Simultaneous inversion of multiple land surface parameters from MODIS optical–thermal observations

https://doi.org/10.1016/j.isprsjprs.2017.04.007Get rights and content

Abstract

Land surface parameters from remote sensing observations are critical in monitoring and modeling of global climate change and biogeochemical cycles. Current methods for estimating land surface variables usually focus on individual parameters separately even from the same satellite observations, resulting in inconsistent products. Moreover, no efforts have been made to generate global products from integrated observations from the optical to Thermal InfraRed (TIR) spectrum. Particularly, Middle InfraRed (MIR) observations have received little attention due to the complexity of the radiometric signal, which contains both reflected and emitted radiation.

In this paper, we propose a unified algorithm for simultaneously retrieving six land surface parameters – Leaf Area Index (LAI), Fraction of Absorbed Photosynthetically Active Radiation (FAPAR), land surface albedo, Land Surface Emissivity (LSE), Land Surface Temperature (LST), and Upwelling Longwave radiation (LWUP) by exploiting MODIS visible-to-TIR observations. We incorporate a unified physical radiative transfer model into a data assimilation framework. The MODIS visible-to-TIR time series datasets include the daily surface reflectance product and MIR-to-TIR surface radiance, which are atmospherically corrected from the MODIS data using the Moderate Resolution Transmittance program (MODTRAN, ver. 5.0). LAI was first estimated using a data assimilation method that combines MODIS daily reflectance data and a LAI phenology model, and then the LAI was input to the unified radiative transfer model to simulate spectral surface reflectance and surface emissivity for calculating surface broadband albedo and emissivity, and FAPAR. LST was estimated from the MIR–TIR surface radiance data and the simulated emissivity, using an iterative optimization procedure. Lastly, LWUP was estimated using the LST and surface emissivity. The retrieved six parameters were extensively validated across six representative sites with different biome types, and compared with MODIS, GLASS, and GlobAlbedo land surface products. The results demonstrate that the unified inversion algorithm can retrieve temporally complete and physically consistent land surface parameters, and provides more accurate estimates of surface albedo, LST, and LWUP than existing products, with R2 values of 0.93 and 0.62, RMSE of 0.029 and 0.037, and BIAS values of 0.016 and 0.012 for the retrieved and MODIS albedo products, respectively, compared with field albedo measurements; R2 values of 0.95 and 0.93, RMSE of 2.7 and 4.2 K, and BIAS values of −0.6 and −2.7 K for the retrieved and MODIS LST products, respectively, compared with field LST measurements; and R2 values of 0.93 and 0.94, RMSE of 18.2 and 22.8 W/m2, and BIAS values of −2.7 and −14.6 W/m2 for the retrieved and MODIS LWUP products, respectively, compared with field LWUP measurements.

Introduction

Land surface parameters from remote sensing observations are critical for monitoring and modeling land surface processes and global climate change. Identified as essential climate variables, Leaf Area Index (LAI), which is widely used in agriculture and ecology, Fraction of Absorbed Photosynthetically Active Radiation (FAPAR), which expresses the energy absorption capacity of vegetation and plays a critical role in the carbon cycle, and surface albedo, which regulates energy exchange between the land surface and the atmosphere, are three essential climate variables recognized by the UN Global Climate Observing System (GCOS, 2006). Land surface emissivity, Land Surface Temperature (LST), and Upwelling Longwave radiation (LWUP) are also key parameters for studying the energy and water balance between the atmosphere and land surfaces (Z.-L. Li et al., 2013, Liang et al., 2010). Recently, many algorithms have been developed to retrieve these parameters from remote sensing datasets, including parametric or non-parametric regression methods, physically-based methods and hybrid methods (Verrelst et al., 2015), and the corresponding global land surface products are already routinely produced by various countries and organizations. Since 2000, the MODIS land team has produced global LAI and FAPAR, surface albedo, and LST products from Terra and Aqua satellite data (Justice et al., 2002). The CYCLOPES project has generated global albedo, LAI, vegetation cover fraction, and FAPAR products (1999–2007) from SPOT/VEGETATION data (Baret et al., 2007). The long-term Global LAnd Surface Satellite (GLASS) products are based on both AVHRR and MODIS data, spanning from 1981 to the present (Liang et al., 2013). In addition, many other typical land surface parameters have been produced from sensors, such as Advanced Very-High-Resolution Radiometer (AVHRR), Polarization and Directionality of the Earth's Reflectance (POLDER), Multi-angle Imaging Spectroradiometer (MISR), Medium-Resolution Imaging Spectrometer (MERIS), and so on (Liang et al., 2012). Although significant progress has been made in generating global land products from satellite observations, two main problems remain.

Firstly, due to cloud contamination and insufficient observations, current products are often incomplete and discontinuous in time and space, since they are generally produced by using only single-phase remote sensing data based on instantaneous physical models. Kobayashi et al. (2010) and Fang et al. (2012) reported that MODIS C5 LAI shows temporal gaps and unrealistically strong variability, especially for forest type during the growing season. Many researchers have pointed out that the MODIS albedo product contains data gaps (20%–40% for the MCD43B3 product) and fails to capture the events of ephemeral snow because of weather or abrupt surface changes (Fang et al., 2007, T. He et al., 2012, Qu et al., 2015). Data assimilation, which combines observations and models, provides a solution by utilizing the multi-temporal signatures of satellite data (Lewis et al., 2012b, Liang and Qin, 2008). Xiao et al. (2011) developed a sequential assimilation method for real-time estimation of LAI from MODIS time series reflectance data, and Liu et al. (2014) extended this to multiple satellite data. The LAI values were updated recursively by combining predictions from dynamic models and MODIS reflectance data using an Ensemble Kalman Filter (EnKF) technique. Lewis et al., 2012a, Lewis et al., 2012b developed a weak constraint variational Earth Observation Land Data Assimilation System (EO-LDAS) to improve mapping of land surface biophysical parameters by exploiting the full information content provided by observations from satellite constellations. By involving the measured LAI in the assimilation procedure, Li et al. (2017) produced high-accuracy LAI of three subtropical forest in China from MODIS data based on the integrated EnKF and PROSAIL model, which is significant for the study of subtropical forest ecosystems. Data assimilation methods were also adapted to retrieve LAI/FAPAR or surface fluxes (Jiang et al., 2014, Qin et al., 2007, Stöckli et al., 2008).

Secondly, current methods for estimating land surface variables usually focus on individual parameters separately, even from the same satellite observations, resulting in inconsistent products. For example, the MODIS land surface products are produced separately by different teams; the main algorithm for MODIS LAI/FAPAR products is based on the three-dimensional (3D) radiative transfer model (Knyazikhin et al., 1998), while the main algorithm for the MODIS BRDF/albedo product is based on the semi-empirical kernel-based model (Schaaf et al., 2002). Garrigues et al. (2008) found that MODIS C4 LAI performs well for grass and crop types, while Fang et al. (2012) and De Kauwe et al. (2011) both found that MODIS C5 LAI underestimates the upper range of in situ LAI measurements for forest type. However, by comparing MODIS albedo retrievals and in situ measurements across the global FLUXNET network, Cescatti et al. (2012) concluded that MODIS albedo is systematically lower than in situ measurements for non-forest sites (grasslands, savannas, croplands), whereas the seasonal pattern of MODIS albedo matches extremely well for forest sites. To address this situation, Xiao et al. (2015b) developed a framework for consistent estimation of LAI, FAPAR, and albedo from MODIS time-series data, and demonstrated that the framework can estimate temporally complete land-surface parameters even if some of the reflectance data are contaminated by residual clouds or missing, and that the retrieved parameter values are physically consistent. Shi et al. (2016) extended this work by consistent estimation of land surface (LAI, FAPAR, albedo, PAR) and atmospheric (aerosol optical depth) parameters from MODIS TOA reflectance data. However, none of these studies explored thermal observations, which are suitable to estimate land surface parameters like LST and land surface longwave radiation. In addition, remote sensing observations in the optical- to Thermal InfraRed (TIR) spectrum are usually used separated in different physical or empirical models. Wan and Li (1997) proposed the MODIS day/night algorithm to retrieve LSE and LST from paired day/night observations of seven MODIS MIR and TIR bands. Other studies used only the reflected component of MIR. Libonati et al. (2010) retrieved MODIS MIR reflectance for burned area mapping in tropical environments. MIR observations remain under-utilized for environmental studies due to the complexity of the radiometric signal, which includes both reflected and emitted fluxes (Boyd and Petitcolin, 2004).

This study presents an approach for simultaneous estimation of six land surface parameters (LAI, FAPAR, surface albedo, LSE, LST, and LWUP), by exploiting visible-to-TIR observations from MODIS data, based on a unified radiative transfer model and a data assimilation framework. Section 2 describes the construction of the unified radiative transfer model and the methods used to retrieve the land surface parameters. Section 3 presents the data used in this study, including the field sites and remote sensing products, Section 4 gives the performance of the proposed algorithm, and Section 5 provides discussions and conclusions.

Section snippets

Methods

The flowchart in Fig. 1 illustrates the proposed unified inversion algorithm. At first, VISible/ShortWave InfraRed (VIS/SWIR) cloud-free reflectances from MODIS daily reflectance products were extracted using cloud mask information contained in the MODIS cloud product. An EnKF technique was used to estimate LAI from the unified optical–thermal radiative transfer model by combining predictions from the dynamic model with MODIS cloud-free daily reflectance data. The retrieved free parameters were

Study sites and field measurements

To evaluate the accuracy and consistency of the six parameters retrieved by the unified inversion algorithm, six representative sites from the AmeriFlux network (WWW1), consisting of Bondville, Brooks Field Site 11- Ames, Fort peck, Santa Rita Mesquite, Walker Branch Watershed, and Chestnut Ridge, were selected based on the principles of the concurrent land surface parameter observations during the survey period, typical land-surface biomes, and the landscape homogeneity for representativeness

Results analysis

A validation exercise was conducted for the selected sites with sufficient field measurements. The presently distributed MODIS, GLASS, and GlobAlbedo products were also used for comparison with the proposed method. Detailed validations of the six land surface parameters are discussed in the following sections.

The Bondville site is an agricultural site in the Midwestern United States, near Champaign, Illinois. The field was continuous no-till with alternating years of soybean and corn crops (

Discussion and conclusion

A new, unified inversion algorithm has been proposed for simultaneous estimation of six land surface parameters, by exploiting the visible-to-TIR spectral observations from MODIS sensor, based on a unified optical–thermal soil–canopy–leaf radiative transfer model and a data assimilation framework. LAI was first estimated using a data assimilation method that combines MODIS daily reflectance data and a LAI phenology model. The estimated LAI values were then input to the unified model to produce

WWW sites

WWW1: AmeriFlux Site and Data Exploration System. http://ameriflux.ornl.gov/

WWW2: ASTER spectral library. http://spclib.jpl.nasa.gov/

WWW3: USDA soil taxonomy. http://www.nrcs.usda.gov/wps/portal/nrcs/main/soils/survey/class/

WWW4: MODIS UCSB spectral library. http://www.icess.ucsb.edu/modis/EMIS/html/em.html

Acknowledgements

This research was financially supported by the National Natural Science Foundation of China under grant nos. 41331173 and 41171264, the Chinese Grand Research Program on Climate Change and Response under grant 2016YFA0600103. The authors thank the AmeriFlux network for providing ground validation data. We are grateful for two anonymous reviewers for their constructive comments and suggestions.

References (82)

  • D.K. Hall et al.

    MODIS snow-cover products

    Remote Sens. Environ.

    (2002)
  • L. He et al.

    Global clumping index map derived from the MODIS BRDF product

    Remote Sens. Environ.

    (2012)
  • T. He et al.

    Estimation of surface albedo and directional reflectance from Moderate Resolution Imaging Spectroradiometer (MODIS) observations

    Remote Sens. Environ.

    (2012)
  • C. Justice et al.

    An overview of MODIS Land data processing and product status

    Remote Sens. Environ.

    (2002)
  • A. Kuusk

    A two-layer canopy reflectance model

    J. Quant. Spectrosc. Radiat. Transfer

    (2001)
  • P. Lewis et al.

    An earth observation land data assimilation system (EO-LDAS)

    Remote Sens. Environ.

    (2012)
  • X. Li et al.

    Assimilating leaf area index of three typical types of subtropical forest in China from MODIS time series data based on the integrated ensemble Kalman filter and PROSAIL model

    ISPRS J. Photogram. Remote Sens.

    (2017)
  • Z.-L. Li et al.

    Satellite-derived land surface temperature: current status and perspectives

    Remote Sens. Environ.

    (2013)
  • R. Libonati et al.

    Retrieving middle-infrared reflectance for burned area mapping in tropical environments using MODIS

    Remote Sens. Environ.

    (2010)
  • Q. Liu et al.

    Retrieval of leaf area index using temporal, spectral, and angular information from multiple satellite data

    Remote Sens. Environ.

    (2014)
  • L. Pérez-Planells et al.

    Comparing different profiles to characterize the atmosphere for three MODIS TIR bands

    Atmospheric Res.

    (2015)
  • B. Pinty et al.

    Evaluation of the JRC-TIP 0.01° products over a mid-latitude deciduous forest site

    Remote Sens. Environ.

    (2011)
  • J.C. Price

    On the information content of soil reflectance spectra

    Remote Sens. Environ.

    (1990)
  • C.B. Schaaf et al.

    First operational BRDF, albedo nadir reflectance products from MODIS

    Remote Sens. Environ.

    (2002)
  • G. Schaepman-Strub et al.

    Reflectance quantities in optical remote sensing – definitions and case studies

    Remote Sens. Environ.

    (2006)
  • H. Shi et al.

    Consistent estimation of multiple parameters from MODIS top of atmosphere reflectance data using a coupled soil-canopy-atmosphere radiative transfer model

    Remote Sens. Environ.

    (2016)
  • J. Stroeve et al.

    Accuracy assessment of the MODIS 16-day albedo product for snow: comparisons with Greenland in situ measurements

    Remote Sens. Environ.

    (2005)
  • J. Stroeve et al.

    Re-evaluation of MODIS MCD43 Greenland albedo accuracy and trends

    Remote Sens. Environ.

    (2013)
  • E. Valor et al.

    Mapping land surface emissivity from NDVI: application to European, African, and South American areas

    Remote Sens. Environ.

    (1996)
  • W. Verhoef et al.

    Coupled soil–leaf-canopy and atmosphere radiative transfer modeling to simulate hyperspectral multi-angular surface reflectance and TOA radiance data

    Remote Sens. Environ.

    (2007)
  • J. Verrelst et al.

    Optical remote sensing and the retrieval of terrestrial vegetation bio-geophysical properties – a review

    ISPRS J. Photogram. Remote Sens.

    (2015)
  • Z. Wang et al.

    Evaluation of MODIS albedo product (MCD43A) over grassland, agriculture and forest surface types during dormant and snow-covered periods

    Remote Sens. Environ.

    (2014)
  • Z. Xiao et al.

    Estimating the fraction of absorbed photosynthetically active radiation from the MODIS data based GLASS leaf area index product

    Remote Sens. Environ.

    (2015)
  • Z. Xiao et al.

    Real-time retrieval of Leaf Area Index from MODIS time series data

    Remote Sens. Environ.

    (2011)
  • A. Berk et al.

    MODTRAN 5: a reformulated atmospheric band model with auxiliary species and practical multiple scattering options: update, Defense and Security

    Int. Soc. Optics Photonics

    (2005)
  • D. Boyd et al.

    Remote sensing of the terrestrial environment using middle infrared radiation (3.0–5.0 µm)

    Int. J. Remote Sens.

    (2004)
  • J.M. Chen et al.

    Defining leaf area index for non-flat leaves

    Plant, Cell Environ.

    (1992)
  • J. Cheng et al.

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

    J. Geophys. Res.: Atmospheres

    (2014)
  • J. Cheng et al.

    Estimating the hemispherical broadband longwave emissivity of global vegetated surfaces using a radiative transfer model

    IEEE Trans. Geosci. Remote Sens.

    (2016)
  • J. Cheng et al.

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

    IEEE Geosci. Remote Sens. Lett.

    (2013)
  • R.N. Clark et al.

    USGS Digital Spectral Library splib06a

    (2007)
  • Cited by (24)

    • A machine learning method trained by radiative transfer model inversion for generating seven global land and atmospheric estimates from VIIRS top-of-atmosphere observations

      2022, Remote Sensing of Environment
      Citation Excerpt :

      Verrelst et al. (2019) successfully estimated the LAI and fractional vegetation cover (FVC) at the highest Sentinel 2 spatial resolution of 10 m with a Gaussian processes regression (GPR). Recently, a multi-variable inversion framework has been proposed to simultaneously retrieve multiple land surface and atmospheric variables based on a complex coupled soil-canopy-atmosphere RTM (Ma et al., 2020; Ma et al., 2017a; Ma et al., 2017b; Shi et al., 2016; Zhang et al., 2021). The shuffled complex evolution (SCE) algorithm with the coupled RT model were first used to determine the surface Bidirectional Reflectance Distribution Function (BRDF) by inverting the surface essential variables (Ma et al., 2020), including LAI, chlorophyll concentration (Cab), dry matter content (Cm), equivalent water thickness (Cw), and soil reflectance variable (s1).

    • Extension of the Hapke model to the spectral domain to characterize soil physical properties

      2022, Remote Sensing of Environment
      Citation Excerpt :

      Therefore, this Hapke-HSR model can be incorporated into the soil-leaf-canopy radiative transfer model (Liang, 2004; Liang and Strahler, 1994; Verhoef et al., 2018; Verhoef and Bach, 2007) and used for the simultaneous retrieval of biophysical variables, and a variety of other applications (Liang and Wang, 2019; Ma et al., 2017a, 2017b; Walthall et al., 1985).

    • Cloudy-sky land surface temperature from VIIRS and MODIS satellite data using a surface energy balance-based method

      2021, Remote Sensing of Environment
      Citation Excerpt :

      The MODIS and VIIRS clear-sky LST values can be obtained from the MODIS and VIIRS LST products; however, LST estimated using a simultaneous retrieval method provides a more accurate LST estimation. Ma et al. (2017) developed a simultaneous retrieval scheme to estimate a suite of parameters from both MODIS VNIR and thermal-infrared (IR) bands, based on a unified optical-thermal soil-canopy-leaf (PROSPECT +4 SAIL) radiative transfer model, and an ensemble KF assimilation framework. The LAI was first determined by data assimilation and was then treated as a basic input parameter to produce the Fraction of Absorbed Photosynthetically Active Radiation (FAPAR), surface albedo, and land surface spectral emissivity (LSE).

    View all citing articles on Scopus
    View full text