Elsevier

Agricultural and Forest Meteorology

Volume 252, 15 April 2018, Pages 256-268
Agricultural and Forest Meteorology

Stand-alone uncertainty characterization of GLEAM, GLDAS and MOD16 evapotranspiration products using an extended triple collocation approach

https://doi.org/10.1016/j.agrformet.2018.01.022Get rights and content

Highlights

  • Evaluated uncertainty of AET products through an extended triple collocation approach.

  • GLEAM and GLDAS showed least absolute uncertainties compared to MOD16 dataset.

  • GLEAM revealed low errors in forest but performed similarly as to GLDAS in rice paddy.

  • A merged AET map was developed with least uncertainties from three selected products.

Abstract

An optimal use of the global scale actual evapotranspiration (AET) products for various hydro-meteorological applications requires a systematic characterization of their uncertainties. This study presents the first application of an extended triple collocation (TC) approach to provide mutually uncorrelated absolute and relative error structure among three readily available AET (MOD16, GLEAM, and GLDAS) products on the point and spatial scale within the extent of Asia. The physical evaluation results of GLEAM, GLDAS and MOD16 exhibited reasonable accuracy compared to the in-situ AET with mean Index of Agreement >0.71, 0.59 and 0.58, respectively, thereby yielding Root Mean Square Error between ∼4–13 mm/8 day over nine AsiaFlux sites representing forest, rice paddy, and grassland biomes. Theoretical uncertainty assessment of four AET dataset combinations revealed that an average ∼1.5–5.5 mm/8 day random error was contributed from in-situ AET, thereby reducing the accuracy of other datasets. GLEAM performed consistently better with least absolute and relative uncertainties over forest compared with rice paddy and grassland surfaces where GLDAS had almost similar errors as those obtained from GLEAM, while MOD16 showed high uncertainties over all vegetation conditions. Interestingly, all four datasets had large relative uncertainties (>25%) for low vegetation compared to the errors of tall canopies. A spatially merged product generated from the least uncertainties showed better agreement in order of GLDAS > GLEAM > MOD16 over 47%, 42% and 11% of the study area. Overall, the application of extended TC approach on the quality of three AET products is a step forward to develop the merged near real-time accurate AET dataset by processing of theoretical and systematic uncertainties in the current AET algorithms.

Introduction

Evapotranspiration (ET) is a nexus of energy, water, and the carbon cycle (Yang et al., 2016) which provides the interactions between climatic and hydrological processes to address the changes among plant-land-atmosphere conditions (Miralles et al., 2016). Actual evapotranspiration (AET) is defined as the loss of water from the surface of the Earth to the atmosphere, and it is typically linked with evaporation from the soil surface and plant transpiration in its natural environment under limited water conditions (Brutsaert and Chen, 1996). Availability of accurate and reliable AET information is a prerequisite for many hydro-meteorological applications such as water resource management, irrigation scheduling, crop yield estimation, and drought predictions (Liaqat et al., 2015; Baik and Choi, 2015). However, the involvement of sensitive climate feedback, heterogeneous land surfaces and environmental conditions, and their variability in space and time usually cause small to large-scale quantitative and qualitative uncertainties in the AET quantification (Ferguson et al., 2010; Long et al., 2014).

Despite of the above quantification and uncertainty challenges, the local field scale AET datasets generated through conventional ground-based techniques such as, scintillometers, weighing lysimeters, Bowen ratio, and eddy covariance based FLUXNET systems are usually considered as true representations of AET at a point scale, inspite of their own measurement errors and scaling issues (Park et al., 2017a,b; Sugita et al., 2017; Velpuri et al., 2013; Baik and Choi, 2015). Continuous efforts have been made in last few decades to produce multi-year global AET datasets by combining conventional methods with recent advancements in satellite remote sensing technologies (Rodell et al., 2004; Miralles et al., 2011a; Ghilain and Arboleda, 2011; Zhang et al., 2010; Onogi et al., 2007; Mu et al., 2007, 2011). These datasets have been generated considering the widespread application of the surface energy budget, land surface models, and empirical as well as semi-empirical approaches through the use of satellite imagery and observation based meteorological forcing (Landeras et al., 2008; Miralles et al., 2011a).

Among several global AET products (Rodell et al., 2004; Mu et al., 2007, 2011; Miralles et al., 2011a; Zhang et al., 2010; Ghilain and Arboleda, 2011; Onogi et al., 2007), very few qualify as datasets, which are widely available possess large coverage and find operational applications. Examples of such datasets include the MODerate resolution Imaging Spectroradiometer (MODIS) i.e., MOD16 AET (Mu et al., 2007, 2011) having eight-day, monthly, and yearly temporal and 1 km spatial resolution; Global Land Evaporation and Amsterdam Model (GLEAM) 25 km daily AET products (Martens et al., 2017; Miralles et al., 2011a,b); and Global Land Data Assimilation System (GLDAS) having 25 km AET products at 3 hourly and monthly temporal resolutions (Rodell et al., 2004). Moreover, MOD16, GLEAM, and GLDAS provide continuous AET datasets on a global scale which are based on Penman-Monteith, Priestley-Taylor, and land surface model approaches, respectively. All of these datasets are based on geophysical measurement systems such as models estimation, remote sensing and in-situ network forcing which are subject to various error sources that include the difference between in-situ measurements itself, sensor calibration and support scale, and underlying model assumptions and parametrization. Thus, the uncertainty analysis of these error sources is critical to characterize and use the above AET datasets with greater confidence in operational applications involving hydro-meteorological projections.

Currently, two types of error estimation techniques are available to investigate the accuracy of geophysical measurements. The first method involves measurments of the difference between modeled and in-situ datasets in terms of correlation, bias, and root mean square error (Majozi et al., 2017; Velpuri et al., 2013; Kim et al., 2012; Liu et al., 2016; Ramoelo et al., 2014; Li et al., 2014). This method is mostly used for calibration and validation purposes due to limited in-situ datasets availability (Jia et al., 2012). The second method employs the triple collocation (TC) error estimation technique (Stoffelen, 1998) that uses the statistical relationships to estimate the random error standard deviation for three collocated datasets of the same geophysical variable. At the same time, it does not consider any dataset which belongs to the perfectly observed system. This implies that the errors associated with target dataset are a true representation of uncertainty associated with that system, and this is mutually independent of the other two sources. Moreover, the mathematical simplicity of covariance based extended TC analysis is that the high-quality reference datasets are not really required. Instead the TC method directly estimates the Pearson’s correlation coefficient between noisy and unknown reference datasets (McColl et al., 2014; Gruber et al., 2016). Therefore, TC has emerged as one of the most promising error estimation techniques for earth observations and hydrological applications due to its optimal utilization of an independent reference system. Although, this method has been widely used to map error structures of a number of geophysical variables including ocean, wind and wave data (Caires and Sterl, 2003), precipitation (Alemohammad et al., 2015), sea surface temperature (Gentemann, 2014), leaf area indices (Fang et al., 2012) and more specifically soil moisture (Kim et al., 2018; Gruber et al., 2016; Scipal et al., 2008; Su et al., 2014a,b), but, to best of our knowledge, it has not yet been considered for analyzing AET datasets.

Therefore, the main objective of this study is to examine the error structure of three independently available long-term AET (MOD16, GLEAM, and GLDAS) datasets over a 10-years duration (2000–2010) in Asia using new covariance based extended TC method. The accuracy of each AET dataset was first examined by comparing its results independently against nine different eddy covariance-based flux tower observations collected over a range of rice paddy, forest, and grassland ecosystems. The results of error structures obtained using the TC method were then interpreted to understand the uncertainties and usage of each AET dataset under their observed dynamics. Moreover, we have also examined the spatio-temporal variations of TC error structures corresponding to land use land cover types in order to choose a suitable AET product for different land surface ecosystems in the region. Overall, the analysis of error characteristics in this study represents a significant step towards the evaluation of multiple global AET products under different land use conditions, which may serve as a baseline for suitable data assimilation insights.

Section snippets

Study area

This study was conducted within the extent of Asia (Fig. 1) the Wolrd’s largest and most populous continent with a heterogeneous land cover and fragile ecosystem. It was considered that the selected study region would likely experience accelerated hydrological cycles due to the recent increase in temperature and global warming effects (Zohaib et al., 2017; Liaqat and Choi, 2015; Byun et al., 2014). Recent modeling endeavors indicate that such warming effects would be prominent in the Tibetan

Extended triple collocation (TC) approach

The extended triple collocation (TC) approach was used to investigate the error structure of three globally available (MOD16, GLEAM, and GLDAS) AET datasets. Extended TC uses a statistical technique among three spatially coincident and temporally collocated datasets of the same geophysical variable to derive standard deviation (Std) of random error (εAET), simultaneously (McColl et al., 2014; Stoffelen, 1998; Gruber et al., 2016; Su et al., 2014a,b). The main advantage of the use of extended TC

Physical uncertainty analysis of AET products using flux tower point measurements

The time series and scatter plot analysis presented in Fig. 2 reveal the seasonal patterns predicted by the three AET (MOD16, GLEAM, and GLDAS) datasets over 8-day timescale in comparison to those obtained from the nine eddy-covariance based flux tower measurements collected over varying topography and a range of mixed forest (Fig. 2a), rice paddy (Fig. 2b), grassland and alpine meadow ecosystems (Fig. 2c). These comparison depicted that all the three AET products have relatively well captured

Conclusion

This study characterizes the quality of MOD16, GLEAM, and GLDAS widely available AET products based on their physical and theoretical uncertainties across several land cover systems within the extent of South and East Asia during the period of 2000–2010. In terms of physical accuracy analysis, the point estimations obtained from three (MOD16, GLEAM, and GLDAS) products shows moderate to good correlation (R) with corresponding average value of (0.35–0.60, 0.63–0.84 and 0.48–0.79) compared to

Acknowledgements

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (NRF-2016R1A2B4008312). This research was supported by Space Core Technology Development Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (NRF-2014M1A3A3A02034789).

References (62)

  • S. Li et al.

    Evapotranspiration from a Mongolian steppe under grazing and its environmental constraints

    J. Hydrol.

    (2007)
  • G. Landeras et al.

    Comparison of artificial neural network models and empirical and semi-empirical equations for daily reference evapotranspiration estimation in the Basque country (Northern Spain)

    Agric. Water Manage.

    (2008)
  • W. Liu et al.

    A worldwide evaluation of basin-scale evapotranspiration estimates against the water balance method

    J. Hydrol.

    (2016)
  • Q. Mu et al.

    Development of a global evapotranspiration algorithm based on MODIS and global meteorology data

    Remote Sens. Environ.

    (2007)
  • Q. Mu et al.

    Improvements to a MODIS global terrestrial evapotranspiration algorithm

    Remote Sens. Environ.

    (2011)
  • J. Park et al.

    Satellite-based crop coefficient and evapotranspiration using surface soil moisture and vegetation indices in Northeast Asia

    Catena

    (2017)
  • A.D. Richardson et al.

    A multi-site analysis of random error in tower-based measurements of carbon and energy fluxes

    Agric. For. Meteorol.

    (2006)
  • C. Su et al.

    Remote sensing of environment stand-alone error characterisation of microwave satellite soil moisture using a fourier method

    Remote Sens. Environ.

    (2014)
  • N.M. Velpuri et al.

    A comprehensive evaluation of two MODIS evapotranspiration products over the conterminous united

    Remote Sens. Environ.

    (2013)
  • S.H. Alemohammad et al.

    Characterization of precipitation product errors across the United States using multiplicative triple collocation

    Hydrol. Earth Syst. Sci.

    (2015)
  • W. Brutsaert et al.

    Diurnal variation of surface fluxes during thorough drying (or severe drought) of natural prairie

    Water Resour. Res.

    (1996)
  • J. Baik et al.

    Evaluation of remotely sensed actual evapotranspiration products from COMS and MODIS at two different flux tower sites in Korea

    Int. J. Remote Sens.

    (2017)
  • G. Badlgley et al.

    On uncertainty in global terrestrial evapotranspiration estimates from choice of input forcing datasets *

    J. Hydrometeorol.

    (2015)
  • S. Caires et al.

    Validation of ocean wind and wave data using triple collocation

    J. Geophys. Res.

    (2003)
  • W.A. Dorigo et al.

    Error characterisation of global active and passive microwave soil moisture datasets

    Hydrol. Earth Syst. Sci.

    (2010)
  • L. DAAC

    Land Cover Type Yearly L3 Global 500 m SIN Grid (MCD12Q1)

    (2009)
  • D. Dragoni et al.

    Uncertainty of annual net ecosystem productivity estimated using eddy covariance flux measurements

    J. Geophys. Res.

    (2007)
  • C.R. Ferguson et al.

    Quantifying uncertainty in a remote sensing-based estimate of evapotranspiration over continental USA

    Int. J. Remote Sens.

    (2010)
  • P.L. Fuehrer et al.

    Flux corrections revisited

    Bound. Layer Meteorol.

    (2002)
  • C.L. Gentemann

    Three way validation of MODIS and AMSR-E sea surface temperatures

    J. Geophys. Res.: Oceans

    (2014)
  • N. Ghilain et al.

    Evapotranspiration modelling at large scale using near-real time MSG SEVIRI derived data

    Hydrol. Earth Syst. Sci.

    (2011)
  • Cited by (166)

    • Disaggregation of remote sensing and model-based data for 1 km daily seamless soil moisture

      2023, International Journal of Applied Earth Observation and Geoinformation
    View all citing articles on Scopus
    View full text