Elsevier

Remote Sensing of Environment

Volume 140, January 2014, Pages 267-278
Remote Sensing of Environment

Modeling annual parameters of clear-sky land surface temperature variations and evaluating the impact of cloud cover using time series of Landsat TIR data

https://doi.org/10.1016/j.rse.2013.09.002Get rights and content

Highlights

  • LST temporal variations using time series of Landsat-5 TIR data and data mining

  • Effect of cloud cover on the derivation of annual temperature cycle parameters

  • Selecting only clear-sky images or removing cloudy pixels may overestimate LST by 2 K.

  • Each LST measurement divided into a long-term climatology and a day-specific anomaly

  • Spatial variation of yearly amplitude in LST more complex than that of mean annual LST

Abstract

Land surface temperature (LST) is of primary importance in understanding global environment change, urban climatology, and land–atmosphere energy exchange. Analysis of long-term remotely sensed LST data remains a challenge for researchers. Most previous studies explored urban thermal pattern over space and time using a limited number of clear-sky images or by removing cloud-contaminated pixels. The limitation in the number of image scenes prevents from deriving long-term LST climatology for a particular region. Moreover, simply eliminating cloudy pixels inevitably obscures the spatial and temporal patterns of LST. This research attempts to characterize the annual and seasonal temperature behaviors during the period of year 2000 to year 2010 in Los Angeles by employing an annual temperature cycle (ATC) model. All 115 image scenes (path 41, row 36) of less than 30% of cloud cover available from the Landsat archive were utilized for the analysis. Three ATC parameters, i.e., mean annual surface temperature (MAST), yearly amplitude surface temperature (YAST), and the phase shift, were optimized with the Levenberg–Marquardt minimization scheme to understand the annual and seasonal characteristics of LST. The overall RMSE of 7.36 K was achieved for the optimization for all 115 images; while for the median monthly composite, the RMSE reached 2.85 K. The monthly median composite partly removed day-specific anomalies and the impact of cloud cover and reflected LSTs largely under clear sky conditions, leading to the improvement in the modeling result. The MAST and YAST derived from the modeling result of the monthly median composite data were further analyzed to relate to three normalized indices, i.e., normalized difference vegetation index (NDVI), normalized difference water index (NDWI), and normalized difference built-up index (NDBI). The results showed that the mean temperature of urban areas in LA was as high as in the barren land area, reaching almost 310 K, but the urban areas possessed a less fluctuation. Seasonal analysis suggested that in winter, the urban areas observed a higher temperature than the barren land/desert, implying that urban materials held a larger amount of heat for a longer time than the barren land. The separation of cloudy scenes into cloud percentage groups made it possible for the evaluation of the effect of cloud cover on the ATC modeling process. The sensitivity analysis indicated that the inclusion of cloudy images brought about a decrease in MAST ranging from 0.18 to 2.0 K, depending on the percentage of cloud cover and the number of cloudy scenes used for the modeling. The decrease is due likely to the inclusion of cloud temperatures in analysis rather than shaded or other land surface temperatures. When all cloudy images were included in the modeling, a decrease of 2 K in MAST and an increase of 0.15 in YAST were observed. The regression analysis demonstrated that NDBI and NDVI were the main factors influencing the spatial variations of MAST and YAST with the R2 value of 0.63 and 0.49, respectively. In addition, the spatial variation of YAST was found more complex than that of MAST, since the three indices can only explain up to 53% of its variance.

Introduction

Land surface temperature (LST) and its spatial–temporal variations are vital in studies of hydrological cycles, long-term climatological changes, and land–atmosphere energy balances (Voogt & Oke, 2003). In the process of urbanization, natural landscapes are transformed into a series of built-up lands, such as buildings, roads, driveway, sidewalk, and parking lots (Xian & Crane, 2006). LST is a key component to understand the variants within this transformation. The variations of LST in urban areas have a significant impact on the local climate, as it modifies the air temperature of the lower boundary of urban atmosphere and impacts surface energy exchange and urban climate formation, and interacts with human disturbances (Voogt and Oke, 2003, Weng, 2009, Xian and Crane, 2006). Increase in the area of urban areas, especially densely built-up land, changed the land surface energy balance and brought about higher LST values in the urban than in the surrounding rural areas, contributing to the urban heat island (UHI) effect (Voogt & Oke, 2003). LST variations are found to be associated with land use and land cover (LULC) characteristics, resulting in strong heterogeneity within the urban areas. Based on linear spectral mixture analysis, Weng, Lu, and Schubring (2004) found that LST possessed a slightly stronger negative relationship with vegetation fraction than with NDVI. In addition, Yuan and Bauer (2007) demonstrated a positive linear relationship of LST with percent impervious areas. Zhang, Zhong, Feng, and Wang (2009) suggested that urban LST was closely related with the spatial distribution of vegetation patches, as well as their characteristics. The study by Callejas et al. (2011) showed that the highest variability of LST was detected in bare soils and urbanized areas. In the urban areas, residential land cover was relatively cooler than the central business district, yielding the difference in LST as high as 15 °C (Quattrochl et al., 2000). Therefore, finer spatial resolution remote sensing data are essential in estimating urban thermal patterns.

Currently, only a few space-borne sensors can deliver medium resolution thermal infrared (TIR) data required by the need to address urban LST heterogeneity and to assess UHI effect (Weng, 2009). However, satellite TIR sensors that are characterized by a relatively high spatial resolution, such as Landsat TM, ETM +, and ASTER, typically have a coarser temporal resolution. Given the long repeat-cycle of these satellites, their TIR data are not readily useful for UHI monitoring. Surface UHI is not only a phenomenon of high spatial variability, but also of high temporal variability. Therefore, researchers have also utilized data from those TIR sensors with coarse spatial but high temporal resolution from geostationary platforms (sub-hourly imagery at 3–10 km resolution), such as Geostationary Operational Environmental Satellite (GOES) and Meteosat Second Generation (MSG), for the UHI studies. Gottsche and Olesen (2001) utilized the GOES TIR data with a diurnal temperature cycle (DTC) model to describe thermal behavior of land surface, and demonstrated that the model was effective to characterize surface properties at a spatial resolution of 5 km. Realizing the low spatial resolution of GEOS LST data, Inamdar, French, Hook, Vaughan, and Luckett (2008) presented an approach to employ MODIS TIR data to disaggregate GOES data into half-hourly LST values at 1 km resolution. Gottsche and Olesen (2009) modified their DTC model to account for atmospheric attenuation of solar irradiation caused by the presence of aerosols, dusts, and clouds. Zaksek and Ostir (2012) utilized the principal component and regression analyses to downscale SEVIRI LST over central Europe to 1 km resolution per 15 min for the diurnal cycle analysis of UHI. In addition to DTC analysis, Bechtel (2012) found that modeling annual temperature cycle (ATC) was also feasible with Landsat data archive to extract the mean annual surface temperature and the yearly amplitude of surface temperature. His study in the city of Hamburg, northern Germany, collected 35 TIR images from Landsat-5 and -7, and yielded an estimation accuracy approximating 1 K. These studies provide important insights into how time series of TIR data may be utilized to examine the temporal variability of UHI without looking into the issue of urban heterogeneity in LST.

In the early studies of urban LST variations, a problem encountered by many researchers, especially those from developing countries, was the lack of remotely sensed TIR data; however, at present, the problem has been transformed into how to utilize the large amount of imagery data archive such as those from Landsat. Various data mining methods have been adopted to analyze the LST variability with time series of imagery and/or over a long time period. Rajasekar and Weng (2009) employed a Gaussian process model to explore the evolvement of UHI magnitude over space and time in Indianapolis based on an analysis of 94 images (45 for day and 49 for night). Keramitsoglou, Kiranoudis, Ceriola, Weng, and Rajasekar (2011) proposed an object-based image analysis method for analysis of MODIS LST data to characterize urban thermal patterns in the Greater Athens area, Greece. A total of 3041 images were used to reveal the magnitude, spatial extent, and maxima of UHI. Although the use of data mining methods has, so far, been limited to satellite systems such as MODIS TIR data, they may also be suitable for high spatial resolution TIR image or for high temporal resolution image data. In this research, we developed an approach to monitor UHI and to analyze LST variation in Los Angeles using time series of Landsat-5 TIR data and a data mining method. Specific objectives of the research are three fold. The first was to estimate annual temperature cycle parameters so as to analyze the annual and seasonal characteristics of LST variability over a decade between 2000 and 2010. Secondly, with the long-term Landsat data archive, this research attempted to evaluate the effect of cloud cover on the modeling of LST variations. The majority of previous UHI studies relied on clear-sky satellite observations to derive LST. The LST observed during clear-sky conditions, however, can be quite different from that experienced under cloudy conditions. Finally, the relationship between the derived LST parameters and selected biophysical indexes were analyzed to understand better the spatio-temporal variation of LST data.

Section snippets

Study area

The study area covers nearly the whole Los Angeles County, California, except for two offshore islands, the Santa Catalina Island and the San Clemente Island, and a small portion of the southern region along the coast, which is not covered by selected Landsat scene (path 41/row 36). According to the US Census 2010, Los Angeles County has a population of 9,818,605, being the most populous county in the nation. This area encompasses geographically diverse regions ranging from hilly mountains,

LST computation

Sobrino, Jimenez-Munoz, and Paolini (2004) examined three different single-channel methods for LST retrieval: the radiative transfer equation (RTE) using in situ radiosounding data; the mono-window algorithm (Qin et al., 2001); and the single-channel algorithm (Jiménez-Muñoz & Sobrino, 2003). The error sources that impact the accuracy of LST estimation with the radiative transfer equation may come from atmospheric correction, noise of the sensor, land surface emissivity, aerosols and other

Annual and seasonal temperature dynamics

Three parameters of the annual cycle, i.e., MAST, YAST, and the phase shift (theta) were optimized from LST measurements of the subset 115 images. It should be noted that initial cloud cover of less than 30% was used for acquisition of image scenes, but cloud cover was found to cover up to 65% of the study area. Fig. 3 illustrates LST observations from the remotely sensed data and the ATC fitted LST for a chosen pixel in the barren land area, with the residuals in Panel B. These deviations may

Discussions and conclusions

The study employed an ATC model to analyze the temporal variation of LSTs, which were derived from time series of Landsat TIR data in Los Angeles from 2000 to 2010. The LST data series was decomposed into a modified Fourier series, which allowed dividing each LST measurement into a long-term LST climatology (for largely cloud free conditions) and a day-specific anomaly, which can be explained by the residuals. The three parameters: MAST, YAST, and the phase shift, were then optimized using the

References (51)

  • J.A. Sobrino et al.

    Land surface temperature retrieval from LANDSAT TM 5

    Remote Sensing of Environment

    (2004)
  • J.A. Voogt et al.

    Thermal remote sensing of urban climates

    Remote Sensing of Environment

    (2003)
  • Q. Weng

    Thermal infrared remote sensing for urban climate and environmental studies: Methods, applications, and trends

    ISPRS Journal of Photogrammetry and Remote Sensing

    (2009)
  • Q. Weng et al.

    Estimation of land surface temperature–vegetation abundance relationship for urban heat island studies

    Remote Sensing of Environment

    (2004)
  • G. Xian et al.

    An analysis of urban thermal characteristics and associated land cover in Tampa Bay and Las Vegas using Landsat satellite data

    Remote Sensing of Environment

    (2006)
  • G.J. Yang et al.

    Estimation of subpixel land surface temperature using an endmember index based technique: A case examination on ASTER and MODIS temperature products over a heterogeneous area

    Remote Sensing of Environment

    (2011)
  • 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)
  • K. Zaksek et al.

    Downscaling land surface temperature for urban heat island diurnal cycle analysis

    Remote Sensing of Environment

    (2012)
  • J.A. Barsi et al.

    Validation of a web-based atmospheric correction tool for single thermal band instruments

    (2005)
  • B. Bechtel

    Robustness of annual cycle parameters to characterize the urban thermal landscapes

    Geoscience and Remote Sensing Letters, IEEE

    (2012)
  • I.J.A. Callejas et al.

    Relationship between land use/cover and surface temperatures in the urban agglomeration of Cuiaba–Varzea Grande, Central Brazil

    Journal of Applied Remote Sensing

    (2011)
  • T. Carlson

    An overview of the “triangle method” for estimating surface evapotranspiration and soil moisture from satellite imagery

    Sensors

    (2007)
  • T.N. Carlson et al.

    A method to make use of thermal infrared temperature and NDVI measurements to infer surface soil water content and fractional vegetation cover

    Remote Sensing Reviews

    (1994)
  • P.S. Chavez

    Image-based atmospheric corrections revisited and improved

    Photogrammetric Engineering and Remote Sensing

    (1996)
  • A.V. Eliseev et al.

    Amplitude-phase characteristics of the annual cycle of surface air temperature in the Northern Hemisphere

    Advances in Atmospheric Sciences

    (2003)
  • Cited by (109)

    View all citing articles on Scopus
    View full text