Opportunities and Challenges of Spaceborne Sensors in Delineating Land Surface Temperature Trends: A Review

Understanding the land surface temperature (LST) trends is crucial for policymakers and stakeholders to develop adaptation and mitigation strategies suitable for a sustainable environment coping in the face of climate change. This article presents a systematic review of the studies related to delineating spaceborne sensor-based LST trends, including information on the instruments and constellations of satellites (missions) that provide thermal infrared (TIR) and passive microwave (PMW) observations. About 99% of the studies used TIR, where 76% were Moderate Resolution Imaging Spectroradiometer (MODIS, onboard Terra/Aqua) observations. Opportunities, challenges, and research gaps for using the TIR and PMW observations were also explored, with instruments onboard either polar-orbiting or geostationary satellites. We identified that the calibrated dataset (e.g., processed, harmonized, and standardized) is extremely limited for each constellation, with multiple satellites and instruments, to make it fully useful for the entire mission period. A few problematic methodological concepts were identified, including using a few images in a longer time series. Using only a few images, acquired on different calendar months in different years, would not provide the true annual trends over the study period because they can be influenced by seasonal variations. To estimate the warming or cooling daytime, nighttime, or diurnal LST trends, the use of MODIS observations could be useful, even though it does not acquire images during the maximum or minimum temperature in a daily cycle. This article indicated further investigations into those research gaps and recommended directions to overcome most of these limitations.


Opportunities and Challenges of Spaceborne
Sensors in Delineating Land Surface Temperature Trends: A Review M. Razu Ahmed , Ebrahim Ghaderpour , Anil Gupta, Ashraf Dewan, and Quazi K. Hassan Abstract-Understanding the land surface temperature (LST) trends is crucial for policymakers and stakeholders to develop adaptation and mitigation strategies suitable for a sustainable environment coping in the face of climate change. This article presents a systematic review of the studies related to delineating spaceborne sensor-based LST trends, including information on the instruments and constellations of satellites (missions) that provide thermal infrared (TIR) and passive microwave (PMW) observations. About 99% of the studies used TIR, where 76% were Moderate Resolution Imaging Spectroradiometer (MODIS, onboard Terra/Aqua) observations. Opportunities, challenges, and research gaps for using the TIR and PMW observations were also explored, with instruments onboard either polar-orbiting or geostationary satellites. We identified that the calibrated dataset (e.g., processed, harmonized, and standardized) is extremely limited for each constellation, with multiple satellites and instruments, to make it fully useful for the entire mission period. A few problematic methodological concepts were identified, including using a few images in a longer time series. Using only a few images, acquired on different calendar months in different years, would not provide the true annual trends over the study period because they can be influenced by seasonal variations. To estimate the warming or cooling daytime, nighttime, or diurnal LST trends, the use of MODIS observations could be useful, even though it does not acquire images during the maximum or minimum temperature in a daily cycle. This article indicated further investigations into those research gaps and recommended directions to overcome most of these limitations.

I. INTRODUCTION
T EMPERATURE trends, the spatiotemporal variations of temperature over a longer period, are the key indica- tor of representing climate change [1]. While land surface temperature (LST) is dependent on the spatial variability of solar radiation and land-atmosphere heat exchange [2], the temporal variability of solar might has effects on the Earth's climate [3]. Spatiotemporal distributions of LST reflect not only the variations of climatic factors but also the characteristics of land surface [4]. It has interdependence among climate, ecosystems, biodiversity, and human societies. It can directly or indirectly impact many aspects of society by potentially disrupting the normal natural balance and forcing the change of weather patterns [5]. Increasing temperature trends (warming) and other associated factors are threatening human existence, ecological communities, and socioeconomic development across the world [6]. The Intergovernmental Panel on Climate Change (IPCC) reported that the global mean temperature had risen (global warming) by about 0.85 (±0.2) • C from 1880 to 2012 [1], [7]. Knowing the magnitude and rate of temperature change (temperature trend) would guide the formulation of appropriate levels of mitigation and adaptation strategies. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ It would help minimize the rate magnitude of the changes to keep our globe a habitable place through a sustainable environment [5].
In determining the temperature trends, the method of using in situ measurements of weather stations is the most accurate approach for the station locations. To represent trends, studies used interpolation techniques to derive the temperature for the remaining landscape [8]. The use of interpolation values in determining the temperature trends has some issues, e.g., various interpolation techniques, such as polynomial, nearest neighbor, and Fourier methods, produce different outputs even using the same input data [9], [10]. Considering this, another efficacious alternative is to use spaceborne sensor-derived LST for estimating temperature trends at regular grids, covering nearly the entire globe. A review article [11] also indicated the importance of using spaceborne Earth observational sensors by quantifying the performances and limitations of different methods used in the literature for LST trend analysis. In general, studies in the literature used the Mann-Kendall (MK) test, Sen's slope estimator (SSE), and linear regression methods to determine the trends of LST [12]. The MK test and SSE have the advantage that they can be applied independent of data distribution pattern in the time series, in contrast to the linear regression analysis whose statistical testing depends on the normality assumption [12]. Because of the nonparametric characteristic of the LST data, the MK test was mostly used to detect the trends in time series, including the SSE to determine the upward or downward directions of trend, i.e., positive or negative slope, respectively. However, the MK test may fail for time series with seasonality and may give incorrect results for shorter time series due to serial correlations [13], see also Section VII showing a simulation experiment.
Spaceborne remote sensors operating in specific wavelength ranges, including thermal infrared (TIR, 8-14 µm) and passive microwave (PMW, 0.8-75 cm or frequency 0.4-35 GHz), are useful in studying LST trends at a pleasant spatial resolution. The polar-orbiting satellites can observe both TIR and PMW, and geostationary satellites only TIR. While the datasets derived from these platforms have created opportunities for delineating the LST trends, challenges exist due to lack of long time series from a single instrument, differences in both spatial and temporal resolutions among the datasets. Here, the aim was to understand the opportunities and challenges of using remote sensing-derived LST data in delineating its trends. Hence, the specific goals were to: 1) conduct a systematic review on the LST trends reported in the literature using both TIR and PMW instruments; 2) opportunities and challenges of using the LST time series data; and 3) determine research gaps in the appropriate use of remote sensing data and summarize future research potentials.

II. SYSTEMATIC REVIEW
Web of Science (WoS) and Scopus are the two widely used citation and publication databases. Here, we used the Scopus database for the search of relevant publications, due to its broader coverage of publications with 20% more than WoS [14]. In Scopus, we performed the following search in Title-Abstract-Keyword on June 24, 2022: (LST) and ("warming trend * " or "cooling trend * " or "temperature trend * " or "positive trend * " or "negative trend * ") and not ("ocean" or "lake"). This search returned 174 documents, including 148 English articles, 12 conference papers, three book chapters, two review articles, one letter, and eight non-English articles, where the English articles were analyzed further. Among these, 31 articles did not use remote sensing data, and another 21 articles used remote sensing data but did not perform trend analysis. Therefore, we considered a total of 96 relevant articles in this study. Note that other keywords, such as spatiotemporal and cloud computing, were also searched along with LST and trend, but all results were captured by the keywords mentioned earlier. The first article was published in 2004, and the total number of articles published per year was the highest in the last four years (2019-2022) [see Fig. 1(a)]. National Oceanic and Atmospheric Administration (NOAA) polar-orbiting environmental satellites have been providing LST data since the 1980s, and thus, a steady increasing trend of using them was observed in the literature. The trend of publishing the related articles, started in 2004, may be linked to the cost-free data availability to scientists around the first decade of this century. It is likely that the LST trend related works will be published more in the coming years. Nevertheless, in terms of the location of study area, the most publications were in Asia (60), followed by North America (12) and Africa (11) [see Fig. 1(b)]. In Asia, studies were mostly conducted in China (24), the Tibetan Plateau (13), South India (14), and Iran (6), and middle eastern countries (3) [see Fig. 1(c)]. Also, the sensor-wise group of articles for TIR and PMW instruments onboard polar-orbiting and geostationary satellites is shown in Fig. 1(c), which is discussed in Sections III and IV.

C. Along Track Scanning Radiometer
Three studies, i.e., one ATSR (onboard European Remote-Sensing Satellite-ERS) and two AATSR (onboard ENVISAT-Environmental satellite), used LST time series from the European Space Agency (ESA) mission [see

D. Advanced Very High Resolution Radiometer
We found two studies that used AVHRR data from the NOAA mission [see Fig. 1(c)]. Bhatt et al. [103] used AVHRR-derived LST at 1.1 km from NOAA-7 through NOAA-18 satellites over the 1982-2015 period to study Arctic tundra vegetation productivity. They showed a positive trend for 1982-1998 during growing seasons, and 1999-2015 were positive in May-June but slightly negative for July-August. In another study, Pinheiro et al. [104] delineated warming and cooling trends over Continental Africa during the 1995-2000 period using a 4-km LST product of daily day and nighttime NOAA-14 AVHRR/2 instrument.  I  SELECTED PUBLICATIONS ALONG WITH THEIR KEY FINDINGS FOR MODIS-BASED LST TREND ANALYSIS  TABLE II  SELECTED PUBLICATIONS ALONG WITH THEIR KEY FINDINGS FOR  LANDSAT-DERIVED LST TRENDS, WHERE ALL THE PUBLICATIONS  USED LINEAR REGRESSION FOR TREND ANALYSIS E. Geostationary Instruments TIR instruments of geostationary satellites, i.e., spinning enhanced visible and infrared imager (SEVIRI) and visible infrared spin scan radiometer (VISSR) onboard Geostationary Operational Environmental Satellite (NOAA's GOES), European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT's) Meteosat Second Generation (ESA's MSG), and GMS-5 (Japanese Geostationary Meteorological Satellite, Himawari-5), respectively, were used in three studies in delineating LST trends [see Fig. 1(c)]. Underwood et al. [105] used GOES images-derived LST time series to map the evening cooling trends at 4-km spatial resolution in the Central Valley, USA, during 1997-2000. Oku et al. [106] used VISSR data at 0.1 • × 0.1 • for mapping LST trends in the Tibetan Plateau and identified that the daily minimum had risen faster than the daily maximum during 1996-2002. Zhu et al. [107] utilized SEVIRI data at 1 km during 2010-2017 to observe LST variations following an earthquake in Iran.

IV. PMW-BASED STUDIES
Only one study used PMW instruments to delineate the LST trends, i.e., Special Sensor Microwave Imager (SSM/I), onboard polar-orbiting Defense Meteorological Satellite Program (DMSP) F-series satellites [108]. They used time series data over the Tibetan Plateau to quantify standardized LST anomalies (trends) during 1987-2008 and observed that annual and monthly LST decreased by 0.5 • per decade with the highest at the Central Plateau.

V. OPPORTUNITIES AND CHALLENGES
Incorporating the variability of solar irradiance in the LST analysis is challenging, though the absolute value of total solar irradiance varies by 0.1% over the past two solar cycles [3], [109]. Performing trend analysis of any climate-related variables requires a continuous time series dataset over a longer period (typically 30 years or more). Some satellite constellations have already developed LST time series data over 30 years, such as the Landsat series, ESA, AVHRR, GOES, EUMETSAT, Himawari, and DMSP, except Terra/Aqua MODIS with 22 years. Though MODIS provides a bit shorter time series, however, its usages do not require additional preprocessing in analyzing LST trends. The design lifetime of a satellite, and its onboard instrument/s, is usually 15 years or much less. Therefore, constellations on different missions launch satellites with continuously improving instruments for better spatial and temporal coverage compared to previous generations. While continuity and improvements are important for a mission, it poses challenges to calibrating the data acquired by different instruments launched over time. It is extremely important to preprocess and harmonize the time series data so that the trends observed in the data will represent the real trends, not sensor-related or other artifacts. Due to the calibration-related challenges, we found a few studies that used observations from multiple instruments of a single mission or mixed instruments from multiple missions in delineating LST trends, rather than mostly using data from the same (or similar) instrument. Sections III and IV discuss both the opportunity and challenges of using TIR and PMW observations to delineate LST trends.

A. Polar-Orbiting TIR
In the polar-orbiting constellations, TIR instruments have the capacity to provide higher spatial resolution data, because of acquiring shorter wavelengths, compared to the PMW instruments. Besides, TIR instruments onboard polar-orbiting satellites provide better spatial resolution than the geostationary satellites due to low-altitude orbits. However, atmospheric conditions, e.g., cloud presence, often pose major challenges in acquitting LST time series. While TIR is useful for weather applications such as cloud temperature and optical properties of clouds [110], cloud-contaminated pixels in the TIR bands are critical for its application in the LST trend analysis. Moreover, the availability of cloud-free data is limited for time series analysis, even in an arid/desert climate. For the understanding of available cloud-free data, we performed a search for Landsat 5 TM (the longest lifetime of a Landsat instrument over 1984-2013) image scenes using the Fmask algorithm through Google Earth Engine (GEE). The total numbers of available Landsat 5 TM scenes for six regions-P1-P6 (in different continents with different climates and land cover)-were 300, 218, 301, 354, 356, and 432, respectively, since 1984 (see Fig. 2). Considering the acquisition of a scene every 16 days during the lifetime of TM instrument, at least ∼668 scenes should be available for each location. Moreover, the cloud-free (clear sky) scenes are much fewer in number among the available scenes, even in an arid region such as the Sahara Desert, Africa. Therefore, the limited availability of cloud-free data creates challenges for spatiotemporal analysis. Although such data gaps could be filled by applying appropriate gap-filling algorithm [111], it is potentially effective for removing small amount of cloud contamination, not the cloud-shadow effects. Another approach by NASA is to provide MODIS land products as LST composites of eight-day and monthly scales, prepared from the daily observations, to minimize cloud contamination. However, cloud-shadow pixels, including some clouds, may not be fully removed from the composites in many locations of the world. Despite the opportunities of using MODIS, noise usually affects the inversion errors strongly when the noise equivalent differential temperatures are used [112]. For instance, error bars indicated ±2 standard error for a 95% confidence interval in a monthly land temperature averaged in the Arctic during 2001-2020 from MODIS LST [113]. The good news is that the plankton, aerosol, cloud, ocean ecosystem (PACE), a NASA Earthobserving satellite mission, is scheduled to launch in 2024, where a highly advanced optical spectrometer (PACE's Ocean Color Instrument-OCI) will replace the aged MODIS [114].
NOAA's AVHRR, having the longest data records of the polar-orbiting TIR since 1979 with consistent spatial (1.1 km) and temporal (twice/day) resolutions, should probably the best option. However, we found a very limited number of studies in the literature since it would be quite challenging to generate well calibrated and harmonized time series observations of AVHRR instruments (AVHRR, AVHRR/2, and AVHRR/3) that were flown on 14 different platforms [115] because AVHRR sensors, onboard multiple platforms that have been active over the years, could have sensor degradation, scanline defects, satellite orbit drift, and channel calibration drift of the different AVHRRs [115]. Though a study successfully calibrated the observations of different AVHRR instruments during 1981-2015 over the peninsular Spain and determined the LST trends [116], however, such calibrated dataset across the globe is unavailable to our knowledge.
The Landsat series accumulated longer observations of TIR with a higher spatial resolution (60-120 m) since 1982, which also provides a better opportunity in delineating LST trends. The downside of using Landsat is its limited coverage on each scene (∼182 × 185 km) compared to the AVHRR coverage (∼6400 × 2400 km) and its temporal coverage of 16 days. To perform LST trends for a location, it is hard to find an appropriate number of cloud-free scenes over a season, even nearly impossible to get it for the same day in each year for a longer time series. We did not find, therefore, any global LST trends using Landsat TIR observations.
Another TIR instrument, namely, NASA's advanced spaceborne thermal emission and reflection radiometer (ASTER) has a similar observation period (since 2000) like MODIS with having 11 times higher spatial resolution (90 m) than MODIS, was not found in the literature for LST trend studies because it has: 1) small scene coverage, i.e., 60 × 60 km [117]; 2) not intended to continuously acquire data, and therefore, archived data are missing over time and space [117]; and 3) programed or acquired data were not free of cost from the beginning of the acquisition, which became freely available from April 1, 2016 [118].

B. Geostationary TIR
Geostationary satellite-based TIR observations have the opportunity of delineating LST trends at high temporal resolution (minutes to hours) with a reasonably high spatial resolution (2-6.9 km), because of having the longest observations in the remote sensing history, since 1975. However, the challenges are related to cloud contamination issues, and the observations of each instrument do not cover the entire globe. Because of the objective of observing a targeted part of the globe, it became a challenge to delineate the LST trends from the geostationary TIR on a global scale.

C. Polar-Orbiting PMW
The PMW observations in the longer wavelengths could be a better choice (in comparison to TIR) for not having cloud contamination [110], which has been available since 1978 through the Scanning Multichannel Microwave Radiometer (SMMR) instrument onboard Nimbus-7 Pathfinder satellite. PMW observations have been available from different instruments, such as SMMR (1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987) [119]. Data of these satellites are available as brightness temperature from different data providers, such as SMMR, SSM/I, and ASMR-E from the National Snow and Ice Data Center (NSIDC), TMI from EARTHDATA, AMSR2 from the G-Portal address of Japan Aerospace Exploration Agency (JAXA), and MWRI from the Chinese National Satellite Meteorological Center (CNSMC). Combining the continuous PMW data from the constellations/missions shows a great opportunity in delineating LST trends at the global scale. However, LST retrieval from the PMW brightness temperature is considered quite challenging due to the following reasons.
1) It is hard to attain the atmospheric brightness temperature (downwelling) from the measured brightness temperature, surface emissivity, and decoupling the LST. 2) Atmospheric correction is challenging due to the strong variations of emissivity from surface properties (e.g., soil moisture, surface roughness, and vegetation cover) in the PMW region. 3) It additionally includes subsurface temperature instead of surface (skin) temperature only. 4) Validation of LST at the pixel level is difficult to realize [119]. Moreover, it provides coarser spatial resolution pixels (typically 25 × 25 km) than TIR data because it requires more area to cover on the ground in collecting a sufficient amount of low-frequency PMW energy reaching to the satellite sensors [120]. However, spatial resolution could be enhanced to all-weather (cloud contamination free) 1 km by a spatial-seamless PMW and TIR reconstruction method [121].

D. Cloud Computing and GEE
Recently, the feasibility of cloud computing for big data analytics through GEE, one of the most popular cloud computing systems, is explored in several studies (e.g., [1]). GEE has an interactive Web-based environment, where users can observe, analyze, or download various datasets. Users can develop their own code in Python or JavaScript for various applications and run it through GEE. Techniques, such as data fusion and cloud masking as well as machine learning algorithms, can also be performed through GEE, opening an opportunity for users to explore the available datasets and investigate spatiotemporal dynamics of LST, vegetation, land cover, and so on [122]. Due to its computing infrastructure, GEE can efficiently and rapidly handle big data and extensive computations [123].

VI. RESEARCH GAPS
In the annual LST trend analysis, different statistical methods (e.g., linear regression considering the parametric, and MK test and SSE for the nonparametric data) were applied to understand the patterns and magnitudes. These methods are applicable for any long-term time series data, either observations of weather stations or remote sensing. However, in the representation of annual trends, averaging the daily or monthly data to derive an annual value has an issue because it does not reflect the seasonal trends, either increasing (warming) or decreasing (cooling). Literature showed increasing trends during summer and decreasing trends during winter in many areas of the world. Once we average the data for a year, it neutralizes the seasonal patterns and trends and, thus, misses the true warming or cooling scenario for the area. For example, an area showed trends of +1 • C (increasing) and −1 • C (decreasing) during winter and summer, respectively, that certainly show no change in the annual trend. Therefore, we need to be cautious about presenting annual trends without looking into the seasonal trends. Since the issue is pervasive in nature, researcher should investigate different scenarios. For example, Shawky et al. [5] utilized the MODIS LST products and generated time series for each calendar month since 2000 at both pixel and ecoregion levels, and then, they applied the MK trend SEE methods to estimate daytime and nighttime LST gradients over South Asia. In this way, one can observe how the LST trend is changing for each calendar month since 2000. On the other hand, at an annual scale, i.e., the average of monthly LSTs for each year, one can only observe how the overall annual LST gradient changes over the years, though it also provides useful information. To obtain the annual LSTs, all the months should be considered or at least be estimated in case of missing LST values; otherwise, the annual LST values may get biased by seasonal variations, failing to provide a reliable annual gradient estimate.
Several studies reported daytime cooling (decreasing trends) and nighttime warming (increasing trends) estimated by the MODIS LST [41], [54], [74], [75] and discussed the possible rationales. However, the trends were not derived from the time series of daily maximum or minimum temperature. In a day cycle, either clear or overcast, the maximum and minimum LSTs during the daytime and nighttime are not synchronized with MODIS image acquisition times of 1:30 or 10:30 A.M. and 1:30 P.M. or 10:30 P.M., respectively. Although the Sun is at its highest point during noon when the Earth receives the most direct sunlight (incoming solar radiation), we do not get the maximum surface temperature at that time. The highest solar energy received and absorbed by the Earth's surface at noon starts to emit thermal energy that reaches the maximum in the afternoon, typically at 3-5 P.M. Inversely, the coldest time (i.e., the lowest temperature) of a daily cycle occurs sometime after sunrise when the speed at which the Earth's emitted thermal energy is no longer greater than the incoming solar radiation. Therefore, trends (such as daytime cooling and nighttime warming) estimated from the MODIS LST may require further investigation and should be noted with probable causes and explanations. However, the patterns of LST trends may not be impacted because the image acquisition time is consistent in the time series.
Some studies used a very small number of cloud-free images (e.g., two or three Landsat images) to analyze trends over 20 or 30 years, which may not provide the true trends. It is quite impossible to get cloud-free Landsat images acquired on the same day for all the years in a time series (see Fig. 2). Therefore, in the trend analysis, a very small number of cloud-free images were mostly used, having different acquisition dates in different years. One image might be taken in early summer for a year and others during the middle or end of summer for another year. Due to the location variability of the Earth toward the Sun, the changes in the magnitude of LST occur through the seasons, and even daily and monthly within a season. Seasons on the Earth are found in the temperate zones only, which extend from 23 • 26 ′ to 66 • 34 ′ latitudes in both northern and southern hemispheres. In these zones, we usually observe four seasons, spring, summer, fall (autumn), and winter, where each season is characterized by variations in temperature, precipitation, and daylight length [124]. The longer daylight (i.e., photoperiod) usually causes the contribution of more energy to the Earth's surface and thus higher air temperature and consequently higher LST.
Therefore, considering the variable energy over a season, LST trends derived from only a few images (acquired in different dates over a season in different years) could incorporate the seasonal change in the magnitude of trends, not the true temperature trends. An example case of such erroneous trend is presented in Fig. 3. The temperature trend from only three images over the period 1984-2011 was estimated 0.08 • C/yr. It was not a true trend for the station location because in situ observations showed the trends of −0.01 • C/yr and 0.04 • C/yr for annual and monthly (July), respectively (see Fig. 3). Hence, it is suggested to use as many images as possible over the time series, at least one image in every two to three years, and image dates as close as possible in the same month of a season.
In addition, it matters where a study area is, whether in the northern or southern hemisphere. It is because two hemispheres experience the seasons at different times of the year, and the daylight length, thus solar insolation (energy), varies with the latitudes. Therefore, researchers should be cautious when comparing the LST trends for different regions located in different hemispheres and latitudes to avoid any misrepresentation of the trends in LST.

VII. FUTURE RESEARCH POTENTIAL
This article can be considered as a scope for conducting further studies to overcome the limitations and gaps for LST trend analysis. Research approaches would include developing methods of preparing a single calibrated dataset, with consistent data quality, from multiple instruments for the entire period of each constellation (mission). The seamless reconstruction of finer spatial resolution all-weather data in the entire world, using cloud contamination-free PMW at coarse and TIR at fine spatial resolution [121], would be another research direction. We could develop data fusion methods for the TIR observations of NOAA's AVHRR, MODIS, and Landsat [125] at the global scale, which would facilitate the longest time series LST data with an enhanced spatial resolution. Besides, to develop a consistent and finer spatial resolution over the history of remote sensing, data fusion of the observations in variable spatial resolution from different instruments in a constellation or across the constellations would be a major research opportunity.
Data fusion of AVHRR and MODIS observations would be the first choice due to having nearly similar spatial resolutions and, thus, likely to face less challenges. Furthermore, observations from the common period of instruments' operation, same mission or across missions, could be used for calibration and validation in developing a longer LST dataset. For example, ten years of AVHRR/3 observations could be calibrated with MODIS for a common period of 2000-2010, and observations (both AVHRR/3 and MODIS) of the remaining recent years could be used for validation. It would lead to developing a longer LST dataset since 1979 at 1.1 km from AVHRR observations. In the case of using a small number of images (e.g., Landsat scenes) with different acquisition dates for a longer period, innovative methods could be developed to standardize (normalize) each image of a year to be used in the time series.
More advanced trend analysis methods may also bypass the data availability limitation up to a certain level. For example, Ghaderpour et al. [126] and Ghaderpour [127] showed that the simultaneous season-trend fit models based on the least-squares spectral analysis has the potential of estimating trends more accurately in the presence of uncertainties due to atmospheric noise and missing data or gaps. For example, if LST data are available in certain months but limited in other months, then the simultaneous season-trend fit model can better estimate the season and trend components, whereas the traditional regression methods may overestimate/underestimate the trend. For example, we simulated a set of LST time series using the following equation: with T (t) = b + mt, S (t) = A sin (2π t), and E (t) = wgn(t, p), where b is the intercept, m is the slope, A is the amplitude, wgn is the white Gaussian noise, and p is the power of noise sample. Also, t is the time selected as monthly, i.e., the sampling rate is 12 samples per year. We applied the linear regression, MK test and its associated SSE, and anti-leakage least-squares spectral analysis (ALLSSA) to estimate the slope of such time series at a 99% confidence level. The results for a case when = 10 • C, m = 0.1 • C/yr, and A = 5 • C with randomly eliminated 30% of the samples are shown in Fig. 4. The estimated slopes by linear regression, SEE, and ALLSSA are 0.124 • C/yr, 0.122 • C/yr, and 0.111 • C/yr, respectively. We performed this experiment on one million time series such that each time series was generated by (1), where the intercept, slope, and amplitude of the sinusoid were randomly generated for each time series and a random wgn was added to each time series. Then, up to 60% of samples were randomly selected and eliminated from each time series. The root-mean-square error (RMSE) was calculated for each method as where m i is the simulated slope (the true value),m i is the estimated slope, and N is the number of those estimated slopes that were statistically significant at a 99% confidence level. The RMSEs for the linear regression, SEE, and ALLSSA were 0.0314, 0.0307, and 0.0141, respectively, showing that the simultaneous season-trend fit model ALLSSA is superior in these cases.
Note that if S (t) in (1) is the sum of multiple sinusoids of different amplitudes and frequencies and T (t) is other types of trends such as quadratic or cubic, then ALLSSA can simultaneously estimate their coefficients with a high accuracy. The discussion above clearly shows that methods themselves can make a big difference in estimating LST trends, and thus, researchers should also be cautious about which techniques they apply to process their time series.

VIII. CONCLUSION
This article explored the opportunities and challenges of using remote sensing data, as well as potential research gaps in the methodological and conceptual approaches of utilizing the data. Understanding the extent and magnitude of the LST trend is important to cope with the ongoing climate change across the world. Using the appropriate remote sensing approaches, the mapping of temperature trends would help us in deciding the adaption and mitigation strategies for environmental sustainability. It would also direct local governments across the world to understand better about the warming trend in their jurisdiction that could help them act quickly and to address any potential negative risks posed by the warming.