Evapotranspiration (ET) is critical in linking global water, carbon and
energy cycles. However, direct measurement of global terrestrial ET is not
feasible. Here, we first reviewed the basic theory and state-of-the-art
approaches for estimating global terrestrial ET, including remote-sensing-based physical models, machine-learning algorithms and land surface
models (LSMs). We then utilized 4 remote-sensing-based physical models,
2 machine-learning algorithms and 14 LSMs to analyze the spatial and
temporal variations in global terrestrial ET. The results showed that the
ensemble means of annual global terrestrial ET estimated by these three
categories of approaches agreed well, with values ranging from 589.6 mm yr-1
(6.56×104 km3 yr-1) to 617.1 mm yr-1
(6.87×104 km3 yr-1). For the period from 1982 to 2011, both
the ensembles of remote-sensing-based physical models and machine-learning
algorithms suggested increasing trends in global terrestrial ET (0.62 mm yr-2 with a significance level of p<0.05 and 0.38 mm yr-2 with a significance level of p<0.05,
respectively). In contrast, the ensemble mean of the LSMs showed no
statistically significant change (0.23 mm yr-2, p>0.05),
although many of the individual LSMs reproduced an increasing trend.
Nevertheless, all 20 models used in this study showed that anthropogenic
Earth greening had a positive role in increasing terrestrial ET. The
concurrent small interannual variability, i.e., relative stability, found in
all estimates of global terrestrial ET, suggests that a potential
planetary boundary exists in regulating global terrestrial ET, with the value of this boundary being
around 600 mm yr-1. Uncertainties among approaches were identified in
specific regions, particularly in the Amazon Basin and arid/semiarid
regions. Improvements in parameterizing water stress and canopy dynamics,
the utilization of new available satellite retrievals and deep-learning methods,
and model–data fusion will advance our predictive understanding of global
terrestrial ET.
Introduction
Terrestrial evapotranspiration (ET) is the sum of the water lost to the
atmosphere from plant tissues via transpiration and that lost from the land
surface elements, including soil, plants and open water bodies, through
evaporation. Processes controlling ET play a central role in linking the
energy (latent heat), water (moisture flux) and carbon cycles
(photosynthesis–transpiration trade-off) in the Earth system. Over 60 % of
precipitation on the land surface is returned to the atmosphere through ET
(Oki and Kanae, 2006), and the accompanying latent heat (λET,
λ is the latent heat of vaporization) accounts for more than half of
the solar energy received by the land surface (Trenberth et al., 2009).
ET is also coupled with the carbon dioxide (CO2) exchange between the canopy and
the atmosphere through vegetation photosynthesis. These linkages make ET an
important variable in both short-term numerical weather forecasts and
long-term climate predictions. Moreover, ET is a critical indicator for
ecosystem functioning across a variety of spatial scales. Therefore, in order to enhance our
predictive understanding of the Earth system and sustainability, it
is essential to accurately assess land surface ET in a changing global
environment.
However, large uncertainty still exists in quantifying the magnitude of
global terrestrial ET and its spatial and temporal patterns, despite
extensive research (Allen et al., 1998; Liu et al., 2008; Miralles et
al., 2016; Mueller et al., 2011; Tian et al., 2010). Previous estimates
of global land mean annual ET range from 417 to 650 mm yr-1 for the whole or part of the 1982–2011 period (Mu et al., 2007;
Mueller et al., 2011; Vinukollu et al., 2011a; Zhang et al., 2010). This
large discrepancy among independent studies may be attributed to a lack of
sufficient measurements, uncertainty in forcing data, inconsistent spatial
and temporal resolutions, ill-calibrated model parameters, and deficiencies
in model structures. Of the four components of ET (transpiration, soil
evaporation, canopy interception and open water evaporation), transpiration
(Tv) contributes the largest uncertainty; this is due to the fact that it is modulated not only
by surface meteorological conditions and soil moisture but also by the
physiology and structures of plants. Changes in non-climatic factors such as
elevated atmospheric CO2, nitrogen deposition and land covers also
serve as influential drivers of Tv (Gedney et al., 2006; Mao et al.,
2015; S. Pan et al., 2018; Piao et al., 2010). As such, the global ratio of
transpiration to ET (Tv/ET) has long been a matter of debate, with the most
recent observation-based estimate being 0.64±0.13 constrained by the
global water-isotope budget (Good et al., 2015). Most Earth system
models are thought to largely underestimate Tv/ET (Lian et al.,
2018).
Global warming is expected to accelerate the hydrological cycle (Pan et
al., 2015). For the period from 1982 to the late 1990s, ET was reported to
have increased by about 7 mm (∼1.2 %) per decade driven by an
increase in radiative forcing and, consequently, global and regional
temperatures (Douville et al., 2013; Jung et al., 2010; Wang et al.,
2010). The contemporary near-surface specific humidity also increased over
both land and ocean (Dai, 2006; Simmons et al., 2010; Willett et al.,
2007). More recent studies confirmed that global ET has
showed an overall increase since the 1980s (Mao et al., 2015; Yao et al., 2016; Zeng et
al., 2018a, 2012, 2016; Zhang et al., 2015; Y. Zhang et al., 2016). However, the magnitude and spatial distribution of such a
trend are far from determined. Over the past 50 years, pan evaporation has
decreased worldwide (Fu et al., 2009; Peterson et al., 1995;
Roderick and Farquhar, 2002), implying an increase in actual ET given the
pan evaporation paradox. Moreover, the increase in global terrestrial ET was
found to cease or even be reversed from 1998 to 2008, primarily due to the
decreased soil moisture supply in the Southern Hemisphere (Jung et al.,
2010). To reconcile the disparity, Douville et al. (2013) argued that
the peak ET in 1998 should not be taken as a tipping point because ET was
estimated to increase in the multi-decadal evolution. More efforts are
needed to understand the spatial and temporal variations of global
terrestrial ET and the underlying mechanisms that control its magnitude and
variability.
Conventional techniques, such as lysimeter, eddy covariance, large aperture
scintillometer and the Bowen ratio method, are capable of providing ET
measurements at point and local scales (Wang and Dickinson, 2012).
However, it is impossible to directly measure ET at the global scale because
dense global coverage by such instruments is not feasible, and the
representativeness of point-scale measurements to comprehensively portray
the spatial heterogeneity of global land surface is also doubtful
(Mueller et al., 2011). To address this issue, numerous approaches have
been proposed in recent years to estimate global terrestrial ET and these
approaches can be divided into three main categories: (1) remote-sensing-based physical models, (2) machine-learning algorithms and (3) land
surface models (Miralles et al., 2011; Mueller et al., 2011; Wang and
Dickinson, 2012). Knowledge of the uncertainties in global terrestrial ET
estimates from different approaches is a prerequisite for future
projection and many other applications. In recent years, several studies
have compared multiple terrestrial ET estimates (Khan et al., 2018;
Mueller et al., 2013; Wartenburger et al., 2018; Y. Zhang et al., 2016);
however, most of these studies analyzed multiple datasets of the same
approach or focused on investigating similarities and differences among
different approaches. Few studies have been conducted to identify
uncertainties in multiple estimates of different approaches.
In this study, we integrate state-of-the-art estimates of global terrestrial
ET, including data-driven and process-based estimates, to assess its spatial
pattern, interannual variability, environmental drivers, long-term trend
and response to vegetation greening. Our goal is not to compare the various
models and choose the best one but to identify the uncertainty sources in
each type of estimate and provide suggestions for future model development.
In the following sections, we first give a brief introduction to all of the
methodological approaches and ET datasets used in this study. We then
quantify the spatiotemporal variations in global terrestrial ET during the
period from 1982 to 2011 by analyzing the results from the current state-of-the-art
models. Finally, we discuss some suggested solutions for reducing the
identified uncertainties.
Methodology and data sourcesOverview of approaches to global ET estimationRemote-sensing-based physical models
Satellite remote sensing has been widely recognized as a promising tool for
estimating global ET, because it is capable of providing spatially and
temporally continuous measurements of critical biophysical parameters
affecting ET, including vegetation states, albedo, the fraction of absorbed
photosynthetically active radiation, land surface temperature and plant
functional types (Li et al., 2009). Since the 1980s, a large number of
methods have been developed using a variety of satellite observations
(K. Zhang et al., 2016). However, some of these methods, such as surface
energy balance (SEB) models and surface temperature–vegetation index
(Ts–VI), are usually applied at local and regional scales. At the global
scale, the vast majority of the existing remote-sensing-based physical models
can be categorized into two groups: those based on the Penman–Monteith (PM) equation and those based on the
Priestley–Taylor (PT) equation.
Remote sensing models based on the Penman–Monteith equation
The Penman equation, derived from the Monin–Obukhov similarity theory and
surface energy balance, uses surface net radiation, temperature, humidity,
wind speed and ground heat flux to estimate ET from an open water surface.
For vegetated surfaces, canopy resistance was introduced into the Penman
equation by Monteith (Monteith, 1965), and the PM equation is
formulated as follows:
λET=ΔRn-G+ρaCpVPD/raΔ+γ1+rs/ra,
where Δ, Rn, G, ρa, Cp, γ, rs,
ra and VPD are the slope of the curve relating saturated water vapor
pressure to air temperature, net radiation, soil heat flux, air density, the
specific heat of air, the psychrometric constant, surface resistance,
aerodynamic resistance and the vapor pressure deficit, respectively. The canopy
resistance term in the PM equation exerts a strong control on transpiration.
For example, based on the algorithm proposed by Cleugh et al. (2007),
the MODIS (Moderate Resolution Imaging Spectroradiometer) ET algorithm
improved the model performance via the inclusion of environmental stress
into the canopy conductance calculation and explicitly accounted for soil
evaporation (Mu et al., 2007). Further, Mu et al. (2011) improved
the MODIS ET algorithm by considering nighttime ET, adding the soil heat flux
calculation, separating the dry canopy surface from the wet, and dividing the soil
surface into saturated wet surface and moist surface. Similarly, Zhang
et al. (2010) developed a Jarvis–Stewart-type canopy conductance model based
on the normalized difference vegetation index (NDVI) to take advantage of the
long-term Advanced Very High Resolution Radiometer (AVHRR) dataset. More
recently, this model was improved by adding a CO2 constraint function
into the canopy conductance estimate (Zhang et al., 2015). Another
important revision for the PM approach is proposed by Leuning et al. (2008). The Penman–Monteith–Leuning method adopts a simple biophysical model
for canopy conductance, which can account for the influences of radiation and
the atmospheric humidity deficit. Additionally, it introduces a simpler soil
evaporation algorithm than that proposed by Mu et al. (2007), which
potentially makes it attractive for use with remote sensing. However,
PM-based models have one intrinsic weakness – temporal upscaling – which is
required when translating instantaneous ET estimation into a longer timescale
value (Li et al., 2009). This could easily be done at the daily scale
under clear-sky conditions but faces challenge at weekly to monthly
timescales due to lack of cloud coverage information.
Remote sensing models based on the Priestley–Taylor equation
The Priestley–Taylor (PT) equation is a simplification of the PM equation
that does not parameterize aerodynamic and surface conductance (Priestley and
Taylor, 1972); it can be expressed as follows:
λET=fstress×α×ΔΔ+γ×Rn-G,
where fstress is a stress factor and is usually computed as a function
of environmental conditions. α is the PT parameter with a value of between
1.2 and 1.3 under water unstressed conditions and can be estimated using remote
sensing. Although the original PT equation works well for estimating
potential ET across most surfaces, the Priestley–Taylor coefficient,
α, usually needs adjustment to convert potential ET to actual ET
(K. Zhang et al., 2016). Thus, Fisher et al. (2008) developed a
modified PT model that keeps α constant but scales down potential ET
using ecophysiological constraints and soil evaporation partitioning. The
accuracy of their model has been validated against eddy-covariance
measurements conducted in a wide range of climates and involving many plant functional
types (Fisher et al., 2009; Vinukollu et al., 2011b). Following this
idea, Yao et al. (2013) further developed a modified Priestley–Taylor
algorithm that constrains soil evaporation using the index of the soil water deficit derived from the apparent thermal inertia. Miralles et al. (2011) also
proposed a novel PT-type model: the Global Land Evaporation Amsterdam Model (GLEAM). GLEAM combines a soil water module, a canopy
interception model and a stress module within the PT equation. The key
distinguishing features of this model are the use of microwave-derived soil
moisture, land surface temperature and vegetation density, and the detailed
estimation of rainfall interception loss. In this way, GLEAM minimizes the
dependence on static variables, avoids the need for parameter tuning and
enables the quality of the evaporation estimates to rely on the accuracy of
the satellite inputs (Miralles et al., 2011). Compared with the PM
approach, the PT-based approaches avoid the computational complexities of
aerodynamic resistance and the accompanying error propagation. However, the
many simplifications and semiempirical parameterization of physical
processes in the PT-based approaches may lower its accuracy.
Vegetation-index-based empirical algorithms and machine-learning methods
The principle of empirical ET algorithms is to link observed ET to its
controlling environmental factors through various statistical regressions or
machine-learning algorithms of different complexities. The earliest
empirical regression method was proposed by Jackson et al. (1977). At
present, the majority of regression models are based on vegetation indices
(Glenn et al., 2010), such as the NDVI and the enhanced vegetation index (EVI),
because of their simplicity, resilience in the presence of data gaps,
utility under a wide range of conditions and connection with vegetation
transpiration capacity (Maselli et al., 2014; Nagler et al., 2005; Yuan
et al., 2010). As an alternative to statistical regression methods, machine-learning algorithms have been gaining increased attention for ET estimation
due to their ability to capture the complex nonlinear relationships between
ET and its controlling factors (Dou and Yang, 2018). Many conventional
machine-learning algorithms, such as those based on artificial neural networks, random
forest, and support vector machine algorithms, have been applied in
various ecosystems (Antonopoulos et al., 2016; Chen et al., 2014; Feng et
al., 2017; Shrestha and Shukla, 2015) and have proved to be more accurate in
estimating ET than simple regression models (Antonopoulos et al., 2016;
Chen et al., 2014; Kisi et al., 2015; Shrestha and Shukla, 2015; Tabari et
al., 2013). In upscaling FLUXNET ET to the global scale, Jung et al. (2010) used the model tree ensemble method to integrate eddy-covariance
measurements of ET with satellite remote sensing and surface meteorological
data. In a recent study (Bodesheim et al., 2018), the
random forest approach was used to derive global ET at a 30 min
timescale.
Process-based land surface models (LSMs)
Although satellite-derived ET products have provided quantitative
investigations of historical terrestrial ET dynamics, they can only cover a
limited temporal record of about 4 decades. To obtain terrestrial ET
before the 1980s and predict future ET dynamics, LSMs are needed, as they are
able to represent a large number of interactions and feedbacks between
physical, biological and biogeochemical processes in a prognostic way
(Jimenez et al., 2011). ET simulation in LSMs is regulated by multiple
biophysical and physiological properties or processes, including but not
limited to stomatal conductance, leaf area, root water uptake, soil water,
runoff and (sometimes) nutrient uptake (Famiglietti and Wood, 1991; Huang
et al., 2016; Lawrence et al., 2007). Although almost all current LSMs have
these components, different parameterization schemes result in substantial
differences in ET estimation (Wartenburger et al., 2018). Therefore, in
recent years, the multi-model ensemble approach has become popular in
quantifying the magnitude, spatiotemporal pattern and uncertainty of global
terrestrial ET (Mueller et al., 2011; Wartenburger et al., 2018). Yao
et al. (2017) showed that a simple model averaging method or a Bayesian
model averaging method is superior to each individual model in predicting
terrestrial ET.
Description of ET models used in this study
In this study, we evaluate 20 ET products that are based on remote-sensing-based physical models, machine-learning algorithms and LSMs to
investigate the magnitudes and spatial patterns of global terrestrial ET
over recent decades. Table 1 lists the input data, the ET algorithms adopted,
the advantages and limitations, and the references for each product. We use a simple model
averaging method when calculating the mean value of multiple models.
Descriptions of the models used in this study, including their drivers,
adopted algorithms, key equations, advantages and limitations, and references.
NameInputAlgorithmSpatialTemporalKey equationsAdvantages and limitationsReferencesresolutionresolutionMTEClimate: precipitation, temperature, sunshine hour, relative humidity and wet days Vegetation: fAPARTRIAL + ERROR0.5∘×0.5∘MonthlyNo specific equationLimitations: insufficient flux observations in tropical regions; no CO2 effectJung et al. (2011)RFClimate: land surface temperature, radiation, potential radiation, index of water availability and relative humidity Vegetation: fAPAR, LAI and enhanced vegetation index (EVI)Randomized decision tree0.5∘×0.5∘30 minNo specific equationLimitations: the same as for MTEBodesheim et al. (2018)P-LSHClimate: radiation, air temperature, vapor pressure, wind speed and CO2 Vegetation: AVHRR NDVIModified Penman–Monteith0.083∘×0.083∘MonthlyEv=ΔRn+ρCpVPDgaλvΔ+Υ1+gagsEs=RHVPDkΔRn+ρCpVPDgaλvΔ+Υ1+gagsAdvantages: more robust physical basis; considers the effects of CO2 Limitations: high meteorological forcing requirements; canopy conductance is based on proxiesZhang et al. (2015)GLEAMClimate: precipitation, net radiation, surface soil moisture, land surface temperature, air tempera- ture and snow depth Vegetation: vegetation optical depthModified Priestley–Taylor0.25∘×0.25∘DailyEs=fsSsαsΔλvρwΔ+γRns-GsEsc=fscSscαscΔλvρwΔ+γRnsc-GscEtc=ftcStcαtcΔλvρwΔ+γRntc-Gtc-βEiAdvantages: simple; low requirement for meteorological data; well-suited for remote sensing observable variables; soil moisture is considered Limitations: many simplifications of physical processes; neither VPD nor surface and aerodynamic resistances are explicitly accounted for; strong dependency on net radiationMiralles et al. (2011)MODISClimate: air temperature, shortwave radiation, wind speed, relative humidity and air pressure Vegetation: LAI and fAPAR, albedoPenman–Monteith–Leuning0.05∘×0.05∘MonthlyEi=fwetfcΔRn-G+ρcpVPDrawcλvρwΔ+γrswcrawcEv=1-fwetfcΔRn-G+ρcpVPDratλvρwΔ+ΥrstratEs=fwet+1-fwethVPDβsAsoil+ρcp1-fcVPDrasλvρwS+γrtotrasAdvantages: more robust physical basis; Limitations: requires many variables that are difficult to observe or that are not observable with satellites; canopy conductance is based on proxies; does not consider soil moisture but uses atmospheric humidity as a surrogate; does not consider the effects of CO2Mu et al. (2011)PML-CSIROClimate: precipitation, air temperature, vapor pres- sure, shortwave radiation, longwave radiation and wind speed Vegetation: AVHRR LAI, emissivity and albedoPenman–Monteith–Leuning0.5∘×0.5∘MonthlyEv=ΔRn+ρCpVPDgaλvΔ+Υ1+gagsEs=fΔAsΔ+γEi: an adapted version of the Gash rainfall interception model (Van Dijk and Bruijnzeel, 2001)Advantages: more robust physical basis (compared with Priestley–Taylor equation); biophysically based estimation of surface conductance Limitations: high meteorological forcing requirements; canopy conductance is based on proxies; does not consider the effects of CO2Y. Zhang et al. (2016)
Continued.
TRENDY LSMs Advantages: land surface models are process-oriented and physically based. Given their structure almost all models are capable of allowing factorial analysis, where one forcing can be applied at a time. Most models also consider the physiological effect of CO2 on stomatal closure. Limitations: most models typically do not allow integration/assimilation of observation-based vegetation characteristics. Model parameterizations remain uncertain and the same process is modeled in different ways across models. Model parameters may or may not be physically based and, therefore, measurable in the field. Models participating in the TRENDYv6 comparison were forced by precipitation, air temperature, specific humidity, shortwave radiation, longwave radiation and wind speed based on the CRU-NCEPv8 data, as explained in Le Quéré et al. (2018). It is very difficult to list all key equations for all land surface models. Thus, we only list the stomatal conductance equation for each model in this table. NameAlgorithmSpatialTemporalKey equationsReferencesresolutionresolutionCABLEPenman–Monteith0.5∘×0.5∘Monthlygs=g0+g1fwAca-cp1+VPDVPD0-1Haverd et al. (2018)CLASS-CTEMModified Penman–Monteith2.8125∘×2.8125∘Monthlygc=mAnpcs-Γ11+VPD/VPD0+bLAIMelton and Arora (2016)CLM45Modified Penman–Monteith1.875∘×2.5∘Monthlygs=g0+g1AcaRHOleson et al. (2010)DLEMPenman–Monteith0.5∘×0.5∘Monthlygs=maxgsmaxrcorrbfppdffTminfVPDfCO2,gsminPan et al. (2015)ISAMModified Penman–Monteith0.5∘×0.5∘Monthlygs=mACs/Patm×esei+btβtBarman et al. (2014)JSBACHPenman–Monteith1.9∘×1.9∘Monthlygs=βw1.6An,potca-ci,potKnauer et al. (2015)JULESPenman–Monteith2.5∘×3.75∘MonthlyBare soil conductance: gsoil=1100θ1θc2 Stomatal conductance is calculated by solving the following two equations: Al=gsCs-Ci/1.6; Ci-Γ∗Cc-Γ∗=f01-ΔqcLi et al. (2016)LPJ-GUESSEquations proposed by Monteith (1995)0.5∘×0.5∘Monthlygs=gsmin+1.6Adtca1-λcSmith (2001)LPJ-wslPriestley–Taylor0.5∘×0.5∘Monthlygs=gsmin+1.6Adtca1-λcSitch et al. (2003)LPX-BernModified equation of Monteith (1995)1∘×1∘Monthlygs=gsmin+1.6Adtca1-λcKeller et al. (2017)O-CNModified Penman–Monteith1∘×1∘Monthlygs=gsmin+1.6Adtca1-λcZaehle and Friend (2010)ORCHIDEEModified Penman–Monteith0.5∘×0.5∘Monthlygs=g0+A+Rdca-cpfvpdgsoil=exp8.206-4.255W/Wsatd'Orgeval et al. (2008)ORCHIDEE-MICTModified Penman–Monteith0.5∘×0.5∘Monthlygs=g0+A+Rdca-cpfvpdGuimberteau et al. (2018)VISITPenman–Monteith0.5∘×0.5∘Monthlygs=g0+g1fwAca-cp1+VPDVPD0-1Ito (2010)
Notes: A: net assimilation rate; Adt: total daytime net photosynthesis;
An,pot: unstressed net assimilation rate; b: soil moisture factor;
bt: stomatal conductance intercept; ca: atmospheric CO2
concentration; cc: critical CO2 concentration; ci: internal
leaf concentration of CO2; ci,pot: internal leaf concentration
of CO2 for unstressed conditions; cs: leaf surface CO2
concentration; cp: CO2 compensation point; es: vapor pressure
at leaf surface; ei: saturation vapor pressure inside the leaf;
Es: soil evaporation; Ec: canopy evapotranspiration; Edry:
dry canopy evapotranspiration; Ewet: wet canopy evapotranspiration;
Ev: canopy transpiration; Ei: canopy interception; Etc:
transpiration from tall canopy; Esc: transpiration from short canopy;
f: fraction of P to equilibrium soil evaporation; fs: soil fraction;
fsc: short canopy fraction; ftc: tall canopy fraction; fvpd:
factor of the effect of leaf-to-air vapor pressure difference; fw: a
function describing the soil water stress on stomatal conductance;
fwet: relative surface wetness parameter; f0: the maximum ratio of
internal to external CO2; f(ppdf):limiting factor of photosynthetic photo flux
density; f(Tmin): limiting factor of daily minimum temperature; f(VPD): limiting
factor of vapor pressure deficit; f(CO2): limiting factor of carbon dioxide;
G: ground energy flux; ga: aerodynamic conductance; gm:
empirical parameter; gs: stomatal conductance; gsmax: maximum
stomatal conductance; gsmin: minimum stomatal conductance; gsoil:
bare soil conductance; g0: residual stomatal conductance when the net
assimilation rate is 0 ; g1: sensitivity of stomatal conductance to
assimilation, ambient CO2 concentration and environmental controls; I:
tall canopy interception loss; m: stomatal conductance slope; Patm:
atmospheric pressure; PEs: potential soil evaporation; PEcanopy:
potential canopy evaporation; qa: specific air humidity; qc:
critical humidity deficit; qs: specific humidity of saturated air;
ra: aerodynamic resistance; rs: stomatal resistance; Rn: net
radiation; Rd: day respiration; RH: relative humidity; Ts: actual
surface temperature; VPD: vapor pressure deficit; VPD0: the sensitivity
of stomatal conductance to VPD; W: top soil moisture; Wcanopy: canopy
water; Wsat: soil porosity; α: Priestley–Taylor coefficient;
αm: empirical parameter; β: a constant accounting for the
times when vegetation is wet; βt: soil water availability
factor between 0 and 1; βw: a scaling factor to account for
water stress; βs: moisture availability function; ρ: air
density; γ: psychrometric constant; λv: latent heat of
vaporization; λc: ratio of intercellular to ambient partial
pressure of CO2; rcorr: correction factor of temperature and air
pressure on conductance; Γ∗: CO2 compensation point
when leaf day respiration is zero; θ1: parameter of moisture
concentration in the top soil layer; θc: parameter of moisture
concentration in the spatially varying critical soil moisture; Δ:
slope of the vapor pressure curve.
Four physically based remote sensing datasets, including the Process-based Land
Surface Evapotranspiration/Heat Fluxes algorithm (P-LSH), the Global Land Evaporation Amsterdam Model (GLEAM), the Moderate Resolution
Imaging Spectroradiometer (MODIS) and the PML-CSIRO (Penman–Monteith–Leuning),
and two machine-learning datasets, which are based on random forest (RF) and model
tree ensemble (MTE), are used in our study. Both machine-learning and
physically based remote sensing datasets (six datasets in total) were
considered as benchmark products. The ensemble mean of the benchmark products
was calculated as the mean value of all machine-learning and physically based
satellite estimates as we treated each benchmark dataset equally.
Three of the four remote-sensing-based physical models quantify ET using
PM approaches. P-LSH adopts a modified PM approach coupled with
biome-specific canopy conductance determined from NDVI (Zhang et al.,
2010). The modified P-LSH model used in this study also accounts for the
influences of atmospheric CO2 concentrations and wind speed on canopy
stomatal conductance and aerodynamic conductance (Zhang et al., 2015).
The MODIS ET model is based on the algorithm proposed by Cleugh et al. (2007). Mu et al. (2007) improved the model performance through the
inclusion of environmental stress into the canopy conductance calculation and by
explicitly accounting for soil evaporation by combining the complementary
relationship hypothesis with the PM equation. The MODIS ET product (MOD16A3)
used in this study was further improved by considering nighttime ET,
simplifying the vegetation cover fraction calculation, adding the soil heat flux
item, dividing the saturated wet and moist soil, separating the dry and wet canopy,
as well as modifying algorithms of aerodynamic resistance, stomatal
conductance and boundary layer resistance (Mu et al., 2011). PML-CSIRO
adopts the Penman–Monteith–Leuning algorithm, which calculates surface
conductance and canopy conductance using a biophysical model instead of classic
empirical models. The maximum stomatal conductance is estimated using the
trial-and-error method (Y. Zhang et al., 2016). Furthermore, for each grid
covered by natural vegetation, the PML-CSIRO model constrains ET at the
annual scale using the Budyko hydrometeorological model proposed by
Fu (1981). The GLEAM ET calculation is based on the PT equation, which
requires fewer model inputs than the PM equation, and the majority of these
inputs can be directly achieved from satellite observations. Its rationale
is to make the most of information about evaporation contained in the
satellite-based environmental and climatic observations (Martens et al.,
2017; Miralles et al., 2011). Key variables including air temperature, land
surface temperature, precipitation, soil moisture, vegetation optical depth
and snow water equivalent are satellite-observed values. Moreover, the extensive
usage of microwave remote sensing products in GLEAM ensures the accurate
estimation of ET under diverse weather conditions. Here, we use GLEAM v3.2, which has overall better quality than previous versions
(Martens et al., 2017).
The first used machine-learning model, MTE, is based on the Tree Induction
Algorithm (TRIAL) and Evolving Trees with Random Growth (ERROR) algorithm
(Jung et al., 2009). The TRIAL grows model trees from the root node and
splits at each node with the criterion of minimizing the sum of squared
errors of multiple regressions in both subdomains. ERROR is used to select
the model trees that are independent of one another and have the best
performance based on the Schwarz criterion. The canopy fraction of absorbed
photosynthetic active radiation (fAPAR), temperature, precipitation,
relative humidity, sunshine hours and potential radiation are used as
explanatory variables to train MTE (Jung et al., 2011). The second
machine-learning model is the random forest (RF) algorithm, whose rationale
is generating a set of independent regression trees by randomly
selecting training samples automatically (Breiman, 2001). Each regression
tree is constructed using samples selected by a bootstrap sampling method.
After fixing the individual tree in entity, the final result is determined by
simple averaging. One merit of the RF algorithm is its capability to handle
complicated nonlinear problems and high dimensional data (Xu et al.,
2018). For the RF product used in this study, multiple explanatory variables
including the enhanced vegetation index, fAPAR, the leaf area index, daytime and
nighttime land surface temperature, incoming radiation, top-of-atmosphere
potential radiation, the index of water availability and relative humidity were
used to train regression trees (Bodesheim et al., 2018).
The 14 LSM-derived ET products were from the Trends and Drivers of
the Regional Scale Sources and Sinks of Carbon Dioxide (TRENDY) project
(including CABLE, CLASS-CTEM, CLM45, DLEM, ISAM, JSBACH, JULES, LPJ-GUESS,
LPJ-wsl, LPX-Bern, O-CN, ORCHIDEE, ORCHIDEE-MICT and VISIT). Daily gridded
meteorological reanalyzes from the CRU-NCEPv8 dataset (temperature,
precipitation, long- and shortwave incoming radiation, wind speed,
humidity and air pressure) were used to drive the LSMs. The TRENDY simulations
were performed in year 2017 and contributed to the global carbon budget
reported in Le Quéré et al. (2018). We used the results from the
S3 simulation from TRENDYv6 (with changing CO2, climate and land
use) over the period from 1982 to 2011, which is a time period consistent with other products
derived from remote-sensing-based physical models and machine-learning
algorithms.
Description of other datasets
To quantify the contributions of vegetation greening to terrestrial ET
variations, we used the LAI from the TRENDYv6 S3 simulation. We also used the
newest version of the Global Inventory Modeling and Mapping Studies LAI data
(GIMMS LAI3gV1) as the satellite-derived LAI. GIMMS LAI3gV1 was generated from
AVHRR GIMMS NDVI3g using a model derived from an artificial neural network (ANN; Zhu et al., 2013). It covers the period from 1982 to 2016 with
bimonthly frequency and has a 1/12∘ spatial resolution. To achieve
a uniform resolution, all data were resampled to 1/2∘ using the
nearest neighbor method. Following N. Pan et al. (2018), grids with an
annual mean NDVI of less than 0.1 were assumed to be non-vegetated regions and
were, therefore, masked out. NDVI data are from the GIMMS NDVI3gV1 dataset.
Temperature, precipitation and radiation are from the CRU-NCEPv8 dataset.
Statistical analysis
The significance of ET trends was analyzed using the Mann-Kendall (MK) test
(Kendall, 1955; Mann, 1945). It is a rank-based nonparametric method
that has been widely applied for detecting trends in hydro-climatic time
series (Sayemuzzaman and Jha, 2014; Yue et al., 2002). The Theil–Sen
estimator was applied to estimate the magnitude of the slope. The advantage
of this method over an ordinary least squares estimator is that it limits the
influence of the outliers on the slope (Sen, 1968).
Terrestrial ET interannual variability (IAV) is mainly controlled by variations in temperature,
precipitation and shortwave solar radiation (Zeng et al., 2018b; Zhang
et al., 2015). In this study, we performed partial correlation analyses
between ET and these three climatic variables at an annual scale for each
grid cell to explore climatic controls on ET IAV. Variability caused by
climatic variables was assessed through the square of partial correlation
coefficients between ET and temperature, precipitation and radiation. We
chose partial correlation analysis because it can quantify the linkage
between ET and a single environmental driving factor while controlling the
effects of the other remaining environmental factors. Partial correlation
analysis is a widely applied statistical tool to isolate the relationship
between two variables from the confounding effects of many correlated
variables (Anav et al., 2015; Jung et al., 2017; Peng et al., 2013). All
variables were first detrended in the statistical correlation analysis, as
we focus on the interannual relationship. The study period is from 1982 to
2011 for all models except MODIS and random forest, which have a temporal coverage that is
limited to 2001–2011 because of data availability.
To quantify the contribution of vegetation greening to terrestrial ET, we
separated the trend in terrestrial ET into four components induced by
climatic variables and vegetation dynamics by establishing a multiple linear
regression model between global ET and temperature, precipitation, shortwave
radiation and LAI (Eqs. 3, 4):
3δET=∂ET∂LAIδLAI+∂ET∂TδT+∂ET∂PδP+∂ET∂RδR+ε4δET=γETLAIδLAI+γETTδT+γETPδP++γETRδR+ε,
where γETLAI, γETT, γETP and γETR are the sensitivities of ET to the leaf area
index (LAI), air temperature (T), precipitation (P) and radiation (R),
respectively. ε is the residual, representing the
impacts of other factors.
After calculating γETLAI,
γETT, γETP and γETR, the contribution of the trend in factor i
(Trend(i)) to the trend in ET (Trend(ET)) can be
quantified as follows:
Contri(i)=γETi×Trendi/Trend(ET).
In performing multiple linear regression, we used the GIMMS LAI for both remote-sensing-based physical models and machine-learning methods as well as the individual
TRENDYv6 LAI for each TRENDY model. The gridded data of temperature,
precipitation and radiation are from CRU-NCEPv8.
ResultsThe ET magnitude estimated by multiple models
The multiyear ensemble mean of annual global terrestrial ET from
2001 to 2011 derived by machine-learning methods, remote-sensing-based
physical models and the TRENDY models agreed well, with values ranging from 589.6 to 617.1 mm yr-1. However, substantial differences existed
among individual models (Fig. 1). LPJ-wsl (455.3 mm yr-1) and LPX-Bern
(453.7 mm yr-1) estimated significantly lower ET than other models,
even in comparison with most previous studies focusing on earlier periods
(Table S1 in the Supplement). On the contrary, JULES gave the largest ET estimate (697.3 mm yr-1, which is equal to 7.57×104 km3 yr-1) among all
models, and showed an obvious increase in ET compared with its estimation
from 1950 to 2000 (6.5×104 km3 yr-1, Table S1).
Average annual global terrestrial ET estimated by each model
during the period from 2001 to 2011. Error bars represent the standard deviation of
each model. The four lines indicate the mean value of each category.
Spatial patterns of global terrestrial ET
As shown in Fig. 2, the spatial patterns of the multiyear average annual ET of
different categories were similar. ET was highest in the tropics and low
in the northern high latitudes and arid regions such as Australia, central Asia,
the western US and the Sahel. Compared with remote-sensing-based physical models
and LSMs, machine-learning methods obtained a smaller spatial gradient. In
general, latitudinal profiles of ET estimated using different approaches were
also consistent (Fig. 3). However, machine-learning methods gave higher ET
estimate at high latitudes and lower ET in the tropics compared with other
approaches. In the tropics, LSMs have significantly larger uncertainties than
benchmark products, and the standard deviation of LSMs is about 2 times
higher than that of benchmark products (Fig. 3). At other latitudes, LSMs and
benchmark ET products generally have comparable uncertainties. The largest
difference in ET in the different categories was found in the Amazon Basin (Fig. 2). In most regions of the Amazon Basin, the mean ET of remote sensing
physical models is more than 200 mm yr-1 higher than the mean ET of
LSMs and machine-learning methods. For individual ET estimates, the largest
uncertainty was also found in the Amazon Basin. MODIS, VISIT and CLASS-CTEM
estimated that the annual ET was higher than 1300 mm in the majority of Amazon,
whereas JSBACH and LPJ-wsl estimated an ET value lower than 800 mm yr-1
(Fig. S1). As is shown in Fig. S2, the differences in ET estimates among
TRENDY models were larger than those among benchmark estimates for tropical
and humid regions. The uncertainty of ET estimates from LSMs is particularly
large in the Amazon Basin, where the standard deviation of LSM estimates is
more than 2 times larger than that of benchmark estimates. It is
noteworthy that, in arid and semiarid regions such as western Australia,
central Asia, northern China and the western US, the difference in ET estimates
among LSMs is significantly smaller than those among remote sensing models
and machine-learning algorithms.
Spatial distributions of the mean annual ET derived from (a) remote-sensing-based physical models, (b) machine-learning algorithms, (c) benchmark datasets and (d) the TRENDY LSMs' ensemble mean, respectively.
Latitudinal profiles of the mean annual ET for different categories of
models. Each line represents the mean value of the corresponding category,
and the shading represents the interval of 1 standard deviation.
Interannual variations in global terrestrial ET
The ensemble mean interannual variability (IAV) of remote sensing ET
estimates and LSM ET estimates showed similar spatial patterns (Fig. 4).
Both remote sensing physical models and LSMs presented low IAV in ET in the
northern high latitudes but high IAV in ET in the southwestern US, India, sub-Saharan Africa, the Amazon and Australia. In contrast, the IAV of machine-learning-based ET was much weaker. In most regions, the IAV of machine-learning ET is
lower than 40 % of the IAV of remote sensing physical ET and LSM ET, and
this phenomenon is especially pronounced in tropical regions. Further
investigation into the spatial patterns of ET IAV for individual models
showed that the two machine-learning methods performed equally with respect to estimating
spatial patterns of ET IAV (Fig. S4). In contrast, differences in ET IAV
among remote sensing physical estimates and LSM estimates were much larger.
LSMs showed the largest differences in the IAV of ET in tropical regions. For
example, CABLE and JULES obtained an ET IAV of less than 15 mm yr-1
in most regions of the Amazon Basin, whereas LPJ-GUESS predicted an ET IAV of
more than 60 mm yr-1. Figure 5 shows that remote sensing physical ET and LSM ET had comparable
IAV north of 20∘ S, but the IAV of the machine-learning-based ET was much lower in this region. In the
region south of 20∘ S, TRENDY ET showed the largest IAV,
followed by those of remote sensing physical ET and machine-learning
estimates. The three approaches agreed on that ET IAV in the Southern
Hemisphere was generally larger than that in the Northern Hemisphere.
Spatial distributions of the interannual variability in ET
derived from (a) remote-sensing-based physical models, (b) machine-learning
algorithms, (c) benchmark datasets and (d) the TRENDY LSMs' ensemble mean,
respectively. The study period used in this study for the interannual
variability analysis is from 1982 to 2011.
Latitudinal profiles of ET IAV for different categories of models.
Each line represents the mean value of the corresponding category, and the
shading represents the interval of 1 standard deviation.
Climatic controls on ET
According to the ensemble remote sensing models, temperature and radiation
dominated ET IAV in northern Eurasia, northern and eastern North America,
southern China, the Congo River basin, and the southern Amazon River basin,
while precipitation dominated ET IAV in arid regions and semiarid regions
(Fig. 6a). The ensemble machine-learning algorithms had a similar pattern,
but they suggested a stronger control of radiation in the Amazon Basin and a
weaker control of precipitation in several arid regions such as central Asia
and northern Australia (Fig. 6b). In comparison, the ensemble LSMs suggested
the strongest control of precipitation on ET IAV (Fig. 6). According to the
ensemble LSMs, ET IAV was dominated by precipitation IAV in most regions of
the Southern Hemisphere and at northern low latitudes. Temperature and
radiation only controlled northern Eurasia, eastern Canada and part of the
Amazon Basin (Fig. 6d). As is shown in Fig. S6, the majority of LSMs agreed
on the dominant role of precipitation in controlling ET in regions south of
40∘ N. However, the pattern of climatic controls in the
ORCHIDEE-MICT model is quite unique and different from all of the other LSMs.
According to the ORCHIDEE-MICT model, radiation and temperature dominate ET
IAV in more regions, and precipitation only controls ET IAV in eastern
Brazil, northern Russia, central Europe and a part of tropical Africa. As
ORCHIDEE-MICT was developed from ORCHIDEE, the dynamic root parameterization
in ORCHIDEE-MICT may explain why ET is less driven by precipitation compared
with ORCHIDEE (Haverd et al., 2018). It is noted that
two machine-learning algorithms, MTE and RF, showed significant discrepancies in
the spatial pattern of dominant climatic factors. According to the result from MTE, temperature controlled ET IAV in regions north of 45∘ N, the eastern US, southern China and the Amazon Basin (Fig. S6e). By contrast,
RF suggested that precipitation and radiation dominated ET IAV in these
regions (Fig. S6f).
Spatial distributions of climatic controls on the interannual
variation of ET derived from the ensemble means of remote-sensing-based
physical models (a), machine-learning algorithms (b), benchmark data (c) and the TRENDY LSMs (d). (Red denotes temperature, green denotes precipitation and blue denotes
radiation.)
Long-term trends in global terrestrial ET
All approaches suggested an overall increasing trend in global ET during the
period from 1982 to 2011 (Fig. 7), although ET decreased from 1998 to 2009. This result
is consistent with previous studies (Jung et al., 2010; Lian et al.,
2018; Zhang et al., 2015). Remote sensing physical models indicated the
largest increase in ET (0.62 mm yr-2), followed by the machine-learning
method (0.38 mm yr-2) and land surface models (0.23 mm yr-2).
The mean ET of all categories except LSMs significantly increased during the
study period (p<0.05). It is noted that the ensemble mean ET values of
different categories are statistically correlated with each other
(p<0.001), even if the driving forces of different ET approaches are
different.
Interannual variations in global terrestrial ET estimated using
different categories of approaches.
Interannual variability (IAV – denoted as standard deviation) and the trend of global terrestrial ET from 1982 to 2011 and the contribution of
vegetation greening to the ET trend. (RS refers to remote sensing.)
∗ Suggests significance at the
95 % confidence level (p<0.05).
All remote sensing and machine-learning estimates indicate a significant
increasing trend in ET during the study period (p<0.05), although
the increase rate of P-LSH (1.07 mm yr-2) is more than 3 times as
large as that of GLEAM (0.33 mm yr-2). Nevertheless, there is a larger
discrepancy among LSMs in terms of the ET trend. The majority of LSMs (10 of 14)
suggest an increasing trend with an average trend of 0.34 mm yr-2
(p<0.05), and eight of them are statistically significant (see Table 2). However, four LSMs (JSBACH, JULES, ORCHIDEE and ORCHIDEE-MICT) suggest a
decreasing trend with an average trend of -0.12 mm yr-2
(p>0.05). Among the four decreasing trends, only the trend of
ORCHIDEE-MICT (-0.34 mm yr-2) is statistically significant
(p<0.05).
Spatial distributions of ET trends from 1982 to 2011
derived from (a) remote-sensing-based physical models, (b) machine-learning algorithms, (c) benchmark datasets and (d) the TRENDY LSMs' ensemble mean,
respectively. Regions with nonsignificant trends were excluded.
According to Fig. 8, the ensemble means of all the three approaches showed
increasing trends in ET over western and southern Africa, western India
and northern Australia, and decreasing ET over the western US,
southern South America and Mongolia. Discrepancies in ET trends mainly
appeared in eastern Europe, eastern India and central China. LSMs also
suggested a larger area of decreasing ET in both North America and South
America. Although the differences in ET trends among individual models were
larger than those among the ensemble means of different approaches, the
majority of models agreed that ET increased in western and southern Africa,
and decreased in the western US and southern South America (Fig. S2).
For both remote sensing estimates and LSM estimates, ET trends in the Amazon
Basin had large uncertainty: P-LSH, CLM45 and VISIT suggested a large area of
increasing ET, whereas GLEAM, JSBACH and ORCHIDEE suggested a large
area of decreasing ET.
Impacts of vegetation changes on ET variations
During the period from 1982 to 2011, global LAI trends estimated from remote sensing
data and from the ensemble LSMs are 2.51×10-3 m2 m-2 yr-1 (p<0.01) and 4.63×10-3 m2 m-2 yr-1 (p<0.01), respectively (Table 2). All LSMs
suggested a significant increasing trend in global LAI (greening). For both benchmark estimates and LSM estimates, it was
found that the spatial
pattern of trends in ET matched well with that of trends in the LAI (Figs. 8c, d, S5a, b), indicating significant effects of vegetation dynamics on ET
variations. According to the results of the multiple linear regression, all
models agreed that greening of the Earth since the early 1980s intensified
terrestrial ET (Table 2), although there was a significant discrepancy in
the magnitude of ET intensification which varied from 0.04 to
0.70 mm yr-2. The ensemble LSMs suggested a smaller ET increase (0.23 mm yr-2) than the ensemble remote sensing physical models (0.62 mm yr-2) and machine-learning algorithm (0.38 mm yr-2). Nevertheless,
the greening-induced ET intensification estimated by LSMs (0.37 mm yr-2) is larger than that estimated by remote sensing models (0.28 mm yr-2) and machine-learning algorithms (0.09 mm yr-2), as LSMs
suggested a stronger greening trend than remote sensing models. The
contribution of vegetation greening to ET intensification estimated by the
ensemble LSMs is larger than 100 %, whereas the contributions estimated by
the ensemble remote sensing physical models (0.62 mm yr-2) and
machine-learning algorithm are smaller than 50 %. Although TRENDY LSMs
were driven by the same climate data and remote sensing physical models were
driven by varied climate data, TRENDY LSMs still showed a larger discrepancy
in terms of the effect of vegetation greening on terrestrial ET than remote
sensing physical models due to the significant differences in both LAI
trends (1.74–13.63×10-3 m2 m-2 yr-1) and the
sensitivities of ET to the LAI (4.04–217.39 mm yr-2 m-2 m-2). In comparison, remote sensing physical models had smaller
discrepancies in terms of the sensitivity of ET to LAI (55.78–143.43 mm yr-2 m-2 m-2).
Discussion and perspectivesSources of uncertaintyUncertainty in the ET estimation of the Amazon Basin
LSMs show large discrepancies in the magnitude and trend of ET in the Amazon
Basin (Figs. 3, S3). However, it is challenging to identify the
uncertainty sources. Given that the TRENDY LSMs used uniform meteorological
inputs, the discrepancies in ET estimates among the participating models
mainly arise from the differences in the underlying model structures and
parameters. One potential source of uncertainty is the parameterization of
root water uptake. In the Amazon Basin, a deep root depth has been confirmed using field measurements (Nepstad et al., 2004). However, many LSMs have an
unrealistically shallow rooting depth (generally less than 2 m), thereby neglecting the existence and significance of deep roots. The incorrect root
distributions enlarge the differences in plant available water and root
water uptake, producing large uncertainties in ET. In addition, differences
in the parameterization of other key processes pertinent to ET such as LAI
dynamics (Fig. S5), canopy conductance variations (Table 1), water movements
in the soil (Abramopoulos et al., 1988; Clark et al., 2015; Noilhan and
Mahfouf, 1996) and soil moisture's control on transpiration (Purdy et
al., 2018; Szutu and Papuga, 2019) also increase the uncertainty in ET. The
abovementioned processes are not independent of each other but interact in
complex ways to produce the end result.
Uncertainty in the ET estimation of arid and semiarid regions
In arid and semiarid regions, benchmark products show much larger
differences in the magnitude of ET than LSMs (Fig. S2). One cause of this
phenomenon is the difference in meteorological forcing. Remote sensing and
machine-learning datasets used different forcing data. For precipitation, RF
used the CRUNCEPv6 dataset, MTE used the Global Precipitation Climatology
Centre (GPCC) dataset, MODIS used the Global Modeling and Assimilation
Office (GMAO) dataset, GLEAM used the Multi-Source Weighted-Ensemble
Precipitation (MSWEP) dataset, PML-CSIRO used the Princeton Global Meteorological Forcing (PGF) and the WATCH Forcing Data ERA-Interim (WFDEI) datasets, and P-LSH
used data derived from four independent sources. As precipitation is the
key climatic factor controlling ET in arid and semiarid regions (Fig. 6),
discrepancies between different forcing precipitation (Sun et al., 2018)
may be the main source of large uncertainty there. In comparison, the
uniform forcing data reduced the inter-model range in ET estimates from the TRENDY
LSMs. Nevertheless, it is noted that the congruence across LSM ET estimates
does not necessarily mean that they are the correct representation of ET. The
narrower inter-model range may suggest shared biases. All remote sensing
models and machine-learning algorithms except GLEAM do not explicitly take
the effects of soil moisture into account (Table S1). Given that soil
moisture is pivotal to both canopy conductance and soil evaporation in arid
and semiarid regions (A et al., 2019; De Kauwe et al., 2015; Medlyn et
al., 2015; Purdy et al., 2018), the lack of soil moisture information also
increases the bias in ET estimation. In addition, the accuracy of
remote sensing data itself is also an uncertainty source. The retrieval of
key land surface variables, such as the leaf area index and surface temperature,
is influenced by vegetation architecture, solar zenith angle and satellite
observational angle, particularly over heterogeneous surface (Norman and
Becker, 1995).
Uncertainty in the ET IAV in the Southern Hemisphere
In regions south of 20∘ S (including Australia, southern
Africa and southern South America), the ET IAV of remote sensing models and
machine-learning algorithms is lower than that of LSMs (Figs. 4, 5),
although their spatial patterns are similar. In these regions, GLEAM, the
only remote sensing model that explicitly considers the effects of soil
moisture, has larger ET IAV than other remote sensing models and has
similar ET IAV to LSMs (Fig. S4). This could imply that most of the existing
remote sensing models may underestimate ET IAV in the Southern Hemisphere
because the effects of soil moisture are not explicitly considered. Machine-learning algorithms show much lower IAV than other models (Figs. 4, S4). The main reason for this is that ET IAV is partly neglected
in the training process, as the magnitude of ET IAV
is usually smaller than the spatial and seasonal variability (Anav et
al., 2015; Jung et al., 2019). Moreover, the IAV of satellite-based key land
surface variables such as the LAI, fAPAR and surface temperature may be not
reliable because of the effects of clouds, which, in turn, affects the estimation
of the IAV of satellite-based ET. It is noted that LSM ET IAV shows large
differences at latitudes south of 20∘ S (Fig. 5). This
divergence in ET IAV indicates that LSMs require better representation of the ET
response to climate in the Southern Hemisphere.
Uncertainty in the global ET trend
All three categories of ET models detected an overall increasing trend
in global terrestrial ET since the early 1980s, which is in agreement with
previous studies (Mao et al., 2015; Miralles et al., 2014; Zeng et al.,
2018a, b, 2014; Zhang et al., 2015; Y. Zhang et al., 2016). Benchmark products generally suggested stronger ET
intensification than LSMs. The weaker ET intensification in LSMs may be
induced by the response of stomatal conductance to the increasing atmospheric
CO2 concentration. Increasing CO2 affects ET in two ways: on the
one hand, increasing CO2 can effectively reduce stomatal conductance
and, thus, decrease transpiration (Heijmans et al., 2001; Leipprand and
Gerten, 2006; Swann et al., 2016); on the other hand, it can increase
vegetation productivity and, thus, increase LAI. For benchmarks, the second
effect could be captured by remotely sensed LAI, NDVI or fAPAR, whereas the
first effect was neglected by all models except P-LSH (Zhang et al.,
2015). In contrast, both effects were modeled in all TRENDY LSMs.
LAI dynamics have significant influences on ET. The increased LAI trend
(greening) since the early 1980s has been reported by previous studies (Mao
et al., 2016; Zhu et al., 2016) and has also been confirmed by remote sensing data
and all of the TRENDY LSMs used in this study (Table 2, Fig. S5). Zhang et
al. (2015) found that the increasing trend of global terrestrial ET from
1982 to 2013 was mainly driven by an increase in the LAI and the enhanced
atmosphere water demand. Using a land–atmosphere coupled global climate
model (GCM), Zeng et al. (2018b) further estimated that global LAI
increased by about 8 %, resulting in an increase of 0.40±0.08 mm yr-2 in global ET (contributing to 55%±25 % of the ET
increase). This number is close to the estimates of ensemble LSMs
(0.37±0.18 mm yr-2). In comparison, the remote sensing models and
machine-learning algorithms used in this study suggested smaller
greening-induced ET increases. It is noted that TRENDY LSMs still showed a
larger discrepancy in terms of the effect of vegetation greening on
terrestrial ET than remote sensing physical models (Table 2) due to the
significant differences in the LAI trend (1.74–13.63×10-3 m2 m-2 yr-1) and in the sensitivity of ET to LAI (4.04–217.39 mm yr-2 m-2 m-2). Uncertainties in the LAI trend may arise from
inappropriate carbon allocations and deficiencies in responding to water
deficits (Anav et al., 2013; Hu et al., 2018; Murray-Tortarolo et al.,
2013; Restrepo-Coupe et al., 2017). Additionally, for machine-learning
algorithms, the results from insufficient long-term in situ measurements and
sparse observations in tropical, boreal and arid regions imply that there
are likely deficiencies in representing the temporal variations.
Lack of knowledge of the effects of irrigation
Irrigation accounts for about 90 % of human consumptive water use and
largely affects ET in irrigated croplands (Siebert et al., 2010). Global
water withdrawals for irrigation were estimated to be within the range of
1161–3800 km3 yr-1 around the year 2000, and they largely increased
during the period from 2000 to 2014 (Chen et al., 2019). However, none of the
remote-sensing-based physical models and machine-learning algorithms
explicitly accounted for the effects of irrigation on ET, although these
effects could be taken into account to some extent by using observed LAI,
NDVI or fAPAR to drive the models (Zhang et al., 2015). Considering
that annual ET may surpass annual precipitation in cropland areas, Y. Zhang et
al. (2016) only used the Budyko hydrometeorological model to constrain the PML-CSIRO
model in grids covered by non-crop vegetation. However, the process of
irrigation affecting evaporation was still not taken into consideration. For
TRENDY LSMs, only 2 of 14 models (DLEM and ISAM) included irrigation
processes (Le Quéré et al., 2018). Therefore, the effects of
irrigation are largely neglected in existing global ET datasets, which
reduces the accuracy of local ET estimates in regions with a large
proportion of irrigated cropland.
ET variability across the precipitation gradient and its planetary
boundary
Precipitation is the source of terrestrial ET. According to
Fig. 9a, the vast majority of models agree that ET has the largest IAV in
regions with annual precipitation between 700 and 1000 mm, although the
magnitude of ET IAV has substantial discrepancies among different models.
The low ET IAV in arid and semiarid regions does not mean that ET is stable in
these regions. In fact, ET has the largest coefficient of variation (CoV –
the ratio of ET standard deviation to ET mean value) in arid regions, and
all models show a clear negative trend in the CoV with increasing precipitation
(Fig. 9b). This is mainly caused by the large CoV of precipitation in arid
regions (Fatichi et al., 2012).
Interannual variability (a) and the coefficient of variation (b) of
ET for each 50 mm interval of mean annual precipitation.
In comparison, terrestrial ET shows a much lower IAV at the global scale
(Table 2), with values ranging from 4.8 to 12.2 mm yr-1 (1 standard deviation),
which only equates to 1.0 %–1.8 % of the global annual mean ET. The model results
suggest that global terrestrial ET stabilizes at about 6.74×104 km3 yr-1 (603 mm yr-1), which is close to previous
estimates (Alton et al., 2009; Mueller et al., 2011; Oki and Kanae, 2006;
Zeng et al., 2012). The stability of global terrestrial ET is probably based
on partitioning the solar constant and suggests that
droughts in one place are balanced by excess rain in other places; thus, it all
evens out at a global scale. This implies that ET also has a potential planetary boundary, a
suggestion made by Running (2012) on net primary production (NPP) as a planetary boundary. ET
integrates four aspects of the current planetary boundaries defined by
Steffen et al. (2015): climate change, freshwater use,
land-system change and biochemical flows. Given the importance of ET with respect to linking
terrestrial water, carbon, nutrient and energy cycles, more studies on the
ET planetary boundary are needed in the context of intensifying global
change and increasing anthropogenic perturbations to the Earth system.
In short, the multi-model intercomparison indicates that considerable
uncertainty exists in both the temporal and spatial variations in global ET
estimates, although a large portion of models adopt similar ET algorithms
(Table 1). The major uncertainty source is different for different types of
models and regions. The uncertainty is induced by multiple factors,
including problems pertinent to parameterization of land processes, lack of
in situ measurements, remote sensing acquisition, scaling effects and
meteorological forcing. Based on the results of different approaches, we
suggest that global terrestrial ET also has a potential planetary boundary,
with the value being about 6.74×104 km3 yr-1 (603 mm yr-1), which is consistent with previous estimates.
Recommendations for future developmentRemote-sensing-based physical methods
Over the past few decades, the development of remote sensing technologies has
contributed to a boom in various ET estimation methods. However, there is
still much room for remote sensing technologies to improve (Fisher et
al., 2017). The development of new platforms and sensors that have improved global
spatiotemporal coverage and using multiband, multisource remote sensing
data are the key points. Planned or newly launched satellites, such as
NASA's GRACE Follow-On (GRACE-FO) mission and the ECOsystem Spaceborne Thermal
Radiometer Experiment on Space Station (ECOSTRESS) mission, will improve the
accuracy of terrestrial ET estimates. ECOSTRESS's thermal infrared (TIR)
multispectral scanner is capable of monitoring diurnal temperature patterns
at high resolutions, which provides insights into plant response to water
stress as well as the means to understand sub-daily ET dynamics (Hulley et al., 2017).
GRACE Follow-On observations can be used to constrain subsurface lateral
water transfers, which helps to correct soil moisture and subsequently
improves the accuracy of ET estimates (Rouholahnejad and Martens,
2018). Moreover, building integrated methods that fuse different ET estimates
or the upstream satellite-based biophysical variables from different
platforms and the other forcing data will be helpful to improve the accuracy
and spatiotemporal coverage of ET (Ke et al., 2016; Ma et al., 2018;
Semmens et al., 2016).
The theories and retrieval algorithms of ET and the related key biophysical
variables also need to be further improved. For example, the method for calculating canopy conductance may be improved by integrating remote-sensing-based solar-induced chlorophyll fluorescence (SIF) data. SIF data from
existing Global Ozone Monitoring Experiment-2 (GOME-2), Orbiting Carbon
Observatory-2 (OCO-2) and TROPOspheric Monitoring Instrument (TROPOMI) satellite instruments as well as
the forthcoming OCO-3 and Geostationary Carbon Cycle Observatory (GeoCarb)
satellites provide a good opportunity to diagnose transpiration and to examine ET partitioning at multiple spatiotemporal scales (Pagán et al.,
2019; Stoy et al., 2019; Sun et al., 2017). Theoretical advancements in
nonequilibrium thermodynamics and maximum entropy production (MEP) could be
incorporated into the classical ET theories (Xu et al., 2019; K. Zhang et
al., 2016). In addition, quantifying the effects of CO2 fertilization
on stomatal conductance is pivotal for remote sensing models to capture the
long-term trend of terrestrial ET.
Most existing remote-sensing-based ET studies have focused on total ET; however,
the partitioning of ET between transpiration, soil evaporation, and canopy
interception may show significant divergence even though the total ET is
accurately estimated (Talsma et al., 2018b). In current
remote-sensing-based ET models, soil evaporation, which is sensitive to
precipitation events and soil moisture, is the component with the largest error
(Talsma et al., 2018a). Therefore, incorporating the increasingly
accessible satellite-based precipitation, soil moisture observations and
soil property data will contribute to the improvement of the soil evaporation
estimation. Meanwhile, the consideration of soil evaporation under
herbaceous vegetation and canopy will also reduce the errors.
Machine-learning methods
It is well known that the capability of machine-learning algorithms to
provide accurate ET estimates largely depends on the representativeness of the training datasets with respect to describing ecosystem behaviors (Yao et al., 2017).
As a result, machine-learning algorithms may not perform well outside the
range of the data used for their training. Unfortunately, long-term field
observations in northern temperate regions are still insufficient. This
is an important cause of the small spatial gradient and small IAV of
machine-learning ET. Given that remote sensing is capable of providing broad
coverage of key biophysical variables at reasonable spatial and temporal
resolutions, one way to overcome this challenge is to exclusively use remote
sensing observations as training data (Jung et al., 2019; Poon and
Kinoshita, 2018). Another simple way to make the IAV of machine-learning ET
more realistic is to normalize the yearly anomalies when comparing them with ET
estimates from LSMs and remote sensing physical models (Jung et al.,
2019). New machine-learning techniques, including the extreme learning
machine and the adaptive neuro-fuzzy inference system, can be used to
improve the accuracy of ET estimation (Gocic et al., 2016;
Kişi and Tombul, 2013). Emerging deep-learning methods
such as recurrent neural network (RNN) and long short-term memory (LSTM)
have the potential to outcompete conventional machine-learning methods in
modeling ET time series (Reichstein et al., 2018,
2019). Almost all machine-learning datasets used precipitation rather than soil
moisture as the explanatory variable when training. However, soil moisture
(rather than precipitation) directly controls ET. As more and more global
remote-sensing-based soil moisture datasets become available, using soil
moisture products as input is expected to improve the accuracy of ET
estimates, especially for regions with sparse vegetation coverage (Xu
et al., 2018).
Land surface models
In contrast to observation-based methods, LSMs are able to project future
changes in ET, and can disentangle the effects of different drivers of ET
through factorial analysis. However, results from LSMs are only as good as
their parameterizations of complex land surface processes, which are limited
by our incomplete understanding of physical and biological processes
(Niu et al., 2011). Although TRENDY LSMs are the state-of-the-art
process-based global land surfaces models, improvements are still needed
because several important processes are missing or not being appropriately
parameterized. Most of the TRENDY LSMs did not simulate processes
relevant to human management, including irrigation (Chen et al., 2019)
and the application of fertilizers (Mao et al., 2015), or natural
disturbances like wildfire (Poon and Kinoshita, 2018). Incorporating
these processes into present LSMs is critical, although the introduction of new
model parameters also potentially leads to an increase in a model's
uncertainty.
In light of the importance of soil water availability in constraining canopy
conductance and dynamics, the accurate representation of hydrological processes
is a core task for LSMs, particularly in dry regions. Integrating a dynamic
root water uptake function and hydraulic redistribution into the LSM can
significantly improve its performance with respect to estimating seasonal ET and soil
moisture (Li et al., 2012). Moreover, other hydrological processes
including groundwater (Decker, 2015), lateral flow (Rouholahnejad
and Martens, 2018) and water vapor diffusion at the soil surface
(Chang et al., 2018) need to be simulated and correctly
represented to reproduce the dynamics of soil water and ET. As the canopy LAI
plays an important role in regulating ET, correctly simulating vegetation
dynamics is also critical. One way to do this is to correct the initialization,
distribution and parameterization of vegetation phenology in LSMs
(Murray-Tortarolo et al., 2013; Zhang et al., 2019). An appropriate carbon
allocation scheme and parameterization of vegetation's response to water
deficits are also important for reproducing vegetation dynamics (Anav
et al., 2013).
Conclusion
In this study, we evaluated 20 global terrestrial ET estimates including
4 from remote-sensing-based physical models, 2 from machine-learning
algorithms and 14 from TRENDY LSMs. The ensemble mean values of global
terrestrial ET for the three categories agreed well, with values ranging from 589.6 to 617.1 mm yr-1. All three categories detected an
overall increasing trend in global ET during the period from 1982 to 2011 and
suggested a positive effect of vegetation greening on ET intensification.
However, the multi-model intercomparison indicates that considerable
uncertainties still exist in both temporal and spatial variations in global
terrestrial ET estimates. LSMs showed significant differences in the ET
magnitude in tropical regions, especially in the Amazon Basin, whereas benchmark
ET products showed a larger inter-model range in arid and semiarid regions
than LSMs. Trends in ET estimates also showed significant discrepancies
among LSMs. These uncertainties are induced by the parameterization of land
processes, meteorological forcings, the lack of in situ measurements, remote
sensing acquisition and scaling effects. Model developments and
observational improvements provide two parallel pathways towards improving
the accuracy of global terrestrial ET estimation.
Code and data availability
TRENDYv6 data are available from Stephen Sitch (s.a.sitch@exeter.ac.uk) upon reasonable
request. MODIS ET data are available from
http://files.ntsg.umt.edu/data/NTSG_Products/MOD16/ (Running, 2020). GLEAM ET
are available from https://www.gleam.eu/ (Miralles, 2020). Both model tree ensemble and
random forest ET data are available from
https://www.bgc-jena.mpg.de/geodb/projects/FileDetails.php (Jung, 2020). P-LSH ET data are
available from http://files.ntsg.umt.edu/data/ET_global_monthly/Global_8kmResolution/ (K. E. Zhang, 2020).
PML-CSIRO ET data are available from
https://data.csiro.au/dap/landingpage?pid=csiro:17375 (Y. Zhang, 2020). CRU-NCEPv8 data are
available from Nicolas Viovy upon reasonable request (email address: viovy@dsm-mail.saclay.cea.fr). GIMMS LAI3gV1 data are
available from Ranga B. Myneni upon reasonable request (email address: rmyneni@bu.edu). GIMMS NDVI3gV1 data are available from https://ecocast.arc.nasa.gov/data/pub/gimms/3g.v1/ (Pinzon and Tucker, 2020).
The supplement related to this article is available online at: https://doi.org/10.5194/hess-24-1485-2020-supplement.
Author contributions
SP initiated this research and was responsible for the integrity of the
work as a whole. NP carried out the analyses. SP, NP, HT and HS
wrote the paper with contributions from all authors. PF, SS,
VKA, VH, AKJ, EK, SL, DL, JEMSN, CO, BP, HT and
SZ contributed to the TRENDY results. SWR provided the MODIS ET dataset.
Competing interests
The authors declare that they have no conflict of interest.
Acknowledgements
We thank all of the contributors who provided the
data used in this study, particularly the TRENDY modeling groups.
Additional details on funding support for the participating 14 land surface
models is provided by the TRENDY Project.
Financial support
This research has been supported by the National Science Foundation (grant nos. 1903722, 1243232 and AGS 12-43071), the US Department of Energy (grant no. DE-SC0016323), the OUC-AU Joint Center program, the Auburn
University IGP program, the National Institute of Food and Agriculture/US Department of
Agriculture (grant nos. 2015-67003-23489 and 2015-67003-23485), EC H2020 (CCiCC; grant no. 821003), SNSF (grant no. 20020_172476), and the Earth
Systems and Climate Change Hub (funded by the Australian Government's
National Environmental Science Program).
Review statement
This paper was edited by Pierre Gentine and reviewed by two anonymous referees.
References
A, Y., Wang, G., Liu, T., Xue, B., and Kuczera, G.: Spatial variation of
correlations between vertical soil water and evapotranspiration and their
controlling factors in a semi-arid region, J. Hydrol., 574, 53–63,
2019.
Abramopoulos, F., Rosenzweig, C., and Choudhury, B.: Improved ground
hydrology calculations for global climate models (GCMs): Soil water movement
and evapotranspiration, J. Climate, 1, 921–941, 1988.
Allen, R. G., Pereira, L. S., Raes, D., and Smith, M.: Crop
evapotranspiration-Guidelines for computing crop water requirements – FAO
Irrigation and drainage paper 56, Fao, Rome, Italy, 300, D05109, 1998.Alton, P., Fisher, R., Los, S., and Williams, M.: Simulations of global
evapotranspiration using semiempirical and mechanistic schemes of plant
hydrology, Global Biogeochem. Cy., 23, GB4023, 10.1029/2009GB003540, 2009.
Anav, A., Murray-Tortarolo, G., Friedlingstein, P., Sitch, S., Piao, S., and
Zhu, Z.: Evaluation of land surface models in reproducing satellite Derived
leaf area index over the high-latitude northern hemisphere. Part II: Earth
system models, Remote Sens., 5, 3637–3661, 2013.
Anav, A., Friedlingstein, P., Beer, C., Ciais, P., Harper, A., Jones, C.,
Murray-Tortarolo, G., Papale, D., Parazoo, N. C., and Peylin, P.:
Spatiotemporal patterns of terrestrial gross primary production: A review,
Rev. Geophys., 53, 785–818, 2015.
Antonopoulos, V. Z., Gianniou, S. K., and Antonopoulos, A. V.: Artificial
neural networks and empirical equations to estimate daily evaporation:
application to lake Vegoritis, Greece, Hydrolog. Sci. J., 61,
2590–2599, 2016.
Barman, R., Jain, A. K., and Liang, M.: Climate-driven uncertainties in
modeling terrestrial energy and water fluxes: a site-level to global-scale
analysis, Glob. Change Biol., 20, 1885–1900, 2014.Bodesheim, P., Jung, M., Gans, F., Mahecha, M. D., and Reichstein, M.: Upscaled diurnal cycles of land–atmosphere fluxes: a new global half-hourly data product, Earth Syst. Sci. Data, 10, 1327–1365, 10.5194/essd-10-1327-2018, 2018.
Breiman, L.: Random forests, Mach. Learn., 45, 5–32, 2001.
Chang, L.-L., Dwivedi, R., Knowles, J. F., Fang, Y.-H., Niu, G.-Y.,
Pelletier, J. D., Rasmussen, C., Durcik, M., Barron-Gafford, G. A., and
Meixner, T.: Why Do Large-Scale Land Surface Models Produce a Low Ratio of
Transpiration to Evapotranspiration?, J. Geophys. Res.-Atmos., 123, 9109–9130, 2018.
Chen, Y., Xia, J., Liang, S., Feng, J., Fisher, J. B., Li, X., Li, X., Liu,
S., Ma, Z., and Miyata, A.: Comparison of satellite-based evapotranspiration
models over terrestrial ecosystems in China, Remote Sens. Environ.,
140, 279–293, 2014.
Chen, Y., Feng, X., Fu, B., Shi, W., Yin, L., and Lv, Y.: Recent global
cropland water consumption constrained by observations, Water Resour. Res., 55, 3708–3738, 2019.
Clark, M. P., Fan, Y., Lawrence, D. M., Adam, J. C., Bolster, D., Gochis, D.
J., Hooper, R. P., Kumar, M., Leung, L. R., and Mackay, D. S.: Improving the
representation of hydrologic processes in Earth System Models, Water Resour. Res., 51, 5929–5956, 2015.
Cleugh, H. A., Leuning, R., Mu, Q., and Running, S. W.: Regional evaporation
estimates from flux tower and MODIS satellite data, Remote Sens. Environ., 106, 285–304, 2007.
Dai, A.: Precipitation characteristics in eighteen coupled climate models,
J. Climate, 19, 4605–4630, 2006.
Decker, M.: Development and evaluation of a new soil moisture and runoff
parameterization for the CABLE LSM including subgrid-scale processes,
J. Adv. Model. Earth Sy., 7, 1788–1809, 2015.De Kauwe, M. G., Kala, J., Lin, Y.-S., Pitman, A. J., Medlyn, B. E., Duursma, R. A., Abramowitz, G., Wang, Y.-P., and Miralles, D. G.: A test of an optimal stomatal conductance scheme within the CABLE land surface model, Geosci. Model Dev., 8, 431–452, 10.5194/gmd-8-431-2015, 2015.d'Orgeval, T., Polcher, J., and de Rosnay, P.: Sensitivity of the West African hydrological cycle in ORCHIDEE to infiltration processes, Hydrol. Earth Syst. Sci., 12, 1387–1401, 10.5194/hess-12-1387-2008, 2008.
Dou, X. and Yang, Y.: Evapotranspiration estimation using four different
machine learning approaches in different terrestrial ecosystems, Comput. Electron. Agr., 148, 95–106, 2018.
Douville, H., Ribes, A., Decharme, B., Alkama, R., and Sheffield, J.:
Anthropogenic influence on multidecadal changes in reconstructed global
evapotranspiration, Nat. Clim. Change, 3, 59–62, 2013.
Famiglietti, J. and Wood, E. F.: Evapotranspiration and runoff from large
land areas: Land surface hydrology for atmospheric general circulation
models, Surv. Geophys., 12, 179–204, 1991.
Fatichi, S., Ivanov, V. Y., and Caporali, E.: Investigating Interannual
Variability of Precipitation at the Global Scale: Is There a Connection with
Seasonality?, J. Climate, 25, 5512–5523, 2012.
Feng, Y., Cui, N., Gong, D., Zhang, Q., and Zhao, L.: Evaluation of random
forests and generalized regression neural networks for daily reference
evapotranspiration modelling, Agr. Water Manage., 193, 163–173,
2017.
Fisher, J. B., Tu, K. P., and Baldocchi, D. D.: Global estimates of the
land–atmosphere water flux based on monthly AVHRR and ISLSCP-II data,
validated at 16 FLUXNET sites, Remote Sens. Environ., 112, 901–919,
2008.
Fisher, J. B., Malhi, Y., Bonal, D., Da Rocha, H. R., De Araujo, A. C.,
Gamo, M., Goulden, M. L., Hirano, T., Huete, A. R., and Kondo, H.: The
land–atmosphere water flux in the tropics, Glob. Change Biol., 15,
2694–2714, 2009.
Fisher, J. B., Melton, F., Middleton, E., Hain, C., Anderson, M., Allen, R.,
McCabe, M. F., Hook, S., Baldocchi, D., and Townsend, P. A.: The future of
evapotranspiration: Global requirements for ecosystem functioning, carbon
and climate feedbacks, agricultural management, and water resources, Water Resour. Res., 53, 2618–2626, 2017.
Fu, B. P.: On the calculation of the evaporation from land surface, Sci.
Atmos. Sin., 5, 23–31, 1981.
Fu, G., Charles, S. P., and Yu, J.: A critical overview of pan evaporation
trends over the last 50 years, Climatic Change, 97, 193–214, 2009.
Gedney, N., Cox, P., Betts, R., Boucher, O., Huntingford, C., and Stott, P.:
Detection of a direct carbon dioxide effect in continental river runoff
records, Nature, 439, 835–838, 2006.
Glenn, E. P., Nagler, P. L., and Huete, A. R.: Vegetation index methods for
estimating evapotranspiration by remote sensing, Surv. Geophys., 31,
531–555, 2010.
Gocic, M., Petkoviæ, D., Shamshirband, S., and Kamsin, A.: Comparative
analysis of reference evapotranspiration equations modelling by extreme
learning machine, Comput. Electron. Agr., 127, 56–63,
2016.
Good, S. P., Noone, D., and Bowen, G.: Hydrologic connectivity constrains
partitioning of global terrestrial water fluxes, Science, 349, 175–177,
2015.Guimberteau, M., Zhu, D., Maignan, F., Huang, Y., Yue, C., Dantec-Nédélec, S., Ottlé, C., Jornet-Puig, A., Bastos, A., Laurent, P., Goll, D., Bowring, S., Chang, J., Guenet, B., Tifafi, M., Peng, S., Krinner, G., Ducharne, A., Wang, F., Wang, T., Wang, X., Wang, Y., Yin, Z., Lauerwald, R., Joetzjer, E., Qiu, C., Kim, H., and Ciais, P.: ORCHIDEE-MICT (v8.4.1), a land surface model for the high latitudes: model description and validation, Geosci. Model Dev., 11, 121–163, 10.5194/gmd-11-121-2018, 2018.Haverd, V., Smith, B., Nieradzik, L., Briggs, P. R., Woodgate, W., Trudinger, C. M., Canadell, J. G., and Cuntz, M.: A new version of the CABLE land surface model (Subversion revision r4601) incorporating land use and land cover change, woody vegetation demography, and a novel optimisation-based approach to plant coordination of photosynthesis, Geosci. Model Dev., 11, 2995–3026, 10.5194/gmd-11-2995-2018, 2018.Heijmans, M. M., Arp, W. J., and Berendse, F.: Effects of elevated CO2 and
vascular plants on evapotranspiration in bog vegetation, Glob. Change Biol., 7, 817–827, 2001.
Hu, Z., Shi, H., Cheng, K., Wang, Y. P., Piao, S., Li, Y., Zhang, L., Xia,
J., Zhou, L., and Yuan, W.: Joint structural and physiological control on
the interannual variation in productivity in a temperate grassland: A
data-model comparison, Glob. Change Biol., 24, 2965–2979, 2018.
Huang, S., Bartlett, P., and Arain, M. A.: Assessing nitrogen controls on
carbon, water and energy exchanges in major plant functional types across
North America using a carbon and nitrogen coupled ecosystem model,
Ecol. Model., 323, 12–27, 2016.
Hulley, G., Hook, S., Fisher, J., and Lee, C.: ECOSTRESS, A NASA
Earth-Ventures Instrument for studying links between the water cycle and
plant health over the diurnal cycle, 2017 IEEE International Geoscience and
Remote Sensing Symposium (IGARSS), 23–28 July 2017, Fort Worth, USA, 5494–5496, 2017.Ito, A.: Evaluation of the impacts of defoliation by tropical cyclones on a
Japanese forest's carbon budget using flux data and a process-based model,
J. Geophys. Res.-Biogeo., 115, G04013, 10.1029/2010JG001314, 2010.
Jackson, R., Reginato, R., and Idso, S.: Wheat canopy temperature: a
practical tool for evaluating water requirements, Water Resour. Res.,
13, 651–656, 1977.Jimenez, C., Prigent, C., Mueller, B., Seneviratne, S. I., McCabe, M., Wood,
E., Rossow, W., Balsamo, G., Betts, A., and Dirmeyer, P.: Global
intercomparison of 12 land surface heat flux estimates, J. Geophys. Res.-Atmos., 116, D02102, 10.1029/2010JD014545, 2011.Jung, M.: Latent heat flux on land, available at:
https://www.bgc-jena.mpg.de/geodb/projects/FileDetails.php, last access: 27 March 2020.Jung, M., Reichstein, M., and Bondeau, A.: Towards global empirical upscaling of FLUXNET eddy covariance observations: validation of a model tree ensemble approach using a biosphere model, Biogeosciences, 6, 2001–2013, 10.5194/bg-6-2001-2009, 2009.
Jung, M., Reichstein, M., Ciais, P., Seneviratne, S. I., Sheffield, J.,
Goulden, M. L., Bonan, G., Cescatti, A., Chen, J., and De Jeu, R.: Recent
decline in the global land evapotranspiration trend due to limited moisture
supply, Nature, 467, 951–954, 2010.Jung, M., Reichstein, M., Margolis, H. A., Cescatti, A., Richardson, A. D.,
Arain, M. A., Arneth, A., Bernhofer, C., Bonal, D., and Chen, J.: Global
patterns of land-atmosphere fluxes of carbon dioxide, latent heat, and
sensible heat derived from eddy covariance, satellite, and meteorological
observations, J. Geophys. Res.-Biogeo., 116, G00J07, 10.1029/2010JG001566, 2011.Jung, M., Reichstein, M., Schwalm, C. R., Huntingford, C., Sitch, S.,
Ahlström, A., Arneth, A., Camps-Valls, G., Ciais, P., and
Friedlingstein, P.: Compensatory water effects link yearly global land CO2
sink changes to temperature, Nature, 541, 516–520, 2017.Jung, M., Koirala, S., Weber, U., Ichii, K., Gans, F., Camps-Valls, G.,
Papale, D., Schwalm, C., Tramontana, G., and Reichstein, M.: The FLUXCOM
ensemble of global land-atmosphere energy fluxes, Scientific Data, 6, 74, 10.1038/s41597-019-0076-8,
2019.Ke, Y., Im, J., Park, S., and Gong, H.: Downscaling of MODIS One kilometer
evapotranspiration using Landsat-8 data and machine learning approaches, Remote Sens., 8, 215, 10.3390/rs8030215, 2016.Keller, K. M., Lienert, S., Bozbiyik, A., Stocker, T. F., Churakova (Sidorova), O. V., Frank, D. C., Klesse, S., Koven, C. D., Leuenberger, M., Riley, W. J., Saurer, M., Siegwolf, R., Weigt, R. B., and Joos, F.: 20th century changes in carbon isotopes and water-use efficiency: tree-ring-based evaluation of the CLM4.5 and LPX-Bern models, Biogeosciences, 14, 2641–2673, 10.5194/bg-14-2641-2017, 2017.
Kendall, M. G.: Rank correlation methods, Charles Griffin, London, UK, 196 pp., 1955.
Khan, M. S., Liaqat, U. W., Baik, J., and Choi, M.: Stand-alone uncertainty
characterization of GLEAM, GLDAS and MOD16 evapotranspiration products using
an extended triple collocation approach, Agr. Forest Meteorol., 252, 256–268, 2018.
Kişi, Ö. and Tombul, M.: Modeling monthly pan
evaporations using fuzzy genetic approach, J. Hydrol., 477,
203–212, 2013.
Kisi, O., Sanikhani, H., Zounemat-Kermani, M., and Niazi, F.: Long-term
monthly evapotranspiration modeling by several data-driven methods without
climatic data, Comput. Electron. Agr., 115, 66–77, 2015.
Knauer, J., Werner, C., and Zaehle, S.: Evaluating stomatal models and their
atmospheric drought response in a land surface scheme: A multibiome
analysis, J. Geophys. Res.-Biogeo., 120, 1894–1911,
2015.
Lawrence, D. M., Thornton, P. E., Oleson, K. W., and Bonan, G. B.: The
partitioning of evapotranspiration into transpiration, soil evaporation, and
canopy evaporation in a GCM: Impacts on land–atmosphere interaction,
J. Hydrometeorol., 8, 862–880, 2007.Leipprand, A. and Gerten, D.: Global effects of doubled atmospheric CO2
content on evapotranspiration, soil moisture and runoff under potential
natural vegetation, Hydrolog. Sci. J., 51, 171–185, 2006.Le Quéré, C., Andrew, R. M., Friedlingstein, P., Sitch, S., Pongratz, J., Manning, A. C., Korsbakken, J. I., Peters, G. P., Canadell, J. G., Jackson, R. B., Boden, T. A., Tans, P. P., Andrews, O. D., Arora, V. K., Bakker, D. C. E., Barbero, L., Becker, M., Betts, R. A., Bopp, L., Chevallier, F., Chini, L. P., Ciais, P., Cosca, C. E., Cross, J., Currie, K., Gasser, T., Harris, I., Hauck, J., Haverd, V., Houghton, R. A., Hunt, C. W., Hurtt, G., Ilyina, T., Jain, A. K., Kato, E., Kautz, M., Keeling, R. F., Klein Goldewijk, K., Körtzinger, A., Landschützer, P., Lefèvre, N., Lenton, A., Lienert, S., Lima, I., Lombardozzi, D., Metzl, N., Millero, F., Monteiro, P. M. S., Munro, D. R., Nabel, J. E. M. S., Nakaoka, S., Nojiri, Y., Padin, X. A., Peregon, A., Pfeil, B., Pierrot, D., Poulter, B., Rehder, G., Reimer, J., Rödenbeck, C., Schwinger, J., Séférian, R., Skjelvan, I., Stocker, B. D., Tian, H., Tilbrook, B., Tubiello, F. N., van der Laan-Luijkx, I. T., van der Werf, G. R., van Heuven, S., Viovy, N., Vuichard, N., Walker, A. P., Watson, A. J., Wiltshire, A. J., Zaehle, S., and Zhu, D.: Global Carbon Budget 2017, Earth Syst. Sci. Data, 10, 405–448, 10.5194/essd-10-405-2018, 2018.Leuning, R., Zhang, Y., Rajaud, A., Cleugh, H., and Tu, K.: A simple surface
conductance model to estimate regional evaporation using MODIS leaf area
index and the Penman-Monteith equation, Water Resour. Res., 44, W10419, 10.1029/2007WR006562, 2008.
Li, J., Wang, Y. P., Duan, Q., Lu, X., Pak, B., Wiltshire, A., Robertson,
E., and Ziehn, T.: Quantification and attribution of errors in the simulated
annual gross primary production and latent heat fluxes by two global land
surface models, J. Adv. Model. Earth Sy., 8, 1270–1288,
2016.Li, L., Wang, Y. P., Yu, Q., Pak, B., Eamus, D., Yan, J., Gorsel, E., and
Baker, I. T.: Improving the responses of the Australian community land
surface model (CABLE) to seasonal drought, J. Geophys. Res.-Biogeo., 117, G04002, 10.1029/2012JG002038, 2012.
Li, Z.-L., Tang, R., Wan, Z., Bi, Y., Zhou, C., Tang, B., Yan, G., and
Zhang, X.: A review of current methodologies for regional evapotranspiration
estimation from remotely sensed data, Sensors, 9, 3801–3853, 2009.
Lian, X., Piao, S., Huntingford, C., Li, Y., Zeng, Z., Wang, X., Ciais, P.,
McVicar, T. R., Peng, S., and Ottlé, C.: Partitioning global land
evapotranspiration using CMIP5 models constrained by observations, Nat. Clim. Change, 8, 640–646, 2018.
Liu, M., Tian, H., Chen, G., Ren, W., Zhang, C., and Liu, J.: Effects of
Land-Use and Land-Cover Change on Evapotranspiration and Water Yield in
China During 1900–2000, J. Am. Water Resour. As., 44, 1193–1207, 2008.
Ma, Y., Liu, S., Song, L., Xu, Z., Liu, Y., Xu, T., and Zhu, Z.: Estimation
of daily evapotranspiration and irrigation water efficiency at a
Landsat-like scale for an arid irrigation area using multi-source remote
sensing data, Remote Sens. Environ., 216, 715–734, 2018.
Mann, H. B.: Nonparametric tests against trend, Econometrica, 1945, 245–259, 1945.Mao, J., Fu, W., Shi, X., Ricciuto, D. M., Fisher, J. B., Dickinson, R. E.,
Wei, Y., Shem, W., Piao, S., and Wang, K.: Disentangling climatic and
anthropogenic controls on global terrestrial evapotranspiration trends,
Environ. Res. Lett., 10, 094008, 10.1088/1748-9326/10/9/094008, 2015.
Mao, J., Ribes, A., Yan, B., Shi, X., Thornton, P. E., Séférian, R.,
Ciais, P., Myneni, R. B., Douville, H., and Piao, S.: Human-induced greening
of the northern extratropical land surface, Nat. Clim. Change, 2016, 959–963, 2016.Martens, B., Miralles, D. G., Lievens, H., van der Schalie, R., de Jeu, R. A. M., Fernández-Prieto, D., Beck, H. E., Dorigo, W. A., and Verhoest, N. E. C.: GLEAM v3: satellite-based land evaporation and root-zone soil moisture, Geosci. Model Dev., 10, 1903–1925, 10.5194/gmd-10-1903-2017, 2017.
Maselli, F., Papale, D., Chiesi, M., Matteucci, G., Angeli, L., Raschi, A.,
and Seufert, G.: Operational monitoring of daily evapotranspiration by the
combination of MODIS NDVI and ground meteorological data: Application and
evaluation in Central Italy, Remote Sens. Environ., 152, 279–290,
2014.
Medlyn, B. E., Zaehle, S., De Kauwe, M. G., Walker, A. P., Dietze, M. C.,
Hanson, P. J., Hickler, T., Jain, A. K., Luo, Y., and Parton, W.: Using
ecosystem experiments to improve vegetation models, Nat. Clim. Change,
5, 528–534, 2015.Melton, J. R. and Arora, V. K.: Competition between plant functional types in the Canadian Terrestrial Ecosystem Model (CTEM) v. 2.0, Geosci. Model Dev., 9, 323–361, 10.5194/gmd-9-323-2016, 2016.Miralles, D. G.: GLEAM v3, available at:
https://www.gleam.eu/, last access: 27 March 2020.Miralles, D. G., Holmes, T. R. H., De Jeu, R. A. M., Gash, J. H., Meesters, A. G. C. A., and Dolman, A. J.: Global land-surface evaporation estimated from satellite-based observations, Hydrol. Earth Syst. Sci., 15, 453–469, 10.5194/hess-15-453-2011, 2011.
Miralles, D. G., van den Berg, M., Gash, J., Parinussa, R., Jeu, R., Beck, H.,
Holmes, T., Jimenez, C., Verhoest, N., and Dorigo, W.: El Nino-La Nina cycle
and recent trends in continental evaporation, Nat. Clim. Change, 4,
122–126, 2014.Miralles, D. G., Jiménez, C., Jung, M., Michel, D., Ershadi, A., McCabe, M. F., Hirschi, M., Martens, B., Dolman, A. J., Fisher, J. B., Mu, Q., Seneviratne, S. I., Wood, E. F., and Fernández-Prieto, D.: The WACMOS-ET project – Part 2: Evaluation of global terrestrial evaporation data sets, Hydrol. Earth Syst. Sci., 20, 823–842, 10.5194/hess-20-823-2016, 2016.
Monteith, J. L.: Evaporation and environment, Symposia of the society for experimental biology, Cambridge University Press (CUP), Cambridge, UK, 19, 205–234, 1965.
Monteith, J. L.: Accommodation between transpiring vegetation and the convective boundary layer, J. Hydrol., 166, 251–263, 1995.
Mu, Q., Heinsch, F. A., Zhao, M., and Running, S. W.: Development of a
global evapotranspiration algorithm based on MODIS and global meteorology
data, Remote Sens. Environ., 111, 519–536, 2007.
Mu, Q., Zhao, M., and Running, S. W.: Improvements to a MODIS global
terrestrial evapotranspiration algorithm, Remote Sens. Environ.,
115, 1781–1800, 2011.Mueller, B., Seneviratne, S. I., Jimenez, C., Corti, T., Hirschi, M.,
Balsamo, G., Ciais, P., Dirmeyer, P., Fisher, J., and Guo, Z.: Evaluation of
global observations-based evapotranspiration datasets and IPCC AR4
simulations, Geophys. Res. Lett., 38, L06402, 10.1029/2010GL046230, 2011.Mueller, B., Hirschi, M., Jimenez, C., Ciais, P., Dirmeyer, P. A., Dolman, A. J., Fisher, J. B., Jung, M., Ludwig, F., Maignan, F., Miralles, D. G., McCabe, M. F., Reichstein, M., Sheffield, J., Wang, K., Wood, E. F., Zhang, Y., and Seneviratne, S. I.: Benchmark products for land evapotranspiration: LandFlux-EVAL multi-data set synthesis, Hydrol. Earth Syst. Sci., 17, 3707–3720, 10.5194/hess-17-3707-2013, 2013.
Murray-Tortarolo, G., Anav, A., Friedlingstein, P., Sitch, S., Piao, S.,
Zhu, Z., Poulter, B., Zaehle, S., Ahlström, A., and Lomas, M.:
Evaluation of land surface models in reproducing satellite-derived LAI over
the high-latitude northern hemisphere. Part I: Uncoupled DGVMs, Remote Sens., 5, 4819–4838, 2013.
Nagler, P. L., Scott, R. L., Westenburg, C., Cleverly, J. R., Glenn, E. P.,
and Huete, A. R.: Evapotranspiration on western US rivers estimated using
the Enhanced Vegetation Index from MODIS and data from eddy covariance and
Bowen ratio flux towers, Remote Sens. Environ., 97, 337–351, 2005.
Nepstad, D., Lefebvre, P., Lopes da Silva, U., Tomasella, J., Schlesinger,
P., Solórzano, L., Moutinho, P., Ray, D., and Guerreira Benito, J.:
Amazon drought and its implications for forest flammability and tree growth:
A basin-wide analysis, Glob. Change Biol., 10, 704–717, 2004.Niu, G. Y., Yang, Z. L., Mitchell, K. E., Chen, F., Ek, M. B., Barlage, M.,
Kumar, A., Manning, K., Niyogi, D., and Rosero, E.: The community Noah land
surface model with multiparameterization options (Noah-MP): 1. Model
description and evaluation with local-scale measurements, J. Geophys. Res.-Atmos., 116, D12109, 10.1029/2010JD015139, 2011.
Noilhan, J. and Mahfouf, J.-F.: The ISBA land surface parameterisation
scheme, Global Planet. Change, 13, 145–159, 1996.
Norman, J. M. and Becker, F.: Terminology in thermal infrared remote sensing
of natural surfaces, Remote Sens. Rev., 12, 159–173, 1995.
Oki, T. and Kanae, S.: Global hydrological cycles and world water resources,
Science, 313, 1068–1072, 2006.
Oleson, K. W., Lawrence, D. M., Gordon, B., Flanner, M. G., Kluzek, E.,
Peter, J., Levis, S., Swenson, S. C., Thornton, E., and Feddema, J.:
Technical description of version 4.0 of the Community Land Model (CLM),
NCAR technical note, National Center for Atmospheric Research, Boulder, CO, USA, 257 pp., 2010.Pagán, R. B., Maes, H. W., Gentine, P., Martens, B., and Miralles, G.
D.: Exploring the Potential of Satellite Solar-Induced Fluorescence to
Constrain Global Transpiration Estimates, Remote Sens., 11, 413, 10.3390/rs11040413, 2019.
Pan, N., Feng, X., Fu, B., Wang, S., Ji, F., and Pan, S.: Increasing global
vegetation browning hidden in overall vegetation greening: Insights from
time-varying trends, Remote Sens. Environ., 214, 59–72, 2018.Pan, S., Tian, H., Dangal, S. R., Yang, Q., Yang, J., Lu, C., Tao, B., Ren,
W., and Ouyang, Z.: Responses of global terrestrial evapotranspiration to
climate change and increasing atmospheric CO2 in the 21st century, Earth's
Future, 3, 15–35, 2015.Pan, S., Chen, G., Ren, W., Dangal, S. R. S., Banger, K., Yang, J., Tao, B.,
and Tian, H.: Responses of global terrestrial water use efficiency to
climate change and rising atmospheric CO2 concentration in the twenty-first
century, Int. J. Digit. Earth, 11, 558–582, 2018.
Peng, S., Piao, S., Ciais, P., Myneni, R. B., Chen, A., Chevallier, F.,
Dolman, A. J., Janssens, I. A., Penuelas, J., and Zhang, G.: Asymmetric
effects of daytime and night-time warming on Northern Hemisphere vegetation,
Nature, 501, 88–92, 2013.
Peterson, T. C., Golubev, V. S., and Groisman, P. Y.: Evaporation losing its
strength, Nature, 377, 687–688, 1995.
Piao, S., Ciais, P., Huang, Y., Shen, Z., Peng, S., Li, J., Zhou, L., Liu,
H., Ma, Y., and Ding, Y.: The impacts of climate change on water resources
and agriculture in China, Nature, 467, 43–51, 2010.Pinzon, J. E. and Tucker, C. J.: GIMMS NDVI3gV1, available at:
https://ecocast.arc.nasa.gov/data/pub/gimms/3g.v1/, last access: 27 March 2020.Poon, P. and Kinoshita, A.: Estimating Evapotranspiration in a Post-Fire
Environment Using Remote Sensing and Machine Learning, Remote Sens., 10,
1728, 10.3390/rs10111728, 2018.
Priestley, C. and Taylor, R.: On the assessment of surface heat flux and
evaporation using large-scale parameters, Mon. Weather Rev., 100,
81–92, 1972.
Purdy, A. J., Fisher, J. B., Goulden, M. L., Colliander, A., Halverson, G.,
Tu, K., and Famiglietti, J. S.: SMAP soil moisture improves global
evapotranspiration, Remote Sens. Environ., 219, 1–14, 2018.
Reichstein, M., Besnard, S., Carvalhais, N., Gans, F., Jung, M., Kraft, B.,
and Mahecha, M.: Modelling Landsurface Time-Series with Recurrent Neural
Nets, IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing
Symposium, 22–27 July 2018, Valencia, Spain, 7640–7643, 2018.
Reichstein, M., Camps-Valls, G., Stevens, B., Jung, M., Denzler, J.,
Carvalhais, N., and Prabhat: Deep learning and process understanding for
data-driven Earth system science, Nature, 566, 195–204, 2019.
Restrepo-Coupe, N., Levine, N. M., Christoffersen, B. O., Albert, L. P.,
Wu, J., Costa, M. H., Galbraith, D., Imbuzeiro, H., Martins, G., and da
Araujo, A. C.: Do dynamic global vegetation models capture the seasonality
of carbon fluxes in the Amazon basin? A data-model intercomparison, Glob. Change Biol., 23, 191–208, 2017.
Roderick, M. L. and Farquhar, G. D.: The cause of decreased pan evaporation
over the past 50 years, Science, 298, 1410–1411, 2002.
Rouholahnejad, E. and Martens, B.: Improving global estimates of terrestrial
evaporation by integrating remotely-sensed observations of changes in
terrestrial water storage (GRACE) into a global evaporation model (GLEAM),
AGU Fall Meeting, 10–14 December 2018, Washington, D.C., USA, 2018.
Running, S. W.: A measurable planetary boundary for the biosphere, Science,
337, 1458–1459, 2012.Running, S. W.: MOD16A_MONTHLY.MERRA_GMAO_1kmALB, available at:
http://files.ntsg.umt.edu/data/NTSG_Products/MOD16/, last access: 27 March 2020.
Sayemuzzaman, M. and Jha, M. K.: Seasonal and annual precipitation time
series trend analysis in North Carolina, United States, Atmos.
Res., 137, 183–194, 2014.
Semmens, K. A., Anderson, M. C., Kustas, W. P., Gao, F., Alfieri, J. G.,
McKee, L., Prueger, J. H., Hain, C. R., Cammalleri, C., Yang, Y., Xia, T.,
Sanchez, L., Mar Alsina, M., and Vélez, M.: Monitoring daily
evapotranspiration over two California vineyards using Landsat 8 in a
multi-sensor data fusion approach, Remote Sens. Environ., 185,
155–170, 2016.
Sen, P. K.: Estimates of the regression coefficient based on Kendall's tau,
J. American Stat. Assoc., 63, 1379–1389, 1968.
Shrestha, N. and Shukla, S.: Support vector machine based modeling of
evapotranspiration using hydro-climatic variables in a sub-tropical
environment, Agr. Forest Meteorol., 200, 172–184, 2015.Siebert, S., Burke, J., Faures, J. M., Frenken, K., Hoogeveen, J., Döll, P., and Portmann, F. T.: Groundwater use for irrigation – a global inventory, Hydrol. Earth Syst. Sci., 14, 1863–1880, 10.5194/hess-14-1863-2010, 2010.Simmons, A., Willett, K., Jones, P., Thorne, P., and Dee, D.: Low-frequency
variations in surface atmospheric humidity, temperature, and precipitation:
Inferences from reanalyses and monthly gridded observational data sets,
J. Geophys. Res.-Atmos., 115, D01110, 10.1029/2009JD012442, 2010.
Sitch, S., Smith, B., Prentice, I. C., Arneth, A., Bondeau, A., Cramer, W.,
Kaplan, J., Levis, S., Lucht, W., and Sykes, M. T.: Evaluation of ecosystem
dynamics, plant geography and terrestrial carbon cycling in the LPJ dynamic
global vegetation model, Glob. Change Biol., 9, 161–185, 2003.
Smith, B.: LPJ-GUESS-an ecosystem modelling framework, Department of
Physical Geography and Ecosystems Analysis, INES, Sölvegatan, 12, 22362,
2001.Steffen, W., Richardson, K., Rockström, J., Cornell, S. E., Fetzer, I.,
Bennett, E. M., Biggs, R., Carpenter, S. R., de Vries, W., de Wit, C. A.,
Folke, C., Gerten, D., Heinke, J., Mace, G. M., Persson, L. M., Ramanathan,
V., Reyers, B., and Sörlin, S.: Planetary boundaries: Guiding human
development on a changing planet, Science, 347, 1259855, 10.1126/science.1259855, 2015.Stoy, P. C., El-Madany, T. S., Fisher, J. B., Gentine, P., Gerken, T., Good, S. P., Klosterhalfen, A., Liu, S., Miralles, D. G., Perez-Priego, O., Rigden, A. J., Skaggs, T. H., Wohlfahrt, G., Anderson, R. G., Coenders-Gerrits, A. M. J., Jung, M., Maes, W. H., Mammarella, I., Mauder, M., Migliavacca, M., Nelson, J. A., Poyatos, R., Reichstein, M., Scott, R. L., and Wolf, S.: Reviews and syntheses: Turning the challenges of partitioning ecosystem evaporation and transpiration into opportunities, Biogeosciences, 16, 3747–3775, 10.5194/bg-16-3747-2019, 2019.
Sun, Q., Miao, C., Duan, Q., Ashouri, H., Sorooshian, S., and Hsu, K. L.: A
review of global precipitation data sets: Data sources, estimation, and
intercomparisons, Rev. Geophys., 56, 79–107, 2018.Sun, Y., Frankenberg, C., Wood, J. D., Schimel, D. S., Jung, M., Guanter,
L., Drewry, D. T., Verma, M., Porcar-Castell, A., Griffis, T. J., Gu, L.,
Magney, T. S., Köhler, P., Evans, B., and Yuen, K.: OCO-2 advances
photosynthesis observation from space via solar-induced chlorophyll
fluorescence, Science, 358, eaam5747, 10.1126/science.aam5747, 2017.Swann, A. L., Hoffman, F. M., Koven, C. D., and Randerson, J. T.: Plant
responses to increasing CO2 reduce estimates of climate impacts on drought
severity, P. Natl. Acad. Sci. USA, 113, 10019–10024,
2016.
Szutu, D. J. and Papuga, S. A.: Year-Round Transpiration Dynamics Linked
With Deep Soil Moisture in a Warm Desert Shrubland, Water Resour. Res., 55, 5679–5695, 2019.
Tabari, H., Martinez, C., Ezani, A., and Talaee, P. H.: Applicability of
support vector machines and adaptive neurofuzzy inference system for
modeling potato crop evapotranspiration, Irrigation Sci., 31, 575–588,
2013.Talsma, C. J., Good, S., Miralles, D., Fisher, J., Martens, B., Jimenez, C.,
and Purdy, A.: Sensitivity of Evapotranspiration Components in Remote
Sensing-Based Models, Remote Sens., 10, 1601, 10.3390/rs10101601, 2018a.
Talsma, C. J., Good, S. P., Jimenez, C., Martens, B., Fisher, J. B.,
Miralles, D. G., McCabe, M. F., and Purdy, A. J.: Partitioning of
evapotranspiration in remote sensing-based models,
Agr. Forest Meteorol., 260–261, 131–143, 2018b.
Tian, H., Chen, G., Liu, M., Zhang, C., Sun, G., Lu, C., Xu, X., Ren, W.,
Pan, S., and Chappelka, A.: Model estimates of net primary productivity,
evapotranspiration, and water use efficiency in the terrestrial ecosystems
of the southern United States during 1895–2007,
Forest Ecol. Manage., 259, 1311–1327, 2010.
Trenberth, K. E., Fasullo, J. T., and Kiehl, J.: Earth's global energy
budget, B. Am. Meteorol. Soc., 90, 311–324, 2009.
Van Dijk, A. and Bruijnzeel, L. A.: Modelling rainfall interception by vegetation of variable density using an adapted analytical model. Part 1. Model description, J. Hydrol., 247, 230–238, 2001.
Vinukollu, R. K., Meynadier, R., Sheffield, J., and Wood, E. F.:
Multi-model, multi-sensor estimates of global evapotranspiration:
Climatology, uncertainties and trends, Hydrol. Process., 25,
3993–4010, 2011a.
Vinukollu, R. K., Wood, E. F., Ferguson, C. R., and Fisher, J. B.: Global
estimates of evapotranspiration for climate studies using multi-sensor
remote sensing data: Evaluation of three process-based approaches,
Remote Sens. Environ., 115, 801–823, 2011b.Wang, K. and Dickinson, R. E.: A review of global terrestrial
evapotranspiration: Observation, modeling, climatology, and climatic
variability, Rev. Geophys., 50, RG2005, 10.1029/2011RG000373, 2012.Wang, K., Dickinson, R. E., Wild, M., and Liang, S.: Evidence for decadal
variation in global terrestrial evapotranspiration between 1982 and 2002: 1.
Model development, J. Geophys. Res.-Atmos., 115,
D20112, 10.1029/2009JD013671, 2010.Wartenburger, R., Seneviratne, S. I., Hirschi, M., Chang, J., Ciais, P.,
Deryng, D., Elliott, J., Folberth, C., Gosling, S. N., and Gudmundsson, L.:
Evapotranspiration simulations in ISIMIP2a-Evaluation of spatio-temporal
characteristics with a comprehensive ensemble of independent datasets,
Environ. Res. Lett., 13, 075001, 10.1088/1748-9326/aac4bb, 2018.
Willett, K. M., Gillett, N. P., Jones, P. D., and Thorne, P. W.: Attribution
of observed surface humidity changes to human influence, Nature, 449, 710–712,
2007.
Xu, D., Agee, E., Wang, J., and Ivanov, V. Y.: Estimation of
Evapotranspiration of Amazon Rainforest Using the Maximum Entropy Production
Method, Geophys. Res. Lett., 46, 1402–1412, 2019.
Xu, T., Guo, Z., Liu, S., He, X., Meng, Y., Xu, Z., Xia, Y., Xiao, J.,
Zhang, Y., Ma, Y., and Song, L.: Evaluating Different Machine Learning
Methods for Upscaling Evapotranspiration from Flux Towers to the Regional
Scale, J. Geophys. Res.-Atmos., 123, 8674–8690, 2018.
Yao, Y., Liang, S., Cheng, J., Liu, S., Fisher, J. B., Zhang, X., Jia, K.,
Zhao, X., Qin, Q., and Zhao, B.: MODIS-driven estimation of terrestrial
latent heat flux in China based on a modified Priestley–Taylor algorithm,
Agr. Forest Meteorol., 171, 187–202, 2013.
Yao, Y., Liang, S., Li, X., Liu, S., Chen, J., Zhang, X., Jia, K., Jiang,
B., Xie, X., and Munier, S.: Assessment and simulation of global terrestrial
latent heat flux by synthesis of CMIP5 climate models and surface eddy
covariance observations, Agr. Forest Meteorol., 223, 151–167,
2016.
Yao, Y., Liang, S., Li, X., Chen, J., Liu, S., Jia, K., Zhang, X., Xiao, Z.,
Fisher, J. B., and Mu, Q.: Improving global terrestrial evapotranspiration
estimation using support vector machine by integrating three process-based
algorithms, Agr. Forest Meteorol., 242, 55–74, 2017.Yuan, W., Liu, S., Liu, H., Randerson, J. T., Yu, G., and Tieszen, L. L.:
Impacts of precipitation seasonality and ecosystem types on
evapotranspiration in the Yukon River Basin, Alaska, Water Resour. Res., 46, W02514, 10.1029/2009WR008119, 2010.
Yue, S., Pilon, P., and Cavadias, G.: Power of the Mann–Kendall and
Spearman's rho tests for detecting monotonic trends in hydrological series,
J. Hydrol., 259, 254–271, 2002.Zaehle, S. and Friend, A.: Carbon and nitrogen cycle dynamics in the O-CN
land surface model: 1. Model description, site-scale evaluation, and
sensitivity to parameter estimates, Global Biogeochem. Cy., 24, GB1005, 10.1029/2009GB003521, 2010.Zeng, Z., Piao, S., Lin, X., Yin, G., Peng, S., Ciais, P., and Myneni, R.
B.: Global evapotranspiration over the past three decades: estimation based
on the water balance equation combined with empirical models, Environ. Res. Lett., 7, 014026, 10.1088/1748-9326/7/1/014026, 2012.
Zeng, Z., Wang, T., Zhou, F., Ciais, P., Mao, J., Shi, X., and Piao, S.: A
worldwide analysis of spatiotemporal changes in water balance-based
evapotranspiration from 1982 to 2009, J. Geophys. Res.-Atmos., 119, 1186–1202, 2014.Zeng, Z., Zhu, Z., Lian, X., Li, L. Z., Chen, A., He, X., and Piao, S.:
Responses of land evapotranspiration to Earth's greening in CMIP5 Earth
System Models, Environ. Res. Lett., 11, 104006, 10.1088/1748-9326/11/10/104006, 2016.
Zeng, Z., Peng, L., and Piao, S.: Response of terrestrial evapotranspiration
to Earth's greening, Curr. Opin. Env. Sust., 33,
9–25, 2018a.Zeng, Z., Piao, S., Li, L. Z., Wang, T., Ciais, P., Lian, X., Yang, Y., Mao,
J., Shi, X., and Myneni, R. B.: Impact of Earth greening on the terrestrial
water cycle, J. Climate, 31, 2633–2650, 2018b.
Zhang, K.: P-LSH ET, available at:
http://files.ntsg.umt.edu/data/ET_global_monthly/Global_8kmResolution/, last access: 27 March 2020.Zhang, K., Kimball, J. S., Nemani, R. R., and Running, S. W.: A continuous
satellite-derived global record of land surface evapotranspiration from
1983 to 2006, Water Resour. Res., 46, W09522, 10.1029/2009WR008800, 2010.Zhang, K., Kimball, J. S., Nemani, R. R., Running, S. W., Hong, Y., Gourley,
J. J., and Yu, Z.: Vegetation greening and climate change promote
multidecadal rises of global land evapotranspiration, Sci. Rep., 5,
15956, 10.1038/srep15956, 2015.
Zhang, K., Kimball, J. S., and Running, S. W.: A review of remote sensing
based actual evapotranspiration estimation, WIRES:
Water, 3, 834–853, 2016.
Zhang, L., Lei, H., Shen, H., Cong, Z., Yang, D., and Liu, T.: Evaluating
the Representation of Vegetation Phenology in the Community Land Model 4.5
in a Temperate Grassland, J. Geophys. Res.-Biogeo.,
124, 187–210, 2019.Zhang, Y.: Monthly global observation-driven Penman-Monteith-Leuning (PML) evapotranspiration and components, available at:
https://data.csiro.au/dap/landingpage?pid=csiro:17375, last access: 27 March 2020.Zhang, Y., Peña-Arancibia, J. L., McVicar, T. R., Chiew, F. H., Vaze,
J., Liu, C., Lu, X., Zheng, H., Wang, Y., and Liu, Y. Y.: Multi-decadal
trends in global terrestrial evapotranspiration and its components,
Sci. Rep., 6, 19124, 10.1038/srep19124, 2016.
Zhu, Z., Bi, J., Pan, Y., Ganguly, S., Anav, A., Xu, L., Samanta, A., Piao,
S., Nemani, R. R., and Myneni, R. B.: Global data sets of vegetation leaf
area index (LAI) 3g and Fraction of Photosynthetically Active Radiation
(FPAR) 3g derived from Global Inventory Modeling and Mapping Studies (GIMMS)
Normalized Difference Vegetation Index (NDVI3g) for the period 1981 to 2011, Remote Sens., 5, 927–948, 2013.
Zhu, Z., Piao, S., Myneni, R. B., Huang, M., Zeng, Z., Canadell, J. G.,
Ciais, P., Sitch, S., Friedlingstein, P., and Arneth, A.: Greening of the
Earth and its drivers, Nat. Clim. Change, 2016, 791–795, 2016.