The general term originated in the 1960s at a similar time to the launch of the first meteorological satellite, the Television InfraRed Observation Satellite (TIROS‐1). Usage is growing within the fields of meteorology and climatology, and works in unison with the use of Geographical Information Systems (GIS) (Chapman and Thornes, 2003; Dyras et al., 2005) for spatial analysis. Techniques can provide increased spatial coverage when compared to weather station data (Mendelsohn et al., 2007) and the instantaneous observations, global coverage and improving quality of remotely sensed information is proving increasingly useful (Jin and Shepherd, 2005). Remote sensing offers the ability to work at a number of scales, from local/citywide (Tomlinson et al., in press), national (Imhoff et al., 2010) and worldwide (Jin, 2004). Regardless of the scale of the study, remote sensing offers an opportunity to provide a consistent and repeatable methodology, suited equally to both quick pilot studies as well as long term monitoring campaigns. Although the initial cost of remote sensing platforms is high, the ease of data availability to end researchers, combined with the often extensive temporal and spatial coverage available, offers a marked improvement to traditional fieldwork campaign studies.

This review looks at remote sensing as a tool for meteorology and climatology, with a particular focus on using remotely sensed data to calculate land surface temperature (LST). In this field, the urban heat island (UHI) is a well‐documented phenomenon (see reviews by Arnfield (2003), Rizwan et al. (2008) and Stewart (2010)) whereby the climate is unintentionally modified, causing urban areas to be warmer than surrounding rural areas. The UHI was first investigated through satellite techniques in the 1970s (Matson et al., 1978; Price, 1979), but the field is constantly advancing as new developments in technology (increases in sensor resolution, satellite availability, global coverage, verification methods) and increased understanding of scientific processes come together. Exploration of the UHI effect via satellite techniques is the primary focus of this review and specific studies will be discussed under relevant sensor headings. Other uses, such as calculating cooling degree‐days (Stathopoulou et al., 2006) or monitoring heatwaves (Dousset et al., 2010), the impact of urban development on runoff (Herb et al., 2008) and soil surface moisture (Petropoulos et al., 2009) have also been successfully demonstrated. Remotely sensed data can be a useful resource for the modelling community; helping to define input data such as short wave net radiation for land surface models (Kim and Liang, 2010), or increasing the utility of surface energy balance (Senay et al., 2007) and other climate models (Jin et al., 2007). A number of reviews exist in this general area. For example, see Kidd et al. (2009) for an excellent general overview of satellite meteorology and climatology at the start of the twenty‐first century. With respect to LST, other reviews have covered satellite remote sensing of the UHI (Gallo et al., 1995), the physics, methods and theoretical limitations of LST retrieval (Dash et al., 2001) and Thermal InfraRed (TIR) remote sensing (Prata, 1994; Voogt and Oke, 2003; Weng, 2009). This review differs from other articles as it details multiple sources of data (including timing and availability). It is written with a meteorologist in mind rather than a remote sensing expert so as such it purposefully does not detail software (either commercial or open source) or in‐depth techniques required to use the datasets described.

2. Derivation of land surface temperature

This section outlines the theory behind deriving LST from remote sensing techniques, and covers some fundamental details that need to be understood if data are to be used accurately and usefully for sensing the weather. If more detailed information is required, the physics behind deriving LST is explained in more detail in Dash et al. (2002). Several textbooks are also available (e.g. Lillesand et al., 2004). Alternatively, the specification documents of individual sensors or platforms can be inspected (see links in Table II).

A fundamental requirement for remote sensing is the detection of electromagnetic radiation (EMR) by sensors on a remote sensing platform. This is useful as different objects emit EMR in different ways, so the spectral response can be analysed. Within the EMR spectrum (Figure 1), the wavelength of most use for LST measurements is the thermal infrared (TIR), between 8 and 15 µm. However, one exception to this is passive microwave which has been used for LST measurement in China (Chen et al., 2010), USA (McFarland et al., 1990), Canadian sub‐arctic (Fily, 2003) and indeed globally (Peterson et al., 2000; Williams et al., 2000). Passive microwave measurements tend to be limited in the sense that they typically offer a very coarse resolution (in the tens of kilometres). For this reason, this review will focus on TIR sensors, which are more commonly used and offer higher resolution data.

Figure 1

The electromagnetic spectrum arranged by wavelength. Thermal infrared highlighted in bold. Adapted from Lillesand et al. (2004)

Satellite TIR sensors receive EMR which can be quantified in the form of measurements of Top Of Atmosphere (TOA) radiances. This includes upwelling radiance emitted from the ground, upwelling radiance from the atmosphere, and the downwelling radiance emitted by the atmosphere and reflected from the ground. During the day there is both emission and reflection of EMR, but during the night sensed EMR is restricted to only emission. The inverse of Planck's law (the energy emitted by a surface is directly related to its temperature) is used to derive blackbody/brightness temperatures from TOA radiances. TOA radiances are then converted to LST by correcting for three main effects; atmospheric attenuation, angular effects and spectral emissivity values at the surface. Atmospheric attenuation (absorption, reflection or refraction and scattering) will alter the EMR as it passes through the atmosphere, resulting in differences between TOA radiances and LST. Within TIR wavelengths, most attenuation is due to water vapour and aerosols. Angular effects are a product of the variety in viewing angles resulting in wavelength shifting which must be compensated for when estimating radiances (Dash et al., 2002). Spectral emissivity refers to the relative ability of a surface to emit radiation and can be highly variable due to the heterogeneity of land, and is influenced by surface cover, vegetation cover and soil moisture. Quantification of emissivity is achieved by considering the ratio of energy emitted by a surface with respect to the energy emitted by a black body at the same temperature. However, calculations are complicated because natural surfaces do not behave like a black body and thus need correction using typical emissivity values (Table I). These corrections are done through complex algorithms, alongside extensive validation and verification, resulting in a final product that can be used by a meteorologist.

Table I. Typical emissivity values of common materials (Lillesand et al., 2004)

Material	Typical average emissivity (over 8–14 µm)
Wet snow	0.98–0.99
Healthy green vegetation	0.96–0.99
Wet soil	0.95–0.98
Brick	0.93–0.94
Wood	0.93–0.94
Dry vegetation	0.88–0.94
Dry snow	0.85–0.90
Glass	0.77–0.81
Aluminium foil	0.03–0.07

Orbital satellite remote sensing methods are limited by image acquisition time which is set by the orbital characteristics of the relevant satellite and means that readings at specific times cannot be obtained or requested unless they match the orbit. Geostationary satellites, which stay in the same position relative to the Earth, offer a greatly increased temporal resolution at the expense of reducing spatial resolution and coverage area. Examples of sensors on geostationary platforms covered in this review include GOES and SEVERI sensors. However, not all images may be accurate, as high zenith angles result in a lengthened atmospheric path that can result in less accurate images (Streutker, 2003). Many images come with additional metadata (such as quality control scientific data sets) that can help recognize this problem. It is also worth noting that not all images are readily available, despite orbital paths. Archives may be corrupt, or the satellite may have been offline or manoeuvering in such a way that meant observations were not collected. Hence, if a study has a specific temporal requirement it can therefore be useful to check multiple potential sources. Choice of image timing is also important. For example, Rigo et al. (2006) found that MODIS LST was more accurate at night compared to the daytime, and the AATSR target accuracy is 2.5 K for daytime, increasing to 1 K at night time (Noyes et al., 2007). Similarly, Hartz et al. (2006) found night time ASTER images could better observe neighbourhood climatic conditions. Limitations of resolution are being investigated, and algorithms have been developed to sharpen thermal images to increase the resolution (Dominguez et al., 2011). A serious limitation of TIR satellite remote sensing techniques is the requirement for clear skies in order to derive accurate readings. Hence, cloud cover can be a serious problem. Dependent on the research requirements, composite images from multiple passes can often be created in order to construct an image without cloud cover limitations (Neteler, 2010), or algorithms can be used to estimate pixels (Jin and Dickinson, 2000). Alternatively, modelling or passive microwave remote sensing could be used (Wan, 2008) if increased coverage is required. An effect of this is that seasonal differences can influence image availability (increased cloud cover) and accuracy (increased rainfall causing wet surfaces leading to unreliable LST measurements), for example winter study periods can be more difficult (Rajasekar and Weng, 2008).

Two main algorithmic approaches are used for conversions, the radiative transfer equation (RTE) and the generalized split window technique (GSW). These techniques are explained in detail elsewhere (Dash et al., 2001; Weng, 2009) and as such are not covered in detail here. The GSW technique in the 11 and 12 µm channels is used by AATSR, AVHRR, MODIS and SEVIRI products, and in simple terms uses adjacent channels with different properties to calculate atmospheric attenuation. Nine different split window algorithms have been evaluated (Yu et al., 2008), concluding that accuracies are dependent on having reliable a priori emissivity data. This is one difficulty with remotely sensed imagery covering large areas: assumptions of average emissivity across a heterogeneous area. It is important to note that single channel products such as Landsat TM/ETM+ cannot use a GSW technique, and are therefore generally considered less accurate as they will not be correcting for atmospheric attenuation at the time of overpass, although under certain conditions single window methods can provide a reasonable estimate of LST (Platt and Prata, 1993).

The differences between satellite derived LST and ground measured air temperature is one area that is still not fully understood, and is the subject of ongoing work. Reviews (Arnfield, 2003; Weng, 2009) cite research that details both similarities between air and LST (Nichol, 1994) and differences (Weller and Thornes, 2001). Related work includes comparing LST and air temperatures over large areas and multiple ecosystems in Africa (Vancutsem et al., 2010) and using MODIS LST data to estimate air temperature in China (Yan et al., 2009).

3. Satellites and LST sensors

There is a number of different satellite remote sensing platforms with multiple sensors in the TIR spectrum, giving the modern meteorologist a number of potentially useful datasets to measure LST. Datasets are available for different time periods, at different resolutions, with varying accuracy, therefore this section outlines the various datasets available, ordered by launch date (Figure 2). Currently operating satellites are also summarized in Table II. Some comparisons between datasets exist, for example between MODIS and ASTER (Pu et al., 2006) and these are discussed as appropriate. This review will focus on satellite based sensors, as they offer global coverage and good availability. Airborne sensors (e.g. ATLAS (Gluch et al., 2006) or AHS (Sobrino et al., 2006)) can offer greater spatial and thermal resolution, but generally airborne data are only available for small areas and at significant cost to the end user. Similarly this paper does not detail private or commercial satellites, as these are generally not as accessible for researchers.

Figure 2

Timeline of satellite launches and associated sensor data availability. Data availability to 2012 indicates ongoing availability

Table II. Current LST capable sensors and satellite information

Sensor	Satellite	Spatial resolution	Orbital frequency	TIR spectral bands (µm)	Image acquisition (local time)	Data available since	Website
Landsat	Landsat	60 maa Collected at 60 m but resampled to 30 m.	16 days	(6) 10.4–12.5	∼1000	1999bb Landsat 7 ETM+ data from 1999, TM data from Landsat 4 and 5 available since 1982 at 120 m spatial resolution.	http://pubs.usgs.gov/fs/
ETM+	7						2010/3026/
							http://landsat.gsfc.nasa.gov/
MODIS	Aqua	∼1 km	Twice daily	(31) 10.78–11.28	∼1330	2002	http://modis.gsfc.nasa.gov/
				(32) 11.77–12.27	∼0130		https://lpdaac.usgs.gov/lpdaac/
							products/modis_overview
MODIS	Terra	∼1 km	Twice daily	(31) 10.78–11.28	∼1030	2000	http://modis.gsfc.nasa.gov/
				(32) 11.77–12.27	∼2230	2000	https://lpdaac.usgs.gov/lpdaac/
							products/modis_overview
ASTER	Terra	90 m	Twice daily	(10) 8.125–8.475	Request	1999	http://asterweb.jpl.nasa.gov/
				(11) 8.475–8.825	only		index.asp
				(12) 8.925–9.275
				(13) 10.25–10.95
				(14) 10.95–11.65
AVHRR	Multiple	∼1.1 km	Twice daily	(4) 10.3–11.3	dd AVHRR is carried on > 10 NOAA satellites; see http://ivm.cr.usgs.gov/tables.php for full orbital details of each.	1979	http://noaasis.noaa.gov/
	NOAA			(5) 11.5–12.5cc AVHRR/3 characteristics.			NOAASIS/ml/avhrr.html
							http://eros.usgs.gov/#/Find_
							Data/Products_and_Data_
							Available/AVHRR
AVHRR	MetOP	∼1.1 km	29 days	(4) 10.3–11.3	∼0930	2006	http://www.esa.int/esaLP/ESA7
				(5) 11.5–12.5			USVTYWC_LPmetop_0.html
AATSR	Envisat	∼1 km	35 days	11	∼1000	2004ee LST product currently available since 2004. Planned application to historical data will result in data from 1991 onwards.	http://envisat.esa.int/
				12			instruments/aatsr/
SEVIRI	Meteosat‐8	∼3 km	Geostationary	10.8	Every	2005	http://landsaf.meteo.pt/
				12	15 min
GOES	GOES	∼4 km	Geostationary	(4) 10.2–11.2	Every 3 h	1974	http://goespoes.gsfc.nasa.gov/
Imager	networkff Status of network available: http://www.oso.noaa.gov/goesstatus/.			(5) 11.5–12.5	(full disc)		goes/index.html

• a Collected at 60 m but resampled to 30 m.
• b Landsat 7 ETM+ data from 1999, TM data from Landsat 4 and 5 available since 1982 at 120 m spatial resolution.
• c AVHRR/3 characteristics.
• d AVHRR is carried on > 10 NOAA satellites; see http://ivm.cr.usgs.gov/tables.php for full orbital details of each.
• e LST product currently available since 2004. Planned application to historical data will result in data from 1991 onwards.
• f Status of network available: http://www.oso.noaa.gov/goesstatus/.

3.1. AVHRR

The Advanced Very High Resolution Radiometer (AVHRR) sensor has been on a number of National Oceanic and Atmospheric Administration (NOAA) satellites and is currently operational on NOAA‐15,‐16,‐17,‐18 and 19, offering at least daily coverage, but restricted to daytime images. The spatial resolution is ∼1.1 km and LST is derived from TIR channels 4 (10.3–11.3 µm) and 5 (11.5–12.5 µm), with a global dataset provided through the sun‐synchronous orbit. Data are available from the NOAA Comprehensive Large Array Stewardship System (http://www.nsof.class.noaa.gov/saa/) and the High Resolution Picture Transmission software (http://www.satsignal.eu/software/hrpt.htm) can be useful for analysis. MetOP, the EUMETSAT satellite platform, also has an AVHRR sensor with an orbital repeat time of 29 days. Comparative studies of AVHRR algorithms exist which offer more details (Ottle and Vidal‐Madjar, 1992; Vázquez et al., 1997).

A strength of the AVHRR sensor is that there is a relatively long historical record of data, and correspondingly a significant body of research that has used the sensor for many different uses. A notable use of AVHRR data has been in the creation of an 18 year (1981–1998) diurnal LST dataset (Jin, 2004) at 8 km resolution globally for snow free land surfaces. It gives monthly diurnally‐averaged, minimum and maximum skin temperatures. This long term record is not possible with most other sensors as the historical data are not available, as the satellites and sensors were not developed or in space. Matson et al. (1978) used VHRR (the forerunner to AVHRR) data for UHI analysis of the US, detecting over 50, and LST investigations in Northern Italy used AVHRR (Ulivieri and Cannizzaro, 1985). Other studies using AVHRR include Gallo et al. (1993) who investigated the surface temperature and vegetation index for 37 cities in the United States, particularly noting the consistent nature of the data when studying UHI. Lee (1993) used AVHRR to study the UHI in South Korea and more recently AVHRR data have been used to study the growth of the UHI in Houston, Texas, USA between 1985–1987 and 1999–2001, with the results showing a growth in magnitude of 35%, and a growth in area between 38 and 88% depending on method (Streutker, 2003). Stathopoulou and Cartalis (2009) used AVHRR data from Greece and applied downscaling techniques to increase the output resolution (1 km > 120 m), helping to address the inevitable balancing between spatial and temporal resolution. A significant weakness of AVHRR includes the lack of availability of night time images.

3.2. Landsat

The Landsat series of satellites are probably the most well known, with the longest record of Earth observations from space. The Thematic Mapper (TM) on Landsat 4 and 5 had a visible resolution of 30 m and a TIR resolution of 120 m (band 6, 10.4–12.5 µm). Landsat 4 and 5 are no longer continually collecting data, but Landsat 7's Enhanced Thematic Mapper (ETM+) collects thermal data at a 60 m resolution (also band 6, 10.4–12.5 µm). Landsat 7 has a near polar Sun‐synchronous orbit with a revisit time of 16 days, meaning that a given point on Earth should be imaged at approximately the same local time (∼1000 h) every 16 days. The ETM+ offers some of the highest resolution thermal resolution measurements from space, and data are available freely from the U.S. Geological Survey (USGS) (http://earthexplorer.usgs.gov/or http://glovis.usgs.gov), however data from 2003 onwards are impaired due to failure of the scan line corrector. This results in only ∼80% of each scene being captured. The Landsat data archive has only been freely available since 2008, therefore the number of studies has increased in recent years. A disadvantage of data from Landsat is that they are not collected at night, and the thermal calibration is limited. More details on the Landsat project is available (http://pubs.usgs.gov/fs/2010/3026) and the Landsat Data Continuity Mission (LCDM) aims to continue the long term Landsat record.

In the USA, Aniello et al. (1995) used Landsat TM data to help map micro UHIs (hot spots within a city) in Dallas, Texas, USA by combining both the thermal band (6) and extracted tree cover data from an unsupervised classification. One satellite image was used and the results showed that micro UHIs were highest in the centre and were generally resulting from a lack of tree cover. Weng et al. (2004) use Landsat ETM+ to link LST to Normalized Difference Vegetation Index (NDVI) in Indianapolis, USA which resulted in results linking LST to different land cover types and Xian and Crane (2006) use both Landsat TM and ETM+ to explore the thermal characteristics of urban areas in Tampa Bay and Florida, USA finding that land use and land cover fundamentally affect the thermal results. Weng (2003) used three Landsat TM images (from 1989, 1996 and 1997) to study the UHI in Guangzhou, China alongside fractal analysis with the result that showed two significant heat islands existed in the city. Further work has been done in China (Chen et al., 2006; Li et al., 2009), including combining Landsat ETM+ with computational fluid dynamic (CFD) modelling in Wuhan, China (Li and Yu, 2008). The combination of remote sensing and modelling was found to be mutually complementary. In Europe, Stathopoulou and Cartalis (2007) used Landsat ETM+ data to explore the daytime UHI across the major cities in Greece using a method that incorporates the CORINE land cover classification to superimpose land cover based emissivity values to create a mean surface temperature by land cover.

Resampling (generally using the nearest neighbour algorithm) the thermal band to lower resolutions (e.g. 30 m to match the visible spectrum) is a common technique (Weng, 2003; Weng et al., 2004; Xian and Crane, 2006; Cao et al., 2010) in order to simplify analysis. Landsat has a great strength in terms of spatial resolution, however its 16 day revisit time and lack of night time image acquisition is limiting at the temporal scale. Stathopoulou and Cartalis (2007) discusses how future studies may focus on a time series of images as the UHI strongly depends on synoptic weather conditions. The spatial resolution of 60 m on Landsat ETM+ does allow individual hotspots to be picked out (Aniello et al., 1995; Stathopoulou and Cartalis, 2007) and work is still using the ETM+ sensor (Boudhar et al., 2011).

3.3. GOES

The Geostationary Operational Environmental Satellite (GOES) system is a network of geostationary satellites (status available: http://www.oso.noaa.gov/goesstatus/) carrying the GOES Imager, a multispectral instrument offering two channels in the TIR (10.2–11.2 and 11.5–12.5 µm) with an at nadir resolution of ∼4 km. GOES related studies discuss algorithm development for dual thermal channel sensors (e.g. on GOES‐8 and ‐10) (Sun, 2003) and single thermal channel sensors (e.g. GOES M‐Q) (Sun et al., 2004). An evaluation of GOES LST retrievals over the USA is given by Pinker et al. (2009). An illustration of an advantage of geostationary satellites is shown by Sun et al. (2006), which measures the diurnal temperature range across the USA, possible due to the high temporal availability of data. An interesting study links MODIS data as a calibration source for GOES data, resulting in a 1 km LST dataset at half‐hourly temporal resolution and a measured accuracy better than 2 °C (Inamdar et al., 2008).

3.4. MODIS

The MODerate resolution Imaging Spectroradiometer (MODIS) sensor is carried on both NASA's Aqua and Terra satellites that have near polar orbits resulting in two images per satellite per day. Image acquisition on Aqua is ∼1330 and 0130 h and Terra is ∼1030 and 2230 h, all local time. This is a high temporal resolution, and the spatial resolution is ∼1 km. Data are available from the USGS Land Processes Distributed Active Archive Center (https://lpdaac.usgs.gov/) and useful LST products include MYD11A1 (Aqua) and MOD11A1 (Terra) which are the daily LST and emissivity at 1 km. Other products include 8 day 1 km data (M*D11A2) and others. These LST products primarily use TIR bands 31 (10.78–11.28 µm) and 32 (11.77–12.27 µm) combined with split window algorithms (Wan and Dozier, 1996) which multiple studies have tested (Wan, 2002, 2008; Wan et al., 2004; Coll et al., 2005) with results suggesting accuracies greater than 1 K over homogeneous surfaces. A useful tool for processing data in ESRI ArcMap is the Marine Geospace Ecology Tools (MGET) plugin (Roberts et al., 2010), or the standalone MODIS Reprojection Tool (https://lpdaac.usgs.gov/lpdaac/tools/modis_reprojection_tool).

There is a number of studies that use MODIS LST data within the urban climatology fields. Within Europe, Pongrácz et al. (2010) explored the UHI of nine central European cities and find that the most intense UHI occurs during daytime in the summer. The summer UHI of Birmingham has been analysed (Tomlinson et al., in press) and work has looked at the 10 most populated cities of Hungary (Pongrácz et al., 2006). Studies in Bucharest used MODIS to calculate the UHI in summer months (Cheval and Dumitrescu, 2009) and under heatwave conditions (Cheval et al., 2009). Globally, Hung et al. (2006) quantified the UHI in eight Asian mega‐cities using MODIS data, Jin et al. (2005) analysed various cities including Beijing and New York, and Imhoff et al. (2010) used MODIS data averaged over 3 years to calculate UHIs across the United States.

MODIS data have been used extensively outside the UHI field. Other surface measurements include observing the impacts of agriculture on rural surface temperatures in North America (Ge, 2010) and measuring water temperature and heat flux over a hydroelectric reservoir in Brazil (Alcântara et al., 2010). Atmospheric studies estimate aerosol optical depth (an important influence on the radiation budget) in America, Canada, China and Africa (Liang et al., 2006), and help detect clear sky, low level temperature inversions in the polar regions (Liu and Key, 2003). In cooler areas, MODIS has been used for frost risk assessment in Bolivia (Pouteau et al., 2010) and permafrost monitoring in Siberia (Langer et al., 2010). Outside of the meteorology domain, MODIS data have been used to help epidemiological studies of tick‐borne diseases (Neteler, 2005) and more. A strength of the MODIS sensor is the compromise between regular image acquisition and reasonable spatial resolution, in comparison to other sensors that offer higher spatial resolution but lower temporal resolution (e.g. Landsat), or higher temporal resolution but lower spatial resolution (e.g. SEVIRI).

3.5. ASTER

The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) operates at a very high resolution (90 m), and calculates surface temperature (AST08 product—http://asterweb.jpl.nasa.gov/content/03_data/01 _Data_Products/SurfaceTemperature.pdf) using the Temperature Emissivity Separation (TES) algorithm (Gillespie and Rokugawa, 1998). ASTER has five TIR bands, and full technical details are available in Yamaguchi et al. (1998). ASTER is based on the NASA Terra satellite platform, but is fundamentally different from other sensors discussed in this review in that it is request only, with fees payable for data. Hence, data are only acquired if a specific request has been detailed and paid for, and therefore the historical data are limited and costly. This is a significant restriction, given the difficulties of ensuring suitable atmospheric and weather conditions for a specific future request, and obviously limits historical studies. However, the 90 m resolution is high, only comparable with Landsat when considering the spatial scale, and ASTER has the potential for better temporal coverage, given the Terra satellite has a twice daily pass.

ASTER images have been used for a number of studies. They were used to compare LST to urban biophysical descriptors (such as impervious surface, green vegetation and soil) in Indianapolis, USA through linear spectral mixture analysis and multiple regression models, with the results that impervious surfaces and hot objects were positively correlated with LST, whereas vegetation and cold objects were negatively correlated (Lu and Weng, 2006). An ASTER image was used alongside a 148 km vehicle traverse of Hong Kong in order to compare air and remotely sensed temperatures (Nichol et al., 2009) and ASTER (for thermal use) and IKONOS data (for high resolution (4 m) visible and near infrared use) were combined to explore the cooling effect of urban parks in Nagoya, Japan (Cao et al., 2010).

There are frequent comparisons between ASTER and MODIS data, for example in verification. This is because ASTER and MODIS are complementary in scale (∼1 km and 90 m) and are based on the same satellite platform, so image acquisition occurs at the same time, height and location which aids comparison. Land surface emissivity and radiometric temperatures have been compared with good agreement over desert in the USA and savannah in Africa (Jacob et al., 2004). Direct comparisons between three correction approaches over the Loess Plateau in China have reduced the discrepancies between ASTER and MODIS data (Liu et al., 2007). Long term ground based long wave radiation between 2000 and 2007 have been compared to ASTER and MODIS images for both LST and emissivity (Wang and Liang, 2009).

3.6. AATSR

The Advanced Along Track Scanning Radiometer (AATSR) is carried onboard the European Space Agency (ESA) ENVironment SATellite (ENVISAT) which was launched in 2002. This was the third instrument in a series (ATSR‐1 and ATSR‐2) which started with the Along Track Scanning Radiometer (ATSR‐1) in 1991. The primary objective of all missions to date has been for sea surface temperature (SST) collection. ENVISAT is in a Sun‐synchronous polar orbit with a 35 day repeat cycle, which means data availability is lower than others. The LST product is relatively new, being operational from March 2004 for data from the AATSR, and the TIR bands 11 and 12 µm are used to provide LST at ∼1 km resolution. However the algorithms developed will be applied to historical data from the previous sensors (ATSR‐1 and ATSR‐2) resulting in an LST dataset starting in 1991, although the timeline for completion is unknown. The AATSR literature is primarily concerned with the theoretical science for algorithm development (Prata, 2002), evaluation of algorithms (Sòria and Sobrino, 2007) or validation (Coll et al., 2005, 2009; Noyes et al., 2007). AATSR has been used for monthly LST mapping over Europe (Joan and Cesar, 2009) and more broadly for drought prediction (Djepa, 2011), estimating evapotranspiration (Liu et al., 2010) and detection of snow covered areas (Istomina et al., 2010). In the future more studies using AATSR can be anticipated, although the long orbital repeat cycle means other sensors may be better suited.

3.7. SEVIRI

The Spinning Enhanced Visible and Infrared Imager (SEVIRI) is an instrument on Meteosat‐8 that uses a generalized split window algorithm (detailed in Sobrino, 2004) to calculate LST from two thermal channels (10.8 and 12 µm). The satellite application facility on land surface analysis (http://landsaf.meteo.pt/) is responsible for generation and archiving of the data. Meteosat Second Generation (MSG) is a geostationary satellite so therefore has different characteristics to other orbital satellites this review has examined. It has a very high temporal resolution of 15 min (theoretical maximum of 96 images per day) but the area covered is constant and not global. All the land pixels within the Meteosat disc that are below a 60° viewing angle are processed for LST measurements, to avoid excessive atmospheric attenuation and reduced accuracy at higher angles. This results in a spatial pixel resolution of 3 km at nadir (increasing to ∼6 km at > 60°). Schmetz et al. (2002) offer a useful introduction to the MSG instrument. The high temporal resolution has a number of advantages, namely it has a much greater chance of getting cloud free images of a study area due to the number that are taken and it enables the potential to study the diurnal LST pattern. Meteosat data have been available since July 2005 for the complete Meteosat disc (February 2005 for Europe).

Trigo et al. (2008) compare Meteosat LST with MODIS LST over three locations and find that Meteosat temperatures are warmer than MODIS, particularly in the daytime. A comparison between MODIS and Meteosat LST has also been carried out focussing on the heatwave in Athens, Greece during July 2007 (Retalis et al., 2010) and the results show significant correlation both between each other and between air temperature measurements, which agrees with other air temperature and Meteosat LST comparisons that also perform well (Nieto et al., 2011).

Due to the high temporal resolution, it is theoretically possible to study the diurnal UHI. In practice this is limited by cloud cover, however recent work outlines a methodology for reconstructing cloud contaminated pixels (Lu et al., 2011) that allows the diurnal variation to be studied in detail. In other fields this high temporal resolution is useful, for example for hazard modelling such as near real time forest fire monitoring (Umamaheshwaran et al., 2007).

4. Future developments

The future for remote sensing LST retrievals is focussed on two main areas, that of improved or replacement physical sensors and platforms, and that of improvements in data manipulation of current, historical and future data. In terms of data manipulation there is potential for improved algorithms, for example improved cloud masking or emissivity calculations. These will rely on ongoing validation and testing across a variety of landscapes and sensors, and could improve existing as well as future data.

Regarding the near future of sensors and satellite platforms, a number of relevant projects are in development. The Landsat Data Continuity Mission (LDCM) (http://ldcm.nasa.gov/or http://pubs.usgs.gov/fs/ 2007/3093/) intends to continue the long Landsat data series, and is due to be launched in December 2012 with 120 m resolution in two thermal channels. The European Space Agency (ESA) Sentinel‐3 satellites are planned for launch from 2013, offering a Sea and Land Surface Temperature Radiometer (SLSTR) with a 1 km resolution in the thermal channels and a daily revisit time. The geostationary GOES‐R satellite is due in 2015, with a 2 km resolution in the thermal channels from a new Advanced Baseline Imager (ABI) (Yunyue et al., 2009). The National Polar‐orbiting Operational Environmental Satellite System (NPOESS) is due to launch in 2016, designed to replace NASA's Aqua, Terra and Aura satellites and offering the Visible and Infrared Imagery Radiometer Suite (VIIRS) sensor for LST. An interesting sensor in development is the Hyperspectral InfraRed Imager (HyspIRI) from NASA that is hopefully planned for launch in 2015, offering a ∼60 m resolution in the thermal bands and a repeat cycle of 5 or 16 days. This is still in a planning phase and more details are available online (http://hyspiri.jpl.nasa.gov/) but this offers the next generation of space based thermal sensors. Coupled with these large ‘traditional’ missions, in the future there is likely to be an increase in ‘small satellites’ (Sandau et al., 2010) that enable relatively quick and inexpensive missions, which could for example help to observe dynamic weather systems. Future increases in spatial resolution of sensors combined with the high temporal resolution that geostationary platforms can provide is likely to offer the most useful data, however this offers considerable scientific challenges.

5. Conclusion

This review has given an overview of remote sensing techniques, sensors and research of interest to the meteorological and climatological community for LST detection and monitoring. It is clear that the focus of research has been surrounding the UHI phenomenon, but a significant research gap still exists which is the quantification of the relationship between measured air temperatures and remotely sensed LST data. Indeed, as Nichol et al. (2009) state this ‘remains the greatest unknown in remotely sensed studies of heat islands’, and this statement is still applicable to any study using LST data as a proxy for air temperature. The importance of being able to relate LST to air temperature is especially important when such datasets are being used to inform policy decisions or communicate outside of the scientific community.

A significant advantage of remote sensing data and techniques is their truly global coverage and scope, but despite this there is a low number of studies focussing on many geographical areas, and a limited number that integrate additional ground data. Remote sensing techniques offer access to data that would otherwise be unobtainable, therefore the requirement for defensible verification and accuracy measurements is considerable. Alongside this, the increasing need for data and intensifying analysis will necessitate using remote sensing data alongside other datasets from numerous sources, resulting in an integral role for remote sensing techniques within the meteorological and climatological communities.

Acknowledgements

This research has been funded by a Doctoral Training Award issued by the Engineering and Physical Sciences Research Council and supported by Birmingham City Council.

List of Acronyms

• ATLAS Advanced Thermal and Land Applications Sensor

• ATSR Along Track Scanning Radiometer

• AVHRR Advanced Very High Resolution Radiometer

• CFD Computational Fluid Dynamics

• EMR Electromagnetic Radiation

• ENVISAT ENVIronment SATellite

• ESA European Space Agency

• ETM+ Enhanced Thematic Mapper+

• EUMETSAT European Organisation for the Exploitation of Meteorological Satellites

• GIS Geographical Information Systems

• GOES Geostationary Operational Environmental Satellite

• GSW Generalised Split Window Technique

• HyspIRI Hyperspectral InfraRed Imager

• LDCM Landsat Data Continuity Mission

• LST Land Surface Temperature

• MODIS MODerate resolution Imaging Spectroradiometer

• MSG Meteosat Second Generation

• NASA National Aeronautics and Space Administration

• NDVI Normalized Difference Vegetation Index

• NOAA National Oceanic and Atmospheric Administration

• NPOESS National Polar‐orbiting Operational Environmental Satellite System

• RTE Radiative Transfer Equation

• SEVIRI Spinning Enhanced Visible and Infrared Imager

• SLSTR Sea and Land Surface Temperature Radiometer

• SST Sea Surface Temperature

• TES Temperature Emissivity Separation

• TIR Thermal InfraRed

• TIROS‐1 Television InfraRed Observation Satellite

• TM Thematic Mapper

• TOA Top of Atmosphere

• UHI Urban Heat Island

• USGS United States Geological Survey

• VHRR Very High Resolution Radiometer

• VIIRS Visible and Infrared Imagery Radiometer Suite