Influence of emissivity angular variation on land surface temperature retrieved using the generalized split-window algorithm

https://doi.org/10.1016/j.jag.2019.101917Get rights and content

Highlights

  • Anisotropy in emissivity in the split-window channels was analyzed.

  • Influence of emissivity angular variation on temperature estimation was quantified.

  • Incorporating the directional emissivities in the split-window algorithm improves LST accuracy.

Abstract

The angular variation of land surface emissivity (LSE) is rarely considered in the split-window algorithm for retrieving land surface temperature (LST), and this can cause large uncertainties in LST retrievals. To analyze the influence of angular LSE variation on LST retrievals, we built a look-up table (LUT) of directional emissivities from the MYD21A1 LST/LSE product in the Moderate Resolution Imaging Spectroradiometer (MODIS) split-window channels. The extracted directional emissivities were then input into the MODIS generalized split-window (GSW) algorithm to substitute for the classification-based emissivities. A simulation analysis was first conducted based on the LUT. Furthermore, the LST retrievals estimated from MODIS observations using the directional emissivities were compared with those estimated using the classification-based emissivities. In-situ measurements from the US SURFRAD and China’s HiWATER networks were used to evaluate LST retrievals obtained using the two different emissivities. The results showed that angular LSE variations in the split-window channels for vegetated surfaces were generally minor during the daytime, but more pronounced during the night-time (approximately 0.005 between nadir and 60°). For barren surfaces, the angular LSE variation in the ˜12 μm channel was negligible but reached approximately 0.01 in the ˜11 μm channel. In the simulation, the influence of angular LSE variation was minor for view-zenith angles (VZA) <40°, but pronounced for VZA >40° reaching approximately 1.0 and 0.7 K at VZA 65° for barren and vegetated surfaces, respectively. In the evaluation, the LST estimated using the directional emissivities showed a higher accuracy than those estimated using the classification-based emissivities, especially over barren surfaces where the improvement reached >1 K. We conclude that angular LSE variation cannot be ignored in LST estimation using the GSW algorithm when VZA is >40°, especially over barren surfaces. The accuracy of the GSW algorithm is improved pronouncedly by using the directional emissivities extracted from the MYD21 product.

Introduction

As a driving force of energy exchange at the surface-atmosphere interface, land surface temperature (LST) plays an important role in the surface-atmosphere interactions at regional and global scales (Li et al., 2014a, b; Wan, 2014; Wang et al., 2017, Wang et al., 2008). LST is a key variable in climatological, environmental, agricultural, and hydrological studies (Anderson et al., 2016; Arnfield, 2003; Quan et al., 2014; Weng, 2009) and has been recognized as one of the high-priority parameters in the International Geosphere and Biosphere Program (IGBP) (Li et al., 2013b).

Because of the limited spatial coverage of ground measurements, satellite measurements offer the only possibility for acquiring LST at global scale with sufficient temporal and spatial resolution (Li et al., 2014a, b). Thermal infrared remote sensing provides an effective method of retrieving LST due to the direct link between thermal infrared radiance and LST through the radiative transfer equation, and has attraced much attention over the last 40 years (Li et al., 2013b; McMillin, 1975). Different algorithms for retrieving LST from thermal infrared remote sensing data have been developed, e.g. the single-channel algorithm (Jiménez-Muñoz et al., 2009; Jiménez‐Muñoz and Sobrino, 2003; Qin et al., 2001; Wang et al., 2019), the split-window algorithm (Sun and Pinker, 2003; Wan and Dozier, 1996; Yu et al., 2009), the day/night algorithm (Wan and Li, 1997), and the temperature emissivity separation (TES) algorithm (Gillespie et al., 1998; Hulley and Hook, 2011). Among these algorithms, the split-window algorithm is the most widely used one due to its simplicity and robustness (Li et al., 2014a).

Taking advantage of the different atmospheric absorption in two adjacent channels centered around 11 μm and 12 μm, the split-window algorithm has been applied to many satellite data sources, including the Advanced Very High Resolution Radiometer (AVHRR; Li and Becker, 1993), Moderate Resolution Imaging Spectroradiometer (MODIS; Wan and Dozier, 1996), Spinning Enhanced Visible and Infrared Imager (SEVIRI; Niclòs et al., 2011) and Geostationary Operational Environmental Satellites (GOES; Yu et al., 2009). In the split-window algorithm, land surface emissivity (LSE) is specified priori to solve the ill-posed inversion of LST and LSE, which is due to their coupling in the thermal infrared signal (Ren et al., 2011; Wan, 2014; Wan and Dozier, 1996). Snyder et al. (1998) developed a classification-based algorithm to estimate emissivity for LST measurements from satellite data and the method was applied to retrieve LST for MODIS using a generalized split-window algorithm (GSW) (Wan and Dozier, 1996). However, the classification-based emissivity is view-zenith angle (VZA)-independent, which is not consistent with the characteristic angular variation in LSE (Hu et al., 2019; García-Santos et al., 2015; Masiello et al., 2018; Ren et al., 2011). It has been reported that an error of 0.005 in emissivity can result in an LST error of 1 K or more (Freitas et al., 2010; Yu et al., 2008). Thus, ignoring the anisotropy of emissivity may cause large uncertainties in the subsequent LST retrieval (Ren et al., 2011; Wan et al., 2004).

To the best of our knowledge, there are only few studies of the influence of angular LSE variation on LST estimation. Wan et al. (2004) pointed out the uncertainties introduced by ignoring angular LSE variation in LST retrieval, but no details were reported on the influence of angular LSE variations. Yu et al. (2006) built a look-up table (LUT) of directional emissivity using the modified geometric project model to correct the angular LSE variation effect in LST retrievals and reported that an error of >1.4 K in the LST retrieval could arise if angular variation was not accounted for. However, the LUT was built based on the assumption that each pixel is composed of tree crown and soil background, which is an obvious generalization of the variation in land surfaces. Moreover, variables such as fraction of tree cover and tree type are not always available. Ren et al. (2011) generated a LUT of directional emissivities from the ˜5-km MOD11B1 product retrieved using the day/night algorithm and applied the LUT to LST retrievals. They found that the differences between the LST retrieved using emissivities from the LUT and those in the MOD11_L2 product were between -1 and +3 K. However, the MOD11B1 product has limitations over very warm targets due to saturation of the middle infrared bands, and can only be produced if several assumptions are met (Hulley et al., 2016). Moreover, the relatively coarse spatial resolution may influence its subsequent applications.

Based on the TES algorithm originally developed for Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) (Gillespie et al., 1998), Hulley and Hook (2011) adapted the TES algorithm to the three thermal bands of MODIS (Bands 29 centered around 8.55 μm, 31 around 11 μm, and 32 around 12 μm) and proposed the MODIS TES algorithm for the MxD21 (i.e. MOD21 and MYD21) LST/LSE products, retrieving LST and LSE simultaneously. The retrieved 1-km LSE is considered to contain the angular information given that the effect of observation angles is kept in the retrieval. Thus, these emissivities in the MxD21 product can provide an approach to accounting for angular LSE variation in LST retrieval at satellite scale.

Our objective was to analyse the influence of angular LSE variation on LST retrieval, and to refine the MODIS GSW algorithm by considering angular LSE variation. To achieve this, a LUT containing directional emissivities of different land surface types in the two split-window channels was built using the MYD21A1 product. The directional emissivities in the LUT were then used in the GSW algorithm to substitute for the classification-based emissivities. A simulation analysis was conducted based on the LUT. Furthermore, LST estimated from MODIS observations using the GSW algorithm with the directional emissivities were compared with those estimated with the classification-based emissivities. In-situ measurements from the US surface radiation budget (SURFRAD) network and the Heihe Watershed Allied Telemetry Experimental Research (HiWATER) automatic meteorological stations in northwestern China were used to evaluate the LST retrievals.

Section snippets

Data and method

Considering the simultaneous retrieval of LST and LSE in the MxD21A1 products and the possible effect of crosstalk in Band 29 of the sensor onboard Terra on the MOD21A1 product, the emissivity retrievals in the MYD21A1D product were used to build the LUT of directional emissivities based on land surface types. The directional emissivities in the LUT were input to the GSW algorithm to consider angular LSE variation.

Result and analysis

Based on the selection criteria of directional emissivity curves in the LUT, 80 curves were constructed for 10 land surface types for the daytime and night-time in the four seasons over the study area. Land cover types included water, evergreen needleleaf forest, deciduous broadleaf forest, mixed forest, open shrubland, woody savannas, grassland, cropland, urban and built-up, cropland and natural vegetation mosaic in the IGBP scheme, which met the selection criteria. In addition, 6 curves were

Land cover types in the LUT

In this study, the LUT of directional emissivities was built based on the land surface types over the contiguous US and northwestern China. Following the selection criteria, 11 out of the total 17 IGBP land surface types were collected in total, encompassing the SURFRAD and HiWATER sites. The other 6 land surface types, including evergreen broadleaf and deciduous needleleaf forests, closed shrubland, savannas, permanent wetland, and snow/ice, were not included in the LUT. Moreover, only 3 out

Conclusion

We built a LUT of directional emissivities using the MYD21A1 product. The influence of angular LSE variation on LST retrievals was analyzed by simulation analysis and applications to MODIS observations.

For vegetated surfaces, the angular LSE variation in the split-window channels was generally minor (<0.005) for the daytime; while for the night-time, the angular variation was approximately 0.005. For barren surfaces, the angular LSE variation in the ˜12 μm channel was almost negligible; while

Acknowledgement

We thank the anonymous reviewers who helped to improve the manuscript. This study was supported by Open Fund of State Key Laboratory of Remote Sensing Science of China (Grant No. OFSLRSS201903), and partially supported by the National Key Research and Development Program of China (Grant No. 2018YFA0605503), partially supported by the Chinese Natural Science Foundation Projects (Grant No. 41571357 and 41871258), and partially supported by the Strategic Priority Research Program of Chinese

References (63)

  • Z.-L. Li et al.

    Feasibility of land surface temperature and emissivity determination from AVHRR data

    Remote Sens. Environ.

    (1993)
  • Z.-L. Li et al.

    Satellite-derived land surface temperature: current status and perspectives

    Remote Sens. Environ.

    (2013)
  • N.K. Malakar et al.

    A water vapor scaling model for improved land surface temperature and emissivity separation of MODIS thermal infrared data

    Remote Sens. Environ.

    (2016)
  • R. Niclòs et al.

    Accuracy assessment of land surface temperature retrievals from MSG2-SEVIRI data

    Remote Sens. Environ.

    (2011)
  • J. Quan et al.

    Multi-temporal trajectory of the urban heat island centroid in Beijing, China based on a Gaussian volume model

    Remote Sens. Environ.

    (2014)
  • H. Ren et al.

    Angular effect of MODIS emissivity products and its application to the split-window algorithm

    ISPRS J. Photogramm. Remote Sens.

    (2011)
  • D.E. Sabol et al.

    Field validation of the ASTER temperature–emissivity separation algorithm

    Remote Sens. Environ.

    (2009)
  • Z. Wan

    New refinements and validation of the Collection-6 MODIS land-surface temperature/emissivity product

    Remote Sens. Environ.

    (2014)
  • K. Wang et al.

    Evaluation of ASTER and MODIS land surface temperature and emissivity products using long-term surface longwave radiation observations at SURFRAD sites

    Remote Sens. Environ.

    (2009)
  • W. Wang et al.

    Validating MODIS land surface temperature products using long-term nighttime ground measurements

    Remote Sens. Environ.

    (2008)
  • Q. Weng

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

    ISPRS J. Photogramm. Remote Sens.

    (2009)
  • A.J. Arnfield

    Two decades of urban climate research: a review of turbulence, exchanges of energy and water, and the urban heat island

    Int. J. Climatol.

    (2003)
  • E. Borbas et al.

    Global profile training database for satellite regression retrievals with estimates of skin temperature and emissivity

    Proceedings of the Fourteenth International TOVS Study Conference

    (2005)
  • A. Chedin et al.

    The improved initialization inversion method: a high resolution physical method for temperature retrievals from satellites of the TIROS-N series

    J. Clim. Appl. Meteorol.

    (1985)
  • J. Cheng et al.

    Effects of thermal-infrared emissivity directionality on surface broadband emissivity and longwave net radiation estimation

    IEEE Geosci. Remote Sens. Lett.

    (2014)
  • J. Cheng et al.

    Estimating the broadband longwave emissivity of global bare soil from the MODIS shortwave albedo product

    J. Geophys. Res. Atmos.

    (2014)
  • J. Cheng et al.

    Estimating the optimal broadband emissivity spectral range for calculating surface longwave net radiation

    IEEE Geosci. Remote Sens. Lett.

    (2013)
  • F. Chevallier et al.

    A neural network approach for a fast and accurate computation of a longwave radiative budget

    J. Appl. Meteorol.

    (1998)
  • L. Davies et al.

    The identification of multiple outliers

    J. Am. Stat. Assoc.

    (1993)
  • S.C. Freitas et al.

    Quantifying the uncertainty of land surface temperature retrievals from SEVIRI/Meteosat

    IEEE Trans. Geosci. Remote Sens.

    (2010)
  • V. García-Santos et al.

    On the angular variation of thermal infrared emissivity of inorganic soils

    J. Geophys. Res. Atmos.

    (2012)
  • Cited by (19)

    • A retrieval method for land surface temperatures based on UAV broadband thermal infrared images via the three-dimensional look-up table

      2022, Building and Environment
      Citation Excerpt :

      Accordingly, numerous LST-retrieval methods have been proposed based on the characteristics of thermal infrared (TIR) data, which can be classified into single-channel and multi-channel algorithms [16]. One commonality between such methods is that they require the input parameters of land surface emissivity (LSE), atmospheric parameters and effective wavelength [17–19]. However, the spatial resolutions of satellite data (e.g. Landsat 8, ASTER and MODIS) [20–22] or atmospheric products (e.g. ERA5 and NCEP) [23–25] are too low to allow accurate LST-retrieval for heterogeneous urban surfaces, especially on the micro scale.

    • Investigating the effects of surface moisture content on thermal infrared emissivity of urban underlying surfaces

      2022, Construction and Building Materials
      Citation Excerpt :

      Given the influence from the higher emissivity of grasses, the overall emissivity of the permeable brick pavement exceeded other manmade underlying surfaces. Conversely, such troughs were distinctively absent for natural surfaces: the emissivity values of grasses and of shrubs were higher, given the higher SMC of their foliage (Fig. 6d and 6e), a finding consistent with the literature [56–58]. Finally, the spectra of the gravel road exhibited irregular double troughs between 8.0 and 9.7 μm, with its minimum emissivity of ∼ 0.89 at ∼ 9.3 μm (Fig. 6f).

    • Global long-term mapping of surface temperature shows intensified intra-city urban heat island extremes

      2022, Global Environmental Change
      Citation Excerpt :

      The satellite Aqua was preferred to Terra because it provides measurements in the early afternoon, approximately at the time of highest solar radiation and maximum daily surface temperature. The MYD11A1 data product was preferred over the more recent MODIS Aqua MxD21 LST product (MYD21A1, Hu et al., 2019) as it is less affected by noise, and provides more conservative estimations of temperature maxima and trends (Yao et al., 2020). The extent of built-up surfaces used to define urban and non-urban areas has been mapped by means of satellite imagery classification.

    • Thermal infrared remote sensing of vegetation: Current status and perspectives

      2021, International Journal of Applied Earth Observation and Geoinformation
      Citation Excerpt :

      In contrast, it had previously been revealed that there is no relation between canopy temperature and NDVI and that instead, the differences in canopy conductance and water stress could affect canopy temperature (Berni et al. 2009). Nonetheless, the impact of the canopy structure and view angles on canopy temperature should have been taken into account (Hu et al. 2019; Jones et al. 2009). In addition, the normalized relative canopy temperature is associated closely and significantly with canopy water content and relative water content under arid and semi-arid conditions using hyperspectral VNIR/SWIR data and thermal imaging data (Elsayed et al. 2017).

    • Cross-calibration of Chinese Gaofen-5 thermal infrared images and its improvement on land surface temperature retrieval

      2021, International Journal of Applied Earth Observation and Geoinformation
      Citation Excerpt :

      In the generalized SW algorithm, a quadratic term (Ti − Tj)2 was added in the latest version to improve the LST retrieval accuracy in wet atmospheric conditions (Wan, 2014). However, most of the current studies focused on improving the SW algorithm by parameterizing the emissivity effect and atmospheric effect (Chen et al., 2018; Chen et al., 2017; Hu et al., 2019; Tang, 2018; Wang et al., 2019) but ignoring the residuals of the SW algorithm. In the present study, the relationship between multiple land and atmospheric parameters with the LST residuals from Eq. (5) was further investigated.

    View all citing articles on Scopus
    View full text