ESDDEarth System Dynamics DiscussionsESDDEarth Syst. Dynam. Discuss.2190-4995Copernicus GmbHGöttingen, Germany10.5194/esdd-6-217-2015Appraising the capability of a land biosphere model as a tool in modelling land surface interactions: results from its validation at selected European ecosystemsNorthM. R.PetropoulosG. P.george.petropoulos@aber.ac.ukIrelandG.McCalmontJ. P.Department of Geography & Earth Sciences, Aberystwyth University, Aberystwyth, SY23 3DB, UKInstitute of Biological, Rural and Environmental Sciences, Aberystwyth University, Aberystwyth, SY23 3EB, UKG. P. Petropoulos (george.petropoulos@aber.ac.uk)9February2015612172657January201512January2015This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://esd.copernicus.org/preprints/6/217/2015/esdd-6-217-2015.htmlThe full text article is available as a PDF file from https://esd.copernicus.org/preprints/6/217/2015/esdd-6-217-2015.pdf
In this present study the ability of the SimSphere Soil Vegetation
Atmosphere Transfer (SVAT) model in estimating key parameters
characterising land surface interactions was evaluated.
Specifically, SimSphere's performance in predicting Net Radiation
(Rnet), Latent Heat (LE), Sensible Heat (H) and Air
Temperature (Tair) at 1.3 and 50 m was
examined. Model simulations were validated by ground-based
measurements of the corresponding parameters for a total of 70 days
of the year 2011 from 7 CarboEurope network sites. These included a variety
of biomes, environmental and climatic conditions in the models evaluation.
Overall, model performance can largely be described as
satisfactory for most of the experimental sites and evaluated parameters.
For all model parameters compared, predicted H
fluxes consistently obtained the highest agreement to the in-situ data in
all ecosystems, with an average RMSD of 55.36 W m-2. LE
fluxes and Rnet also agreed well with the in-situ data
with RSMDs of 62.75 and 64.65 W m-2 respectively. A good
agreement between modelled and measured LE and H fluxes was found,
especially for smoothed daily flux trends. For both
Tair 1.3 m and Tair 50 m
a mean RMSD of 4.14 and 3.54 ∘C was reported respectively.
This work presents the first all-inclusive evaluation of SimSphere,
particularly so in a European setting. Results of this study
contribute decisively towards obtaining a better understanding of
the model's structure and its correspondence to the real world
system. Findings also further establish the model's capability as
a useful teaching and research tool in modelling Earth's land
surface interactions. This is of considerable importance in the
light of the rapidly expanding use of the model worldwide,
including ongoing research by various Space Agencies examining its
synergistic use with Earth Observation data towards the development
of operational products at a global scale.
Introduction
Global climate change is currently facilitating large scale changes
within the atmosphere, biosphere, geosphere and hydrosphere
(Steinhauser et al., 2012). Quantification and management of such
changes and a better understanding of the interactions between
different components of the Earth system has been identified nowadays
as an important and urgent research direction to be addressed within
numerous scientific disciplines (Coudert et al., 2008; Petropoulos
et al., 2013a). It also serves as essential information for policy
makers and the wider global community (IPCC, 2009). Accurate
monitoring of water and vegetation stress is now of prominent global
concern and it is regarded as a high priority issue within several
European Union (EU) frameworks. This is particularly so for
communities in water limited environments, or areas which rely on rain
fed agriculture, such as some regions in the Mediterranean basin
(European Commission, 2009; Amri et al., 2014).
Accurate estimation of energy fluxes and their partitioning has never
been more important in the face of increasing climate change (WMO,
2002; ESA, 2014). The terrestrial boundary layer and its vegetation
play a critical role in regulating the partitioning of incoming energy
(into latent (LE), sensible (H), and ground (G) heat fluxes) and in
the land–atmosphere exchange of carbon dioxide (CO2), and the
close relationship between photosynthesis and the energy and water
vapour cycles (Prentice et al., 2014). On this basis, the need to
develop a thorough understanding of how heat and water fluxes are
characterised in different ecosystems is imperative. This is due to
the profound contribution these parameters make to various
biogeophysical processes at the planetary boundary layer (Feddema
et al., 2005). Currently, the physical interactions behind land
surface processes are relatively well-documented within the global
scientific community. However, there is a need for further research
towards improving our understanding of temporal and spatial dynamics
of energy and water fluxes (Quintana-Segui et al., 2008) and the
complexity of regional energy and water exchanges (Braud et al.,
1995). Also, there is a requirement to provide, at increased
estimation accuracy, parameters characterising the energy and water
cycles at different observations scales (Anderson et al., 2008; Amri
et al., 2014).
Research undertaken to improve our understanding on the representation of land
atmosphere interactions has lead to the development and exploration of
a wide variety of modelling schemes. Since the 1970's the
global scientific community has developed numerous land surface models
(LSMs) to assess a multitude of parameters associated with land
surface interactions with varying degrees of complexity and
applicability (Olchev et al., 2008). LSMs have evolved from simple
bucket models without vegetation consideration (e.g. Manabe, 1969)
into credible representations of the exchanges of energy, water and
carbon dioxide in the soil-vegetation-atmosphere continuum. The use of
Soil Vegetation Atmosphere Transfer (SVAT) models represent one of the
most common approaches in studying land surface processes and the
interactions between the Earth's system components. SVAT models are
mathematical representations of vertical “views” of the physical
mechanisms controlling energy and mass transfers in the
soil-vegetation-atmosphere continuum. Those models are able to provide
deterministic estimates of the time course of soil and vegetation
state variables at time-steps compatible with the dynamics of
atmospheric processes.
Those models have arisen as a convergence of several needs (Petropoulos
et al., 2009a), namely: (i) to better understand land/atmosphere
boundary transfers, (ii) to investigate how vegetation responds to
climate change and (iii) to assess hydrological balances and measure
conditions at a given boundary level. One of their main relative
advantages, compared to traditional techniques, is the ability to
simulate at a fine temporal resolution (often less than 1 h); this
subsequently allows simulations to be in satisfactory agreement with
the timescale of the physical process being simulated. In addition to
this, SVATs comprehensively analyse a large array of parameters
associated with the hydrological, radiative and physical domains of
the Earth's energy and water cycles. To this end, such models are
widely regarded as the most suitable tool to analyse various complex
land surface interactions. SVAT models can be employed as
“decision making tools” within policy implementation because of
their ability to holistically and accurately assess numerous
parameters in past, present, and future environments. Yet, their
predictions have an undefined spatial coverage and are limited in
their ability to simulate energy and water transfers only within an
area representative of their initial parameterisation. Therefore,
surface heterogeneity presents itself as a pertinent problem in the
application of those models to more fragmented landscapes, where the
high levels of internal biophysical variability cannot be fully
represented within the model's parameterisation (Oltchev et al., 2002;
Falge et al., 2005; Olioso et al., 2005; Samaali et al.,
2007). Additionally, SVAT models often require a large amount of input
parameters for initialisation. This makes the widespread application
and transferability of those models in some cases troublesome. This is
because obtaining site specific parameters in remote and data scarce
areas is often very difficult (Oltchev et al., 2002). Current
research has led to the development of SVATs incorporating sub-grid
scale heterogeneity and with improved representation of plant
physiological processes. Evidently, the incorporation of these
additional processes has further increased the complexity and number
of input parameters required to implement such models.
It is important to note, however, that uncertainty is inevitable in
any model since it will never be as complex as the reality it portrays
(Denti, 2004). As such, the process of validating a mathematical model
is an essential step in its development. Generally speaking, the
validation of a model consists of determining how well the model
performs when comparing its simulated results with those from the real
world. Numerous model validation techniques exist; for a comprehensive
overview see Bellocchi et al. (2010). A common
strategy is to quantitatively compare the model's predictions
vs. actual in-situ observations on the basis of various appropriate
statistical metrics. Validation techniques are often also implemented
over numerous land cover types, helping to further identify how energy
and water fluxes are characterised within different ecological
settings. Such techniques help develop confidence in the model's
ability to be used within these settings and also contribute to our
overall understanding on how land cover types characterise local
energy and water fluxes (Coudert et al., 2008). Sensitivity analysis
(SA) can also be performed as a key component of any model evaluation,
including SVATs. SA utilises mathematical techniques which aim to
quantify the relative influence of each input parameter on the model's
output variability (Tomlin, 2013; Vanuytrecht et al., 2014). It allows
for an objective assessment of model structure and coherence (Petropoulos et al., 2013a; Gan
et al., 2014). In addition, Kramer et al. (2002), in an attempt to
holistically assess the capability of a model in portraying a real
world system, has proposed a set of model assessment criteria, namely:
accuracy, generality and realism. Accuracy is described
as the “goodness of fit” of a models estimations to
in-situ measurements. Generality is described as the applicability of
the model in numerous ecosystems. Realism is described as the ability
of the model to address relationships between modelled phenomena. It
is widely agreed however, that sometimes discrepancies between the
modelled and observed datasets can be partly attributed to uncertainty
within the observational dataset itself (Denti, 2004; Wang et al.,
2004; Verbeeck et al., 2009). Therefore validation attempts not only
require a highly accurate observational dataset (Wang et al., 2004),
but also a wider understanding of problems associated to equifinality,
insensitivity and uncertainty when assessing biophysical models
(Verbeeck et al., 2009).
SimSphere is one example of a SVAT model, developed by Carlson and
Boland (1978) to increase our understanding of boundary layer
processes. Since its original development, the model has diversified
and become highly varied in its applicational use (for a comprehensive
overview of the model use refer to Petropoulos
et al., 2009a). SimSphere's development as a research, educational and
training tool is currently expanding within several universities
worldwide. Furthermore, its use synergistically with Earth
Observation (EO) data is at present being considered by several Space
Agencies towards the development of spatio-temporal estimates of
evapotranspiration (ET) rates and surface soil moisture (Mo) products
at an operational scale globally (Chauhan et al., 2003; ESA STSE,
2012). These investigations have been based around the implementation
of a data assimilation technique termed the “triangle” on which
SimSphere is used synergistically with EO data
(Carlson, 2007; Petropoulos and Carlson, 2011). Furthermore,
a variant of this “triangle” approach is already in use in Spain to
deliver an operational product of surface soil moisture at
1 km spatial resolution from the Soil Moisture and Ocean Salinity (SMOS) satellite launched by
the European Space Agency (ESA) (Piles et al., 2011).
Thus, it is understandable that it is of primary importance to perform
a variety of validatory tests to appraise SimSphere's adequacy and
coherence in terms of its ability to realistically represent Earth
surface processes. In this respect, a series of SA experiments have
already been conducted on SimSphere (Petropoulos et al., 2009b, 2013a,
b, 2014a, b). Those studies provided for the first time independent
evidence to enhance our understanding of the model's behaviour,
coherence and correspondence to what it has been built to
simulate. Yet, validation studies performing direct comparisons of
model predictions against corresponding in-situ data on the basis of
statistical metrics proposed in the classic literature have been
scarce and incomprehensive, only performed over a very small range of
land use/cover types (e.g. Todhunter and Terjung, 1987; Ross and Oke,
1988). This despite the fact that this type of validation approach is
a common strategy in examining the accuracy of model predictions (e.g.
Falge et al., 2005; Giertz et al., 2006; Marshall et al., 2013).
Given SimSphere's current global expansion, this type of validation
is both timely and of fundamental importance in further
establishing the model's structure, coherence and representativeness.
With regards to the elements discussed above, this paper investigates
the applicability of SimSphere in reproducing a series of observed
parameter validations characterising land surface interactions at
a total of 7 European ecosystems. The objective was to thoroughly
understand the model's ability to simulate, at a local scale, key
parameters characterising Earth's energy and water budgets, namely:
net Radiation (Rnet), Latent Heat (LE), Sensible Heat
(H), and Air temperature (Tair) at 1.3 and
50 m. Model validation is performed through a comparison of
the model predictions against corresponding data belonging to
CarboEurope, the largest in-situ monitoring network in Europe
which provides validated measurements of key micrometeorological
parameters.
Model formulation
This work deals with the SimSphere 1-D boundary layer model devoted to
the study of energy and mass interactions of the Earth
system. Formerly known as the Penn-State University
Biosphere–Atmosphere Modeling Scheme (PSUBAMS) (Carlson and Boland,
1978; Lynn and Carlson, 1990), it was considerably modified to its
current state by Gillies et al. (1997) and later by Petropoulos
et al. (2013c). It is currently maintained and freely distributed by
the Department of Geography and Earth Sciences at Aberystwyth
University (http://www.aber.ac.uk/simsphere). This section aims
at providing an overview of the model architecture, based on the most
recent implementation by Gillies et al. (1997).
SimSphere represents various physical processes taking place in
a column that extends from the root zone below the soil surface up to
a level well above the surface canopy, the top of the surface mixing
layer. Essentially, SimSphere is a 1-dimensional two-source SVAT model
with a plant component. Three main systems are represented within
SimSphere's structure, namely the physical, the
vertical and the horizontal layers (Fig. 1).
The physical components ultimately determine the
microclimate conditions in the model and are grouped into three
categories, radiative, atmospheric and hydrological. The
primary forcing of this component is the available clear sky radiant
energy reaching the surface or the plant canopy, calculated as
a function of sun and earth geometry, atmospheric transmission factors
for scattering and absorption, the atmospheric and surface
emissivities and surface (including soil and plant) albedos. The
vertical structure components, effectively
correspond to the components of the Planetary Boundary Layer (PBL)
which is divided into three layers – a surface mixing layer,
a surface of constant flux layer and a surface vegetation or bare soil
layer, where the depths of the first layer is somewhat variable with
time, growing throughout the day as H flux is added from below.
The depth of the constant flux and vegetation layers are set in the
model input, although the depth of a bare soil transition (between
soil and air) layer is variable in time depending on the wind speed
and the surface roughness. In addition, the vertical structure also
contains a fourth layer, the substrate layer, which refers to the
depth of the soil over which heat and water is conducted.
The vegetation component is dormant at night, that is, after radiation
sunset. The night-time dynamics for the surface fluxes differ from
those during the day time. LE and H fluxes are exchanged
between both the ground and foliage, between plant and inter-plant
airspaces through stomatal and cuticular resistances in the leaf (for
water vapour) and the air, between soil and the interplant air spaces
and between the entire vegetation canopy and the air. A separate
component exists for the bare soil fluxes between the surface and the
air. Vegetation and soil fluxes merged at the top of the vegetation
canopy. Their relative weights depend on the fractional vegetation
cover, specified as an input to the model. As such, SimSphere is
referred to as a form of two-stream or two-source model. An important
factor in controlling, in particular, the partitioning between
LE and H is the stomatal resistance component within
the vegetation parameterisation settings. SimSphere provides a choice
of two stomatal resistance parameterisations, Deardoff (1978) and
Carlson and Lynn (1991). The first is inclusive of the stomatal
resistance behaviour that is affected by soil, water and sunlight.
However, the inability to measure plant hydraulics (a major
attributing factor to vegetation transpiration) is seen to be
a prominent disadvantage. The second measures stomatal resistance as
a function of leaf–atmosphere vapour pressure difference. This is
measured by the difference within the mesophyll and epidermal leaf
water potentials, as the stomatal resistance is directly proportional
to the vapour pressure difference. The main advantage of this
parameterisation setting is the ability to analyse the transpiration
plateau effect, described in more detail in Carlson et al. (1991). Further details about the
model architecture can be found in Gillies (1993).
The processes and interactions simulated by the model are allowed to
develop over a 24 h cycle at a chosen time step, starting from a set
of initial conditions given in the early morning (at 05:30 LT – local
time) with a continuous evolving interaction between soil, plant and
atmosphere layers. A large amount of input parameters are required for
the model parameterisation, 53 in total, categorised into 7 defined
groups; time and location, vegetation, surface, hydrological,
meteorological, soil and atmospheric (Table 1). From initialisation,
over a 24 h cycle SimSphere assesses the diurnal evolution of more
than 30 prognostic variables associated with the radiative,
hydrological and atmospheric physical domains. Numerous physical
processes are simulated and all parameters are evaluated as a function
of time and their diurnal evolution. Outputs of the model include,
between others, the surface energy fluxes (LE and H fluxes) below and
at the soil surface, around and above the vegetation canopy and the
transfer of water in the soil and in the plants. It also simulates the
CO2 (carbon dioxide) flux between the atmosphere and the plants and the surface
O3 (ozone) flux. Several meteorological parameters are also assessed
such as the radiometric surface temperature, wind velocity, air
temperature, and humidity at various levels in and above the canopy,
plus a number of other plant parameters, such as stomatal resistance
and leaf water potential.
Materials and methods
This section provides a synopsis of the methodology followed in
evaluating SimSphere's ability to simulate key parameters
characterising land surface interactions. An overview of the main
steps included in this process is furnished in Fig. 2.
In-situ datasets collection
Reliable data is needed to calibrate and evaluate the predictions of
any model (Wang et al., 2004). Therefore, in this study, in-situ data
from selected sites belonging to the CarboEurope ground monitoring
network were obtained. The latter is part of a larger observational
network, FLUXNET (Baldocchi et al., 2001), which is currently the
largest global network acquiring ancillary information of
micro-meteorological flux and a number of ancillary parameters. Once
the data reaches FLUXNET, it is quality controlled and gap-filled
using techniques described by Papale et al. (2006) and Moffat
et al. (2007). As a result, the in-situ data can be provided to the
end users community at different processing levels.
In this study, SimSphere's ability to provide estimates of key
parameters characterising our water and energy balance was evaluated
at 7 CarboEurope sites. These sites were representative of different
ecosystem types with markedly different site characteristics
(Table 2). All available in-situ data for each site was obtained for
the year 2011, allowing for a sufficient database for model
parameterisation and validation to be developed. All data was
acquired from the European Fluxes database Cluster
(http://gaia.agraria.unitus.it/). In particular Level 2 data
was obtained across all selected sites for consistency. This product
includes the originally acquired in-situ measurements from which only
the removal of erroneous data caused by obvious instrumentation error
has been undertaken. In addition, atmospheric profile (i.e.
radiosonde) data were obtained for each site/day by the University of
Wyoming (http://weather.uwyo.edu/upperair/sounding.html). This
data included the atmospheric profile of temperature, dew
point temperature, wind direction, wind speed and atmospheric
pressure.
Validation days selection
Further analysis was implemented to identify the specific days for
which SimSphere would be parameterised and validated for each
experimental site. Initially, for each site, cloudy days were
identified and subsequently excluded from further analysis. Judgment
on which days (or time-periods) were cloud-free was based on analysis
of the diurnal observation of shortwave incoming solar radiation (Rg). Cloud-free days were flagged as
those having smooth and symmetrical Rg curves, a property signifying
clear-sky conditions (Carlson et al., 1991).
Subsequently, for the cloud-free days, the energy balance closure (EBC)
was evaluated. EBC evaluation has been accepted as a valid method for
accuracy assessment of the turbulent fluxes derived from eddy
covariance measurements (Wilson et al., 2002; Li et al., 2005).
Evaluation of EBC using the above equation is only directly relevant
to the assessment of LE and H fluxes, and not to other scalar fluxes
such as CO2 (e.g. Wilson et al., 2002; Foken et al.,
2006). Energy imbalance derived from implementation of the EBC
principle has been found to have implications for the way these energy
flux measurements should be interpreted, and therefore, on how they
should be compared with model simulations (e.g. Twine et al., 2000;
Culf et al., 2002).
EBC was evaluated herein principally by performing a regression
analysis (e.g. see Wilson and Baldocchi, 2000; Wilson et al., 2002;
Oliphant et al., 2004). The linear regression coefficients (slope and
intercept) as well as the coefficient of determination (R2) were
calculated from the ordinary least squares (OLS) relationship between
the half-hourly estimates of the dependent flux variables
(LE +H) and the independently derived available energy
(Rnet-G-S). In addition to this, the Energy Balance Ratio
(EBR) was computed by cumulatively summing Rnet-G-S and
LE +H from the 30 min mean average surface energy flux
components, and then rationing each of the cumulative sums as follows
(e.g. Oliphant et al., 2004; Liu et al., 2006):
EBR=∑(LE+H)∑(Rnet-G-S).
This index ranges generally from zero to one, with values closer to
one highlighting a satisfactory diurnal energy closure, indicating
a good quality of in-situ measurements. All days with poor EBC
(EBR <0.750, slope <0.85, R2<0.930) were excluded from
further analysis.
Further constraints were subsequently employed to ensure that selected
days were of the highest possible quality in terms of in-situ data
quality. Firstly, all days selected were within the growing season of
April–October; this eliminated the main effects ascribed to the
inter-annual variability in vegetation phenology. Secondly, selected
simulation days were assessed for atmospherically stable conditions,
namely low wind speeds and small available energy (Maayar et al.,
2001). Such conditions were identified by the evaluation of the
in-situ dataset, where direct measurements of wind speed and energy
flux amplitude and diurnal trend were used as indicators of
atmospherically stable conditions. However, it should be noted that
for the IT_Ro3 site no in-situ measurements of air temperature were
available for August and September. As a result it was not possible to
evaluate the model's performance for this period. In the end, a final set
of a total of 70 non-consecutive days from the 7 different CarboEurope
sites were identified as being suitable to proceed with the SimSphere
validation.
SimSphere parameterisation and implementation
As already stated (Sect. 2), SimSphere has been developed to simulate
the various physical processes that take place as a function of time
in a column that extends from the root zone below the soil surface up
to a level higher than the surface vegetation canopy. In the
horizontal domain, SimSphere implicitly refers to a horizontal area of
undefined size that can be composed of a mixture of bare soil and
vegetation. Thus, it is conceivable that the horizontal scale for the
model is defined by the degree to which the model's initial conditions
are representative of the horizontal area to be simulated. In theory,
this scale should also be used for the validation process. Consequently,
SimSphere parameterisation was carried out at the measurement scale of
the flux tower observations. That is a function of the area of the
fetch around which the tower is built and the footprint of the
turbulent flux measurements, representing an area of ∼1km2 for the test sites as they are relatively
homogeneous.
On this basis SimSphere was parameterised to the daily conditions
existing at the flux tower for each of the selected days. Initial
conditions for Tair, humidity, wind velocity and direction
soundings were acquired at 06:00 GMT from the University of Wyoming
database to correspond to the model's initialisation and were used
within the parameterisation. Ancillary information on vegetation and
soil parameters (e.g. Leaf Area Index (LAI), Fractional Vegetation
Cover (FVC), vegetation height, soil type etc.) was
also used directly within the model's initialisation. Such information
was acquired in most cases directly from communication with the
principal investigators of each respective site, and occasionally from
standard literature sources (e.g. Mascart et al., 1991; Carlson
et al., 1991). The soil type parameters were obtained from the
classifications of Clapp and Hornberger (1978) and Cosby
et al. (1984), using the soil texture data provided at each
CarboEurope test site and information supplied in some instances by
the site managers for each experimental site. Similarly, this was also
the case for the topographical information that was required in model
initialisation. Upon the model initialisation, the latter was executed
for each site/day and the 30 min average value of each of the evaluated
parameters per site for the period 05:30–23:30 LT was subsequently
exported in SPSS for comparisons against the corresponding in-situ
data.
Validation approach
Six statistical metrics were used to evaluate how well
the SimSphere predictions matched the observed data for each day on
which the model was parameterised and executed. The model's coherence
to the observational data was undertaken using the statistical terms
suggested by Wilmott (1982). These specifically included the
Root Mean Square Difference (RMSD), the linear regression fit model
coefficient of determination (R2), the Bias or Mean Bias Error
(MBE), the Scatter or Mean Square Difference (MSD), the Mean Absolute Error (MAE) and
the NASH index. The MBE term expresses the accuracy of the model
outputs in relation to the in-situ measurements (i.e. low
bias = high accuracy) and is used to correct for systematic
errors. The MSD term expresses model precision (i.e. low
scatter = high precision) and is used to correct for
non-systematic errors. The sum of both can be utilised to evaluate
overall model accuracy. Table 3 lists the formulae that express the
above statistical terms; a detailed description of which can be found
for example in Silk (1979), Burt and Barber (1996) and Wilmott
(1982). These statistics have also been widely used in similar
validation experiments carried out previously (e.g. Wang et al., 2004;
Falge et al., 2005; Giertz et al., 2006; Marshall et al., 2013).
In addition, SimSphere's ability to reproduce the diurnal evolution of
the examined parameters was evaluated according to the Kramer
et al. (2002) criteria described earlier (Sect. 1). All statistical
metrics were computed from comparisons performed at identical
0.5 hourly intervals between the two datasets for each day of
comparison. In addition, the same statistical parameters where
computed as a summary per experimental CarboEurope site to provide an
overview of the model performance per site.
ResultsNet Radiation (Rnet) flux
Table 3 summarises the results of the statistical analysis concerning
the comparisons of Net Radiation between the SimSphere estimations and
the in-situ measurements. Furthermore, Fig. 3 illustrates the
agreement between the in-situ and the predicted Rnet for
all days of comparisons from all experimental sites. Generally, the
diurnal variation of the simulated Rnet was in close
correspondence with the observed Rnet in both shape and
magnitude for most of the compared days (although results not shown
here for brevity). In overall, SimSphere was able to simulate
Rnet relatively satisfactorily with an average RMSD of
64.65 Wm-2 and a correlation coefficient of 0.95.
A minor underestimation of the in-situ data was also evident for all
sites and days combined (MBE =-2.07 Wm-2). The
correspondence between predicted and observed Rnet fluxes
was variable between the individual sites and days included in our
study. Indeed, Rnet showed a significant range of
agreement, with RMSD ranging from 24.38 to 98.26 Wm-2
between the different validation days. Notably, there were increased
periods within a number of test sites where simulation accuracy
increased depending on the period in which the simulation days were
located. For example, for the IT_Ro3 cropland site, error ranges
decreased for the period between late April (21 April 2011) and late
August (28 August 2011), before increasing in early September
(9 September 2011). However, the periods of increased accuracy varied
on a per site basis and were only prevalent within the olive
plantation (ES_Lju), grassland (IT_Mbo), cropland (IT_Ro3) and
deciduous broadleaf forest (IT_Col) sites. Daily R2 values
exhibited less variance with generally more comparable ranges
(0.909–0.998) between all the study days, suggesting a satisfactory
agreement between both datasets, also illustrated by the distribution
of the points around the 1:1 line in Fig. 3. This was also reflected
within the NASH index values reported (0.897–0.999).
As can be seen from Table 3, when averaged per site, RMSD showed
significantly less variance, exhibiting a range from
55.86 Wm2 (IT_Lav) to 68.19 Wm-2
(IT_Col). This trend was also reflected by lower variance in
correlation coefficients (R2=0.936–0.970) and NASH index values
(0.943–0.981) for the per site averages. The evergreen needle-leaf
forest site, IT_Lav, consistently demonstrated the highest model
performance in simulating Rnet with a mean absolute error
value of 55.86, 8.79 Wm-2 lower than the overall
average. A weaker agreement was apparent between model predictions of
Rnet and the corresponding observed data in the deciduous
broadleaf site, IT_Col (68.19 Wm-2), which exhibited the
highest RMSD of all sites. MBE between sites showed significant
variability, ranging from a moderate underestimation of the in-situ
measurements over the evergreen broadleaf forest site
(-15.99 Wm-2), to a moderate overestimation within the
shrubland site (15.02 Wm-2). No clear trends in model
prediction accuracy dependent on site or land cover type could be identified
in our study results.
All in all, SimSphere was able to reproduce the evolution of
Rnet reasonably well in terms of both amplitude and
trend which is reflected in the low MSD values of all sites
(55.01–68.03 Wm-2), particularly so at sites such as
IT_Lav (55.01 Wm-2) and ES_Agu
(60.92 Wm-2). Generally, sites which recorded higher
scatter results also exhibited higher RMSD results – notably in sites
IT_Col (68.03 Wm-2) and FR_Pue
(66.60 Wm-2). Throughout, consistently high NASH values
further confirmed the high correspondence between model predictions
and observed data.
Latent heat (LE) flux
Results for the comparison between SimSphere estimated LE flux and the
CarboEurope in-situ LE measurements for all days combined exhibited an
overall average RMSD error of 62.75 Wm-2 and
a correlation coefficient value of 0.542 respectively
(Table 4). Figure 4 plots the LE flux from the in-situ measurements
against the corresponding predicted fluxes from SimSphere for all
simulation days from all experimental sites. Although RMSD for the LE
parameter showed a better agreement in comparison to the
Rnet parameter (Sect. 4.1), R2 was significantly
lower (a decrease of 0.408). As can be seen from Fig. 4, the
distribution of points shows an increased dispersion from the 1:1
line in comparison to the Rnet parameter. There was also
an apparent overestimation of the in-situ measurements by the model
for this parameter (MBE =15.78Wm-2). R2 values
varied significantly between all simulation days from 0.020–0.961
(Table 4), suggesting notable discrepancies between the predictions
and observations. Additionally, daily RMSD values also varied
significantly, reflecting the trends observed in the R2
statistics. RMSD varied from 22.08 to 86.45 Wm-2 between
all days of simulation. When analysed on a site by site basis, average
RMSD exhibited comparable ranges to those reported for the individual
simulation days, with RMSD varying from 37.25 Wm-2
(ES_Agu – Shrubland) to 75.36 Wm-2 (IT_Col, deciduous
broadleaf forest). On a per site basis, in overall, there were
noticeable differences in the magnitude of the daily evolution of
simulated LE when compared to the in-situ measurements. Specifically,
the ES_Agu shrubland site, consistently demonstrated above average
alikeness to the in-situ measurements with the lowest RMSD and MAE
values of all sites, 37.25 and 25.58 Wm-2
respectively. Lowest agreement between the LE fluxes predicted from
SimSphere and those from the in-situ measurements was in the IT_Col
deciduous broadleaf forest site (RMSD =75.36Wm-2,
MAE=55.86Wm-2) and IT_Mbo grasslands site
(RMSD =74.66Wm-2,
MAE=52.87Wm-2) respectively.
On the whole, SimSphere was consistent in terms of its ability to
reproduce in-situ LE fluxes, with low MSD values reported across the
majority of sites. However, the IT_Mbo (grassland) and IT_Ro3
(cropland) sites exhibited the largest MSD of 74.58 and
68.48 Wm-2 respectively, an increase of 15.64 and
9.54 Wm-2 on the overall average, suggesting a weaker
systematic replication of LE fluxes over those sites (Table 4). There
was a systematic overestimation of the in-situ measurements by the
model simulations for the majority of sites. The only exceptions were for the
IT_Mbo and IT_Ro3 sites, exhibiting a small average underestimation
(MBE) of -5.11 and -0.87 Wm-2 respectively.
Interestingly, both broad-leaf forest sites, IT_Col (deciduous
broad-leaf forest) and FR_Pue (evergreen broad-leaf forest), showed
the highest overestimation of LE fluxes with moderately high MBE
values of 33.67 and 37.56 Wm-2 respectively.
Sensible heat (H) flux
Figure 5 depicts the scatterplot of observed vs. simulated H flux
for all experimental sites, whilst Table 5 summarises the relevant
statistics concerning the comparisons between the simulated and
observed H fluxes for all the days/sites. Results consistently
indicated a high ability of the model to accurately simulate H
fluxes, with an average RMSD of 55.36 Wm-2 and an R2
value of 0.83. A significant improvement in accuracy of this
parameter in comparison to both the Rnet and LE parameters
was evident. H flux results exhibited a decrease in overall RMSD of
9.29 and 7.39 Wm-2 respectively. Similar trends were also
evident in both the MBE (-0.08 Wm-2) and MSD
(55.36 Wm-2) results for this parameter, where model
performance was better in comparison to both the Rnet and
LE parameters. Although with regards to R2, the H flux
parameter exhibited a minor decrease in correlation (0.83) compared to
the Rnet parameter. When examining the R2 values for
the individual simulation days, there was a significant variation in
both correlation coefficients (R2=0.607–0.982) and RMSD
(RMSD =20.03–91.07 Wm-2). Notably, there was no clear
trend between simulation accuracy and simulation day. Values ranged
from 35.50 Wm-2 (ES_Agu) to 80.41 Wm-2
(IT_Ro3) on a site by site basis. Similarly to LE flux, the ES_Agu
site reported the highest simulation accuracy
(RMSD =35.50Wm-2, R2=0.944,
MBE=-7.01Wm-2,
MSD =34.80Wm-2). On the contrary, the cropland site
IT_Ro3 consistently reported a less satisfactory agreement between
model prediction and in-situ data for H flux. Generally, SimSphere
was often unable to represent the peak of H flux across all sites
diurnally; this is shown by a scatter of peak values as reported in
Fig. 4. However, the model did neither consistently overestimate nor
underestimate H flux, but produced a range of bias values, with an
average error of -0.08 Wm-2. Both the FR_Pue and
ES_Lju sites showed a predominant underestimation of H flux at
-25.88 and -17.17 Wm-2 respectively. Yet, for the
IT_Mbo site, a moderate overestimation of 16.41 Wm-2 was
reported, suggesting land cover type may be related to simulation
accuracy, which can be subject of future investigations.
Air temperature at 1.3 m
(Tair1.3m)
Results obtained confirmed the ability of the model to simulate
Tair1.3m well, indicating a low average RMSD
of 4.1 ∘C and an average correlation coefficient of 0.631 for
all sites and days (Table 6, Fig. 6). Notably, results for R2
for the specific test days and study sites exhibited significant
variance, ranging from 0.237 to 0.939. Such results suggest that time
of year and land cover type, and in particular their effect on
vegetation, has a noticeable effect on the model's capability to
predict Tair1.3m. RMSD results also
exhibited variation between different test days and sites, with values
ranging from 1.32 to 7.13 ∘C.
When simulation accuracy was assessed on a site by site basis, average
RMSD ranged from 3.15 ∘C (IT_Ro3) to 5.12 ∘C
(IT_Col). All sites showed an overestimation of
Tair1.3m, with an average MBE of
3.33 ∘C. In addition to this, all sites reported low MSD,
with an average of just 2.30 ∘C. This appraises the model's
ability to repetitively simulate Tair1.3m to
a highly acceptable accuracy. The results for the specific sites
varied markedly. Simulation over the ES_Agu and IT_Ro3 sites
exhibited minor overestimation of the in-situ measurements, with an
MBE of 0.72 and 1.01 ∘C for both sites
respectively. Scatter results for both the IT-Lav and It Ro3 sites
were very low (and 2.84 and 2.99 ∘C), appraising
the model's ability to produce accurate and stable outputs over these
sites. Furthermore, the IT_Ro3 site also produced the highest
correlation coefficient (R2=0.769) and NASH index (0.769) of all
sites. Results for the IT_Col and ES_Lju sites exhibited an
increased overestimation of the in-situ measurements
(MBE =3.49∘C) compared to all other sites, with MSD
values of 3.74 and 3.95 ∘C respectively
indicating weaker model stability over these sites for all days
combined. Results for the ES_Lju site also exhibited lower NASH
(-0.054) and R2 (0.517) values in comparison to all other
sites. The latter indicated that the model had some difficulty in
reproducing the conditions represented by the in-situ data over the
olive orchard experimental site.
Air temperature at 50 m
(Tair50m)
Figure 7 shows the agreement difference in the simulated
Tair50m and corresponding in-situ from all
experimental sites/days included in this study. The results from the
statistical comparisons between the simulated and the measured diurnal
Tair50m for all the days of the experiment
for which observational data were available are summarised in
Table 7. As can be observed, the model showed slightly superior
performance in predicting Tair50m compared to
Tair1.3m, with a decrease of 0.45 ∘C
in overall RMSD to an average value of 3.54 ∘C. There was
a minor overestimation of Tair50m by the
model (1.40 ∘C); however, again, an improvement on the
results exhibited by the Tair1.3m parameter was
apparent.
R2 values per study day for Tair50m
showed an increased variability in comparison to the
Tair1.3m parameter, with the overall range
in values increasing by 0.173 (0.055 to 0.930). However, daily average
RMSD exhibited significantly less variability between sites, ranging
from 1.06 ∘C (IT_Lav) to 6.49 ∘C (FR_Pue), an
improvement of 0.038 ∘C on the ranges displayed by the
Tair1.3m results. On a site by site basis,
the average range in RMSD in comparison to
Tair1.3m decreases considerably again, with
RMSD ranging from 2.87 ∘C (IT_Mbo) to 4.25 ∘C
(ES-Lju) for the Tair50m parameter. When
considering simulation accuracy on a per site basis, it was evident
that there were significant differences between the accuracy of the
model in simulating both the Tair1.3m and
the Tair50m parameters over the different
sites included in our study. Highest simulation accuracy was reported
within the It_Mbo (grassland) and IT_Lav (evergreen needle leaf)
sites for the Tair50m parameter, whereas in
comparison, IT_Ro3 (cropland) and ES_Agu (shrubland) were the most
accurate sites in terms of Tair1.3m
prediction by SimSphere. Both parameters were consistently simulated
to high statistical accuracy over the IT_Lav study site. Less
satisfactory simulation accuracy was exhibited within the ES_Lju
(olive orchards) site for both.
As a whole, the diurnal course of the temperatures predicted by
SimSphere was also found to be largely realistically reproduced by the
model for most days. A minor overestimation of
Tair50m was reported for all validation sites
used in this study, with an overall MBE of just
1.35 ∘C for all days simulated. The extent to which each site
overestimated Tair50m was comparable, with
a very low range in MBE results from 0.03 ∘C (ES_Agu) to
2.66 ∘C (FR_Pue), further appraising the model's ability to
produce accurate outputs. Furthermore, such results are a significant
improvement on those reported earlier for the
Tair1.3m parameter. For all days of
simulation, low MSD values were also obtained, with an average MSD of
just 3.15 ∘C. Although there was a slight increase on values
reported for the Tair1.3m parameter, results
reported still indicate a satisfactory agreement with the in-situ
data.
Discussion
This study evaluated the ability of the SimSphere land biosphere model
to simulate key parameters characterising the Earth's energy and water
budget in several European ecosystems. The model was parameterised for
a total of 7 CarboEurope sites, representative of a range of ecosystem
and environmental conditions. A total of 70 days (10 days per site)
from the year 2011 were selected to validate the model's ability to
predict Net Radiation (Rnet), Latent Heat (LE), Sensible
Heat (H), and Air temperature (Tair) at 1.3 and
50 m. The agreement between the two datasets was evaluated
based on a series of computed statistical metrics.
At all sites, Rnet was systematically well represented by
the model, with an average overall RMSD of 64.65 Wm-2. In
comparison to previous similar validation experiments conducted on
earlier SimSphere versions, simulation accuracy of Rnet
reported here is higher, for example more than 20 Wm-2 in
comparison to Ross and Oke (1988) who validated Rnet over
an urban environment. Respectively, ecosystems which presented high
inter-annual change of vegetation phenology, namely olive plantation
(ES_Lju), grassland (IT_Mbo), cropland (IT_Ro3) and deciduous
broadleaf forest (IT_Col) sites all exhibited distinct periods where
model performance was increased. However, these results are
significantly better in comparison to those reported by Marshall
et al. (2013), who in a similar validation study of the model reported
average RMSE of up to 118.46 Wm-2. Akkermans
et al. (2014) noted that Rnet prediction accuracy is also
largely dependable on the vegetation and surface characteristics of
the respective site and model performance is highly reliant on its
representation of the surface vegetation and soil optical properties,
most notably surface albedos and emissivities (Falge et al., 2005).
SimSphere showed increased model performance in simulating both LE and
H fluxes in comparison to Rnet; this is confirmed by the
low average RMSD and high overall R2 as reported in Tables 4 and
5. Apart from the general overestimation of LE
(MBE =15.78Wm-2), results reported show largely
acceptable simulation accuracies compared to other analogous
studies. Ross and Oke (1988) performed a validation of a previous
version of SimSphere over an urban environment of Vancouver, BC.
Authors reported acceptable agreement between model output and
observed in-situ for H flux (average RMSE =56Wm-2)
but significant average error distributions for LE fluxes
(RMSE =107Wm-2). Todhunter and Terjung (1987) further
described in detail how earlier versions of the SimSphere model
dissipated too much of Rnet as LE and too little to be
lost to H, this correlates well to Ross and Oke's (1988) findings but
also the findings reported within; where average bias values indicate
general net overestimations of LE flux in the order of
15.78 Wm-2, compared to the slight average
underestimation of H at -0.08 Wm-2.
The shrubland site ES_Agu consistently showed remarkably low average
RMSD in all parameters assessed, particularly so for LE and H
fluxes. This is likely to be a function of the site's location within
a water limited environment, where transpiration effects are much
lower in amplitude and thus more predictable, especially given the
site's relative homogeneity (Maayar et al., 2001). Marshall
et al. (2013) have also suggested that ecosystems which exhibit
increased stand complexity and heterogeneity, such as forested
environments (particularly those with understory vegetation) can have
a profound effect on the overall exchange of mass and energy. The
latter cannot be fully represented within the model's
parameterisation, therefore accounting for poorer simulation
accuracies of LE and H. Additionally, it is widely reported that soil
water content is an imperative control to the simulation accuracy of
LE and H (Oltchev et al., 2002; Falge et al., 2005). Within our
study, soil moisture availability and root zone moisture availability,
two of the most sensitive parameters to LE and H flux partitioning
(see for example SA study of Petropoulos et al., 2013a, 2014a), were
acquired directly from the corresponding daily in-situ
measurements. Akkermans et al. (2014) stated that underestimations of
LE can largely be attributed to overestimations of H fluxes. Such
effects were seen most prominently in our validation site ES_Lju,
where a general underestimation of LE (MBE =-17.17 Wm-2) partly contributed to the significant
overestimation of H flux (MBE =21.09Wm-2).
The model also consistently indicated a satisfactory capability
in simulating Tair1.3m and
Tair50m in all ecosystems in which it was
assessed, with average RMSD similar to values reported in other
analogous studies (Ross and Oke, 1988). Poorer simulation accuracies
of Tair1.3m were reported in stands where
vegetation height exceeds 1.3 m; this is most noticeable in
sites ES_Lju, IT_Col and FR_Pue. This suggests that the in-situ
data at 1.3 m has a limited representation of the overall
transfer of energy and heat seen within the stand; this can
explain in part why the model often portrays a general overestimation
of Tair1.3m at these particular sites.
However, when model predictions are evaluated at 50 m the
agreement between modelled and predicted Tair is much
stronger, with an average RMSD error of 0.6 ∘C lower than
Tair1.3m. Ross and Oke (1988) noted that
peak values of air temperature should be observed between
10:30–14:30 LST, this is in close correlation to this present study,
further appraising SimSphere's representation of Tair at
both 1.3 and 50 m.
It is also apparent that SimSphere fulfils all 3 of Kramer
et al.'s (2002) model assessment criteria, namely accuracy, generality
and realism. No significant prediction errors occurred within all of
the parameters analysed, further appraising the model's ability to
represent numerous environments accurately. Temporal patterns of the
predicted parameters were consistent with the patterns found in the
corresponding field data, indicating a strong influence of
environmental forcing variables (such as global radiation or vapour
pressure deficit) on model output. This result is also in agreement to
previous SimSphere validation studies (Ross and Oke, 1988). SimSphere
has shown high levels of generality, with acceptable simulation
accuracies attained in all evaluated sites. In order to improve the
model's generality, the inclusion of more northern European sites
would act to further test the models applicability within European
ecosystems. Realism has been most notable in the simulation of LE and
H fluxes, where slight changes in the vegetation phenology or soil
surface moisture was accountable for characterising the diurnal
evolution of fluxes in all validated sites. On this basis, SimSphere
has shown itself to be highly capable of simulating the observed
fluxes in both terms of trend and amplitude, with systematically
accurate representation of the seasonal effects of vegetation change
to flux characteristics.
In the overall evaluation of the results reported, instrumentation
uncertainty in the measured parameters themselves should also be
partially taken into account when attempting to explain the
disagreement between the simulated and observed parameters (Baldocchi
et al., 2001; Oncley et al., 2007; Verbeeck et al., 2009). Generally,
Rnet measurement accuracy error is in the order of
10 %, although, an additional 10 % instrumentation uncertainty
should be added due to limited view angle/measuring volume (especially
in the case of rugged terrains) (Baldocchi et al., 2001). Typical
uncertainty in the estimation of the LE and H fluxes using the eddy
covariance method generally varies between 10 to 20 % but can be
much higher during periods of low flux magnitude and/or limited
turbulent mixing such as at night (Petropoulos et al., 2013c). For
example, Hollinger and Richardson (2005) showed that uncertainty in
flux measurements are inversely proportional to magnitude; the smaller
the flux the greater the relative uncertainty. Also, it should be
noted that for some days included in our comparisons, a characteristic
of the acquired in-situ data for those days was the presence of many
spikes (indicative of very high or very low values). Probable reasons
for those spikes could be instrumental errors, horizontal advection of
H2O and CO2, footprint changes as well as
a non-stationarity of turbulent regime within the atmospheric surface
layer (Papale et al., 2006; Olchev et al., 2008). For those days,
comparisons resulted in a somewhat lower accuracy of model predictions
as such conditions cannot be replicated by the model which assumes
homogeneity of vegetation canopy and ignores horizontal advection. In
terms of SimSphere parameterisation, it is important to note that
understory effects of vegetation is a critical influence missing from
the model's parameterisation, along with the model's representation of
multiple vegetation types. The latter can have a significant effect in
more complex vegetation stands (for example the increased presence of
understory vegetation in forested environments). This might also be in
part responsible for the comparatively poorer overall simulation
accuracies exhibited by the model at times.
On the whole, despite the occasionally inferior performance of the
model in simulating the examined parameters for some days/sites,
SimSphere predictions are significant in terms of the representation
of the physical and dynamic processes involved in the interactions of
the complex nature of the soil-land–atmosphere system. Moreover, it
is important to recognise that uncertainty is inevitable in any model,
as a model will never be as complex as the reality it
portrays (Denti, 2004). In this way, SimSphere fulfils its objective as a tool
to identify expected patterns of change, if not always the
magnitudes. The latter indicates its usefulness in practical
applications either as a stand-alone tool or in combination with EO
data, as done for instance through the implementation of the
“triangle” data assimilation technique of Carlson (2007).
Concluding remarks
In this paper, key findings from a large scale validation of the
SimSphere land biosphere model in numerous European environments are
reported. In total, 7 different ecosystems were chosen for validation
with 70 simulations made for cloud free days in 2011. A systematic
statistical analysis was employed to assess the agreement between
model predictions and corresponding in-situ measurements. To our
knowledge, this is the first study of its kind, reporting results from
an in-depth validation of this models' ability in accurately
simulating key parameters characterising land surface processes,
particularly so in European ecosystems.
In overall, model performance can largely be described as satisfactory
for most of the experimental sites and parameters which were
evaluated. Results were also largely comparable to other similar
validation attempts of earlier versions of the model performed in
dissimilar experimental settings (Todhunter and Terjung, 1987; Ross
and Oke, 1988). SimSphere was found to be able to
reproduce the diurnal evolution of key parameters at accuracies
similar to those reported by others evaluating different SVAT models
(Ridler et al., 2012; Marshall et al., 2013; Akkermans et al., 2014).
Many factors were identified as having a noteworthy effect on
simulation accuracy.
Model comparisons similar to the one conducted in this study can
advance our understanding on the amount of complexity required for
adequate representation of land surface processes and interactions
between different components of our Earth system. An evaluation and
analysis of a model performance allows for an increased understanding
of the model's representation and helps to identify possible
misrepresentations within the observational data. Thus, reported
discrepancies found in any validation study such as ours should indeed be
regarded as a positive step when evaluating model performance (Denti,
2004; Verbeeck et al., 2009). However, as noted by Denti (2004), any
land surface model, by its definition, will never be as complex as the
reality it portrays. Nevertheless, in overall, the validation results
of this study provide further independent evidence that
SimSphere has a high capability of simulating parameters associated
with the Earth's energy balance.
Further efforts should be made to validate SimSphere to numerous
global ecosystems to assess its applicability as a universally applied
SVAT model. Moreover, as SimSphere's use is being explored
synergistically with EO data, perhaps future efforts should be
directed towards performing a detailed error budget assessment and
evaluating the overriding effects of SimSphere predictions to the
overall prediction error of the spatio-temporal estimates of energy
fluxes and soil moisture derived from its implementation within the
“triangle” technique. These topics and results will be discussed in
the next issues.
Acknowledgements
G. P. Petropoulos gratefully acknowledges the financial support
provided by the European Commission under the Marie Curie Career
Re-Integration Grant “TRANSFORM-EO” project for the completion of
this work. Authors would also like to thank Daisy Rendall and
Thalia Tataris for their positive contributions to data processing
and model implementation. We would also like to thank the PI's of
the CarboEurope network for sharing the acquired data of their
experimental sites which made this study possible.
ReferencesAkkermans, T., Lauwaet, D., Demuzere, M., Vogel, G.,
Nouvellon, Y., Ardö, J., and Van Lipzig, N.: Validation and
comparison of two soil–vegetation–atmosphere transfer models for
tropical Africa, J. Geophys. Res.-Biogeo., 117, G02013, 10.1029/2011JG001802, 2012.Akkermans, T., Thiery, W., and Van Lipzig, N. P.: The
regional climate impact of a realistic future, J. Climate, 27, 2714–2734,
10.1175/JCLI-D-13-00361.1, 2014. Amri, R., Zribi, M., Lili-Chabaane, Z., Szczypta, C.,
Calvet, J. C., and Boulet, G.: FAO-56 Dual Model Combined with
Multi-Sensor Remote Sensing for Regional Evapotranspiration
Estimations, Remote Sens., 6, 5387–5406, 2014. Anderson, M. C., Norman, J. M., Kustas, W. P.,
Houborg, R., Starks, P. J., and Agam, N.: A thermal-based remote
sensing technique for routine mapping of land-surface carbon, water
and energy fluxes from field to regional scales, Remote
Sens. Environ., 112, 4227–4241, 2008. Baldocchi, D., Falge, E., Gu, L., Olson, R.,
Hollinger, D., Running, S., and Wofsy, S.: FLUXNET: a new tool to
study the temporal and spatial variability of ecosystem-scale carbon
dioxide, water vapour, and energy flux
densities, B. Am. Meteorol. Soc., 82, 2415–2434, 2001. Bellocchi, G., Rivington, M., Donatelli, M., and
Matthews, K.: Validation of biophysical models: issues and
methodologies. A review, Agron. Sustain. Dev., 30, 109–113, 2010. Braud, I., Dantas-Antonino, A. C., Vauclin, M.,
Thony, J. L., and Ruelle, P.: A simple soil–plant–atmosphere
transfer model (SiSPAT) development and field
verification, J. Hydrol., 166, 213–250, 1995. Burt, J. E. and Barber, G. M.: Elementary Statistics for
Geographers, Longman Ed., London, 504 pp., 1996. Carlson, T. N.: An overview of the “triangle method”
for estimating surface evapotranspiration and soil moisture from
satellite imagery, Sensors, 7, 1612–1629, 2007. Carlson, T. N. and Boland, F. E.: Analysis of urban–rural
canopy using a surface heat flux/temperature model, J. Appl. Meteorol., 17, 998–1014, 1978. Carlson, T. N. and Lynn, B.: The effects of plant water
storage on transpiration and radiometric surface temperature,
Agr. Forest Meteorol., 57, 171–186, 1991. Carlson, T. N., Dodd, J. K., Benjamin, S. G., and
Cooper, J. N.: Satellite estimation of the surface energy balance,
moisture availability and thermal inertia, J. Appl. Meteorol., 20, 6–87, 1981. Carlson, T. N., Belles, J. E., and Gillies, R. R.:
Transient water stress in a vegetation canopy: simulations and
measurements, Remote Sens. Environ., 35, 175–186, 1991. Chauhan, N. S., Miller, S., and Ardanuy, P.: Spaceborne
soil moisture estimation at high resolution: a microwave-optical/IR
synergistic approach, Int. J. Remote Sens., 22, 4599–4646, 2003. Clapp, R. B. and Hornberger, G. M.: Empirical equations
for some soil hydraulic-properties, Water Resour. Res., 14, 601–604, 1978. Coudert, B., Ottlé, C., and Briottet, X.: Monitoring
land surface processes with thermal infrared data: calibration of
SVAT parameters based on the optimisation of diurnal surface
temperature cycling features, Remote Sens. Environ., 112, 872–887, 2008. Culf, A. D., Folken, T., and Gash, J. H. C.: The energy
balance closure problem, in: Vegetation, Water, Humans and the
Climate, Springer-Verlag, Berlin, 2002. Deardoff, J. W.: Efficient prediction of ground surface
temperature and moisture inclusion of a layer of
vegetation, J. Geophys. Res., 83, 1889–1903, 1978. Denti, G.: Developing a desertification indicator system
for a small Mediterranean catchment: a case study from the Serra De
Rodes, Alt Emporda, Catalunya, NE Spain, PhD thesis, University of
Girona, Girona, 2004. European Commission: White Paper, Adapting to climate
change: towards a European framework for action, COM, Brussels, 1–16, 2009.European Space Agency, Support to Science Element.:
A pathfinder for innovation in Earth Observation, ESA, 2012,
available at: http://due.esrin.esa.int/stse/files/document/STSE_report_121016.pdf,
last access: 10 July 2013. Falge, E., Reth, S., Brüggemann, N.,
Butterbach-Bahl, K., Goldberg, V., Oltchev, A., and Bernhofer, C.:
Comparison of surface energy exchange models with eddy flux data in
forest and grassland ecosystems of Germany, Ecol. Model., 188, 174–216, 2005. Feddema, J. J., Oleson, K. W., Bonan, G. B.,
Mearns, L. O., Buja, L. E., Meehl. G. A., and Washington, W. M.: The
importance of land-cover change in simulating future climates,
Science, 310, 1674–1678, 2005.Foken, T., Wimmer, F., Mauder, M., Thomas, C., and
Liebethal, C.: Some aspects of the energy balance closure problem,
Atmos. Chem. Phys., 6, 4395–4402, doi:10.5194/acp-6-4395-2006, 2006. Gan, Y., Duan, Q., Gong, W., Tong, C., Sun, Y., Chu, W.,
Ye, A., Miao, C., and Di, Z.: A comprehensive evaluation of various
sensitivity analysis methods: a case study with a hydrological
model, Environ. Modell. Softw., 51, 269–285, 2014.Giertz, S., Diekkrüger, B., and Steup, G.:
Physically-based modelling of hydrological processes in a tropical
headwater catchment (West Africa) – process representation and
multi-criteria validation, Hydrol. Earth Syst. Sci., 10, 829–847,
doi:10.5194/hess-10-829-2006, 2006. Gillies, R. R.: A physically-based land sue
classification scheme using remote solar and thermal infrared
measurements suitable for describing urbanisation, PhD thesis,
University of Newcastle, Newcastle, UK, 121 pp., 1993. Gillies, R. R., Carlson, T. N., Cui, J., Kustas, W. P.,
and Humes, K. S.: Verification of the “triangle” method for
obtaining surface soil water content and energy fluxes from remote
measurements of the Normalized Difference Vegetation Index NDVI and
surface radiant temperature, Int. J. Remote Sens., 18, 3145–3166, 1997. Hollinger, D. and Richardson, A.: Uncertainty in eddy
covariance measurements and its application to physiological models,
Tree Physiol., 25, 873–885, 2005.IPCC, Summary report of the IPCC expert meeting on the
science of alternative metrics 18–20 March 2009, Oslo, Norway,
IPCC-XXX/Doc.13 (31.III.2009), available at:
www.ipcc.ch/meetings/session30/doc13.pdf, last access: 8 December 2014.Kramer, K., Leinonen, I., Bartelink, H. H.,
Berbigier, P., Borghetti, M., Bernhofer, C., and Vesala, T.:
Evaluation of six process-based forest growth models using
eddy-covariance measurements of CO2 and H2O fluxes
at six forest sites in Europe, Global Change Biol., 8, 213–230, 2002. Li, Z. Q., Yu, G. R., Wen, X. F., Zhang, L. M.,
Re, C. Y., and Fu, Y. L.: Energy balance closure at ChinaFLUX sites,
Sci. China Ser. D., 48, 51–62, 2005. Liu, Y., Hiyama, T., and Yamaguchi, Y.: Scaling of land
surface temperature using satellite data: a case examination on
ASTER and MODIS products over a heterogeneous terrain area,
Remote Sens. Environ., 105, 115–128, 2006. Lynn, B. H. and Carlson, T. N.: A stomatal resistance
model illustrating plant vs. external control of
transpiration, Agr. Forest Meteorol., 52, 5–43, 1990. Maayar, M., Price, D. T., Delire, C., Foley, J. A.,
Black, T. A., and Bessemoulin, P.: Validation of the Integrated
Biosphere Simulator over Canadian deciduous and coniferous boreal
forest stands, J. Geophys. Res.-Atmos., 106, 14339–14355, 2001. Manabe, S.: Climate and the ocean circulation 1. The
atmospheric circulation and the hydrology of the earth's surface,
Mon. Weather. Rev., 97, 739–774, 1969.Marshall, M., Tu, K., Funk, C., Michaelsen, J.,
Williams, P., Williams, C., Ardö, J., Boucher, M.,
Cappelaere, B., de Grandcourt, A., Nickless, A., Nouvellon, Y.,
Scholes, R., and Kutsch, W.: Improving operational land surface
model canopy evapotranspiration in Africa using a direct remote
sensing approach, Hydrol. Earth Syst. Sci., 17, 1079–1091,
doi:10.5194/hess-17-1079-2013, 2013. Mascart, P., Taconet, O., Pinty, J. P., and Mehrez, M. B.:
Canopy resistance formulation and its effect in mesoscale models:
a HAPEX perspective, Agr. Forest Meteorol., 54, 319–351, 1991. Moffat, A., Papale, D., Reichstein, M., Hollinger, D. Y.,
Richardson, A. D., Barr, A. G., and Beckstein. C.: Comprehensive
comparison of gap-filling techniques for eddy covariance net carbon
fluxes, Agr. Forest Meteorol., 147, 209–232, 2007. Monin, A. S. and Obukhov, A.: Basic laws of turbulent
mixing in the surface layer of the atmosphere, Contrib. Geophys. Inst. Acad. Sci. USSR, 151, 163–187, 1954. Olchev, A., Ibrom, A., Ross, T., Falk, U., Rakkibu, G.,
Radler, K., and Gravenhorst, G.: A modelling approach for simulation
of water and carbon dioxide exchange between multi-species tropical
rain forest and the atmosphere, Ecol. Model., 212, 122–130, 2008. Olioso, A., Inoue, Y., Farias, O., Demarty, J.,
Widneron, J. P., Braud, I., Jacob, F., Lecharpentier, P., Ottle, C.,
Calvet, J. C., and Brisson, N.: Future directions for advanced
evapotranspiration modelling: assimilation of remote sensing data
into crop simulation models and SVAT models, Irrig. Drain., 19, 377–412, 2005. Oliphant, A. J., Grimmond, C. S. B., Zutter, H. N.,
Schmid, H. P., Su, H.-B., Scott, S. L., Offerle, B., Randolph, J. C.,
and Ehman, J.: Heat storage and energy balance fluxes for
a temperate deciduous forest, Agr. Forest Meteorol., 126, 185–201, 2004. Oltchev, A., Cermak, J., Nadezhdina, N., Tatarinov, F.,
Tishenko, A., Ibrom, A., and Gravenhorst, G.: Transpiration of
a mixed forest stand: field measurements and simulation using SVAT
models, Boreal Environ. Res., 7, 389–397, 2002. Oncley, S. P., Foken, T., Vogt, R., Kohsiek, W.,
DeBruin, H. A. R, Berhofer, C., Christen, A., Van Gorsel, E.,
Grantz, D., Feigenwinter, C., Lehner, I., Liebethal, C., Liu, H.,
Mauder, M., Pitacco, A., Ribeiro, L., and Weidinger, T.: The Energy
Balance Experiment EBEX-2000, Part I: Overview and energy balance,
Bound.-Lay. Meteorol., 123, 1–28, 2007.Papale, D., Reichstein, M., Aubinet, M., Canfora, E.,
Bernhofer, C., Kutsch, W., Longdoz, B., Rambal, S., Valentini, R.,
Vesala, T., and Yakir, D.: Towards a standardized processing of Net
Ecosystem Exchange measured with eddy covariance technique:
algorithms and uncertainty estimation, Biogeosciences, 3, 571–583,
doi:10.5194/bg-3-571-2006, 2006. Petropoulos, G. P. and Carlson, T. N.: Retrievals of
turbulent heat fluxes and soil moisture content by remote sensing,
in: Advances in Environmental Remote Sensing: Sensors, Algorithms,
and Applications, chap. 19, Taylor and Francis, Boca Raton, Florida, 556, 667–502, 2011. Petropoulos, G. P., Carlson, T., and Wooster, M. J.: An
overview of the use of the SimSphere Soil vegetation Atmospheric
Transfer (SVAT) model for the study of land atmosphere interactions,
Sensors, 9, 4286–4308, 2009a. Petropoulos, G. P., Wooster, M. J., Kennedy, K.,
Carlson, T. N., and Scholze, M.: A global sensitivity analysis study
of the 1-D SimSphere SVAT model using the GEM SA software,
Ecol. Model., 220, 2427–2440, 2009b. Petropoulos, G. P., Griffiths, H. M., and Tarantola, S.:
Towards operational products development from earth observation:
exploration of simsphere land surface process model sensitivity
using a GSA approach, in: 7th International Conference on
Sensitivity Analysis of 25 Model Output, 1–4 July 2013, Nice, France, 2013a. Petropoulos, G. P., Griffiths, H., and Tarantola, S.:
Sensitivity analysis of the SimSphere SVAT model in the context of
EO-based operational products development, Environ. Modell. Softw.,
49, 166–179, 2013b. Petropoulos, G. P., Konstas, I., and Carlson, T. N.:
Automation of SimSphere Land Surface Model Use as a Standalone
Application and Integration with EO Data for Deriving Key Land
Surface Parameters, European Geosciences Union, 7–12 April 2013,
Vienna, Austria, 2013c.Petropoulos, G. P., Griffiths, H. M., Carlson, T. N.,
Ioannou-Katidis, P., and Holt, T.: SimSphere model sensitivity
analysis towards establishing its use for deriving key parameters
characterising land surface interactions, Geosci. Model Dev., 7,
1873–1887, doi:10.5194/gmd-7-1873-2014, 2014a. Petropoulos, G. P., Griffiths, H. M., and
Ioannou-Katidis, P.: Sensitivity exploration of SimSphere land
surface model towards its use for operational products development
from Earth observation data, in: chap. 14, Advancement in Remote Sensing for
Environmental Applications, edited by: Mukherjee, S., Gupta, M.,
Srivastava, P. K., and Islam, T.,
Springer, Switzerland, 2014b. Piles, M., Camps, A., Vall-Llossera, M., Corbella, I.,
Panciera, R., Rudiger, C., Kerr, Y. H., and Walker, J.: Downscaling
SMOS-derived soil moisture using MODIS visible/infrared data, IEEE
Geosci. Remote S., 49, 3156–3166, 2011.Prentice, I. C., Liang, X., Medlyn, B. E., and
Wang, Y.-P.: Reliable, robust and realistic: the three R's of
next-generation land surface modelling, Atmos. Chem. Phys. Discuss.,
14, 24811–24861, doi:10.5194/acpd-14-24811-2014, 2014. Quintana-Seguí, P., Le Moigne, P., Durand, Y.,
Martin, E., Habets, F., Baillon, M., and Morel, S.: Analysis of
near-surface atmospheric variables: validation of the SAFRAN
analysis over France, J. Appl. Meteorol. Clim., 47, 92–107, 2008. Ridler, M. E., Sandholt, I., Butts, M., Lerer, S.,
Mougin, E., Timouk, F., and Madsen, H.: Calibrating
a soil–vegetation–atmosphere transfer model with remote sensing
estimates of surface temperature and soil surface moisture in
a semi-arid environment, J. Hydrol., 436, 1–12, 2012. Ross, S. L. and Oke, T. R.: Tests of three urban energy
balance models, Bound-Lay. Meteorol., 44, 73–96, 1988. Samaali, M., Courault, D., Bruse, M., Olioso, A., and
Occelli, R.: Analysis of a 3-D boundary layer model at local scale:
validation on soybean surface radiative measurements,
Atmos. Res., 85, 183–198, 2007.Schrodin, R. and Heise, E.: A new multi-layer version of
the DWD Soil Model TERRA_LM, Cosmo Technical Report, No. 2,
available at: www.cosmo-model.org (last access: 16 December 2014), 2002. Sheikh, V., Visser, S., and Stroosnijder, L.: A simple
model to predict soil moisture: Bridging Event and Continuous
Hydrological (BEACH) modelling, Environ. Modell. Softw., 24, 542–556, 2009. Silk, J.: Statistical Concepts in Geography, Harper Collins, London, 275 pp., 1979.Second Space for Hydrology Workshop, European Space
Agency (ESA): available at: http://earth.esa.int/hydrospace07/,
last access: 16 December 2014.State of Hydrological Observation Networks, World
Meteorological Organization (WMO): available at:
http://earth.esa.int/hydrospace07/participants/84231/pres_84231.pdf
(last access: 16 December 2014), 2002. Steinhaeuser, K., Ganguly, A. R., and Chawla, N. V.:
Multivariate and multi-scale dependence in the global climate system
revealed through complex networks, Clim. Dynam., 39, 889–895, 2012. Todhunter, P. E. and Terjung, W. H.: Intercomparison of
three urban climate models, Bound-Lay. Meteorol., 42, 181–205, 1988. Tomlin, A. S.: The role of sensitivity and uncertainty
analysis in combustion modelling, P. Combust. Inst., 34, 159–176, 2013. Twine, T. E., Kustas, W. P., Norman, J. M., Cook, D. R.,
Houser, P. R., Meyers, T. P., Prueger, J. H., Starks, P. J., and
Wesely, M. L.: Correcting eddy-covariance flux underestimates over
a grassland, Agr. Forest Meteorol., 103, 279–300, 2000. Vanuytrecht, E., Raes, D., and Willems, P.: Global
sensitivity analysis of yield output from the water productivity
model, Environ. Modell. Softw., 51, 323–332, 2014. Verbeeck, H., Samson, R., Granier, A., Montpied, P., and
Lemeur, R.: Multi-year model analysis of GPP in a temperate beech
forest in France, Ecol. Model., 210, 85–109, 2008. Wang, Z., Wang, P., and Li, X.: Using MODIS Land Surface
Temperature and Normalised Difference Vegetation Index products for
monitoring drought in the southern Great Plains, USA,
Int. J. Remote. Sens., 25, 61–72, 2004. Willmott, C. J.: Some comments on the evaluation of model
performance, B. Am. Meteorol. Soc., 63, 1309–1313, 1982. Wilson, K., Goldstein, A., Falge, E., Aubinet, M.,
Baldocchi, D., Berbigier, P., and Verma, S.: Energy balance closure
at FLUXNET sites, Agr. Forest Meteorol., 113, 223–243, 2002. Wilson, K. B. and Baldocchi, D. D.: Seasonal and
inter-annual variability of energy fluxes over a broadleaved
temperate deciduous forest in North America, Agr. Forest Meteorol.,
100, 1–18, 2000.
Summary of the main SimSphere inputs. The units of each of the model inputs are also provided in parentheses where applicable.
Name of the model inputProcess in which parameter is involvedMin valueMax valueSlope (degrees)TIME and LOCATION045Aspect (degrees)TIME and LOCATION0360Station Height (meters)TIME and LOCATION04.92Fractional Vegetation Cover (%)VEGETATION0100LAI (m2m-2)VEGETATION010Foliage emissivity (unitless)VEGETATION0.9510.990[Ca] (external [CO2] in the leaf) (ppmv)VEGETATION250710[Ci] (internal [CO2] in the leaf) (ppmv)VEGETATION110400[03] (ozone concentration in the air) (ppmv)VEGETATION0.00.25Vegetation height (meters)VEGETATION0.02120.0Leaf width (meters)VEGETATION0.0121.0Minimum Stomatal Resistance (sm-1)PLANT10500Cuticle Resistance (sm-1)PLANT2002000Critical leaf water potential (bar)PLANT-30-5Critical solar parameter (Wm-2)PLANT25300Stem resistance (sm-1)PLANT0.0110.150Surface Moisture Availability (vol/vol)HYDROLOGICAL01Root Zone Moisture Availability (vol/vol)HYDROLOGICAL01Substrate Max. Volum. Water Content (vol/vol)HYDROLOGICAL0.011Substrate climatol. mean temperature (∘C)SURFACE2030Thermal inertia (Wm-2K-1)SURFACE3.530Ground emissivity (unitless)SURFACE0.9510.980Atmospheric Precipitable water (cm)METEOROLOGICAL0.055Surface roughness (meters)METEOROLOGICAL0.022.0Obstacle height (meters)METEOROLOGICAL0.022.0Fractional Cloud Cover (%)METEOROLOGICAL110RKS (satur. thermal conduct., Cosby et al., 1984)SOIL010Cosby B (see Cosby et al., 1984)SOIL2.012.0THM (satur.vol. water cont.) (Cosby et al., 1984)SOIL0.30.5PSI (satur. water potential) (Cosby et al., 1984)SOIL17Wind direction (degrees)WIND SOUNDING PROFILE0360Wind speed (knots)WIND SOUNDING PROFILE––Altitude (1000's feet)WIND SOUNDING PROFILE––Pressure (mBar)MOISTURE SOUNDING PROFILE––Temperature (Celsius)MOISTURE SOUNDING PROFILE––Temperature-Dewpoint Temperature (Celsius)MOISTURE SOUNDING PROFILE––
Some of the main characteristics of the selected CarboEurope sites used for SimSphere validation.
Site NameSiteCountyGeographicPFTEcosystem TypeDominant SpeciesElevationClimateAbbreviationLocationLlano de los JuanesES_LjuSPAIN36.9266/-2.1521OLIOlive PlantationOleaeuropea, Macchia1622 mWarm Temperate with dry, hot summerCollelongo-SelvaPianaIT_ColITALY41.8493/13.5881DBFDeciduous Broadleaf ForestFagus sylvatica1645 mWarm temperate fully humid with warm summerMonte ModoneIT_MboITALY46.0296/11.0829GRAGrasslandAlpine meadow1547 mSnow fully humid warm summerAguamargaES_AguSPAIN36.8347/-2.2511SHRAnnual Broadleaf ShrubSumac (Rhus), Toyon (Heteromeles) and Coffeeberry (Rhamnus) Species195 mArid Steppe ColdLavaroneIT_LavITALY45.9553/11.2812ENLEvergreen Needle Leaf forestPinus sylvestris1353 mWarm temperate fully humid with warm summerPuechabonFR_PueFRANCE43.7414/3.5958EBFEvergreen Broadleaf forestQuercus ilex211 mWarm Temperate with dry, hot summerRoccarepampaniIT_Ro3ITALY42.3753/11.9154CROCroplandCereal Crop320 mWarm Temperate with dry, hot summer
An overview of the statistical measures implemented in this study to evaluate SimSphere's outputs against the corresponding in-situ data.
NameDescriptionMathematical DefinitionBias/MBEBias (accuracy) or Mean Bias Errorbias=1N∑i=1N(Pi-Oi)R2Linear Correlation Coefficient of Determination of Pi to OiR2=[∑i=1NPi-P‾Oi-O‾/∑i=1NOi-O‾2∑i=1NPi-O‾20.52Scatter/MSDScatter (precision) or Mean Square Differencescatter=1(N-1)∑i=1NPi-Oi-(Pi-Oi)‾2RMSDRoot Mean Square DifferenceRMSD=bias2+scatter2MAEMean Absolute ErrorMAD=N-1∑i=1N|Pi-Oi|NASHNash Sutcliffe EfficiencyNASH=1-∑i=1N(Oi-Si)2∑i=1NOi-O‾2
An overview of Rnet simulation accuracy.
SitePFTDayStatistical Test SitePFTDayStatistical Test BiasScatterRMSDMAENASHBiasScatterRMSDMAENASHES_LjuOLI14 Apr 2011-24.5542.3148.9132.450.921IT_Ro3CRO9 Apr 2011-8.2085.7686.1676.400.9129 May 2011-19.3460.3163.3347.550.97611 Apr 2011-52.8746.2170.2255.970.91324 Jun 201112.1867.5468.6357.970.91618 Apr 201113.7480.8882.0372.170.99027 Jun 20116.0666.9867.2547.260.97821 Apr 201124.9556.3461.6255.090.98219 Jul 201126.0557.3863.0144.210.93420 Jun 2011-12.5153.1554.6048.950.93728 Jul 201134.5256.1265.8947.600.97126 Jun 2011-22.3648.3953.3042.700.9724 Aug 201115.0651.0853.2533.810.93024 Aug 201113.9454.5356.2841.840.96122 Aug 20118.2657.5558.1447.330.89928 Aug 2011-8.9859.9560.6251.200.89925 Aug 201110.2359.0359.9149.440.9789 Sep 2011-19.9267.6270.4962.770.89728 Sep 2011-19.6992.1994.2778.840.99811 Sep 20112.4068.1568.1955.230.971Average4.8864.7864.9648.650.950Average-6.9866.5366.9056.230.943IT_ColDBF26 Jun 2011-29.9167.8274.1252.940.969IT_LavEN L27 Jun 2011-24.6057.5262.5646.130.9718 Jul 2011-23.1546.3451.8041.840.9783 Jul 2011-60.6939.1272.2163.350.98613 Jul 2011-12.9556.8158.2750.160.9349 Jul 2011-35.9057.4367.7358.590.97118 Jul 2011-23.6954.9959.8748.720.97811 Aug 2011-16.5131.2235.3230.060.99811 Aug 2011-10.6763.2364.1250.030.97412 Aug 2011-0.7931.2431.2524.100.99623 Aug 201114.5064.1765.7954.930.94020 Aug 20113.5931.3231.5321.850.97511 Sep 201140.8553.9667.6747.630.89921 Aug 201123.6929.0137.4632.130.98915 Sep 201138.9559.5271.1352.790.96924 Aug 201147.4525.9954.1047.450.99016 Sep 201118.8470.2372.7150.390.9999 Sep 201133.7146.8357.7049.080.97917 Sep 201144.5454.4670.3647.230.92030 Sep 201158.8478.6698.2678.020.954Average4.6168.0368.1951.160.956Average-9.7055.0155.8644.020.981IT_MboGRA10 Apr 2011-45.4954.3470.8747.710.979FR_PueEBF6 Apr 2011-48.9148.8969.1552.630.97810 May 2011-22.0541.0046.5637.140.9369 Apr 2011-39.0351.2764.4350.030.91325 Jun 2011-11.7021.3924.3818.920.90116 Apr 2011-57.0945.6773.1157.570.9323 Jul 2011-12.3866.2067.3556.630.97817 May 2011-27.9849.2256.6246.950.94624 Aug 201140.6155.8469.0446.810.92528 May 2011-38.3648.1461.5550.920.96125 Aug 201141.2261.0473.6650.970.97819 Jun 2011-58.1049.4176.2764.970.94713 Sep 2011-23.8680.9584.3978.380.9638 Jul 2011-27.6238.4147.3137.660.97521 Sep 2011-21.1275.1978.1069.160.91026 Sep 201149.9044.9667.1749.900.96326 Sep 2011-3.4467.2967.3859.950.91214 Sep 201160.0948.5877.2760.090.97830 Sep 2011-5.0549.5549.8143.630.97820 Sep 201147.7162.8578.9151.510.938Average-6.3365.0765.3850.930.946Average-15.9966.6068.4952.470.953ES_AguSHR07 Apr 2011-49.4223.1154.5549.420.97827 Apr 2011-62.8726.1468.0962.870.9638 May 2011-41.1119.6745.5841.110.97414 May 2011-14.8734.1737.2633.380.95423 May 2011-24.0124.7934.5131.380.96013 Jul 201127.9526.7838.7132.170.98029 Jul 201152.8664.5283.4068.430.97914 Aug 201155.6850.2174.9767.510.96826 Aug 201159.1152.3078.9270.460.9897 Sep 201141.8148.7964.2559.210.972Average15.0260.9262.7553.400.972ALL SITESAVERAGE-2.0763.8564.6550.980.96
An overview of LE simulation accurancy.
SitePFTDayStatistical Test SitePFTDayStatistical Test BiasScatterRMSDMAENASHBiasScatterRMSDMAENASHES_LjuOLI14 Apr 201113.1043.6945.6234.000.987IT_Ro3CRO9 Apr 2011-34.8854.1964.4539.690.9969 May 2011-8.4837.5738.5126.450.99311 Apr 2011-39.3543.0258.3041.490.99724 Jun 201142.6262.2275.4263.340.97718 Apr 2011-17.4721.9028.0220.970.99827 Jun 201146.9859.1575.5360.960.96821 Apr 20111.6527.6927.7420.700.99819 Jul 201117.7825.0330.7023.020.95420 Jun 201151.8554.1574.9755.860.95428 Jul 201126.3523.8835.5730.000.96126 Jun 201138.3331.8249.8139.170.9604 Aug 2011-13.9724.0927.8521.570.96624 Aug 201112.1528.2930.7922.730.98422 Aug 2011-3.4038.7738.9228.530.98728 Aug 201118.0526.5132.0723.960.97325 Aug 201122.9733.4340.5629.310.9029 Sep 201146.9345.1765.1447.730.97228 Sep 201122.0028.7636.2126.910.90311 Sep 201149.0954.1373.0751.670.986Average21.0951.4955.6437.220.983Average-0.8768.4868.4847.510.982IT_ColDBF26 Jun 201126.5330.7240.5930.210.915IT_LavEN L27 Jun 2011-9.0938.5439.5929.720.9388 Jul 20112.3471.2071.2451.700.9363 Jul 201123.4041.8847.9738.470.97313 Jul 201133.3353.2362.8147.750.9769 Jul 2011-16.3955.2857.6641.600.91218 Jul 201135.8570.0778.7162.730.93511 Aug 201132.4744.8455.3641.660.89911 Aug 201132.4668.3175.6365.570.89412 Aug 201129.7067.4373.6859.100.93723 Aug 2011-25.3481.1585.0150.980.90020 Aug 201131.4880.5286.4563.160.93611 Sep 201156.1042.2670.2356.100.98621 Aug 2011-12.1345.4447.0433.460.93815 Sep 201160.6949.4278.2761.470.98424 Aug 2011-21.8757.0661.1146.970.98916 Sep 201150.2547.7269.3053.450.9879 Sep 201127.1869.2274.3759.710.93517 Sep 20116.7426.5127.3521.590.99330 Sep 20119.7840.2755.6948.690.913Average33.6767.4375.3655.860.951Average8.4758.3258.9341.390.937IT_MboGRA10 Apr 201116.8525.3930.4721.850.989FR_PueEBF6 Apr 201152.8557.2477.9156.050.98010 May 2011-35.3542.7255.4540.520.9139 Apr 2011-17.4439.3943.0825.790.99625 Jun 20116.8759.9360.3349.330.97616 Apr 201143.7641.6760.4345.930.9773 Jul 2011-26.5173.7578.3756.200.91117 May 201145.0059.7374.7856.060.99024 Aug 2011-19.2951.7955.2737.790.97828 May 201146.2561.5576.9955.460.98525 Aug 201126.8568.1573.2561.210.93619 Jun 201128.6443.4152.0139.130.99313 Sep 2011-8.0944.2044.9336.710.9988 Jul 201122.0538.5244.3833.470.98321 Sep 201114.9353.3455.3934.190.93626 Sep 201149.0444.6066.2850.750.98526 Sep 201114.5252.1254.1039.330.97814 Sep 201162.2839.9774.0062.280.95430 Sep 201126.2137.6545.8833.520.98020 Sep 201111.5419.5622.7118.020.987Average-3.4574.5874.6652.870.959Average37.5657.7768.9147.460.988ES_AguSHR7 Apr 2011-20.7630.0936.5525.020.99027 Apr 2011-21.8629.0336.3428.040.9948 May 2011-9.6821.1223.2316.540.99614 May 20119.0520.1422.0817.510.99023 May 201110.8425.1027.3519.640.98613 Jul 201127.0128.6339.3631.060.88429 Jul 201134.4725.9443.1434.810.75414 Aug 201125.4224.4235.2528.310.94726 Aug 201128.0052.6159.6040.410.9757 Sep 201136.6537.9652.7639.470.953Average13.9934.5337.2525.580.947ALL SITESAVERAGE15.7858.9462.7543.980.964
An overview of H simulation accurancy.
SitePFTDayStatistical Test SitePFTDayStatistical Test BiasScatterRMSDMAENASHBiasScatterRMSDMAENASHES_LjuOLI14 Apr 2011-29.2444.7553.4539.510.985IT_Ro3CRO9 Apr 201110.9239.8041.2726.920.9349 May 2011-11.7632.5734.6330.290.96311 Apr 201131.6730.2443.7934.750.91924 Jun 2011-47.0739.1161.2048.540.94518 Apr 201142.1042.3459.7144.000.95827 Jun 2011-28.8138.9848.4737.580.94821 Apr 201133.3552.2862.0142.530.96119 Jul 2011-27.4638.7447.4835.770.97820 Jun 2011-9.5773.2973.9152.420.95828 Jul 2011-43.8750.4866.8851.270.91526 Jun 201117.2589.4291.0770.440.9834 Aug 201118.9538.4242.8431.950.93424 Aug 201116.3043.6246.5636.970.91722 Aug 2011-3.3951.1451.2539.750.96428 Aug 2011-17.2948.3251.3230.110.91325 Aug 201117.2152.0854.8544.130.9649 Sep 2011-15.8939.2342.3228.030.97828 Sep 201113.2341.6043.6529.290.97811 Sep 2011-22.6161.4565.4844.200.928Average-17.1760.2262.6243.970.957Average15.5370.2371.9347.950.945IT_ColDBF26 Jun 20111.7446.7746.8033.260.899IT_LavEN L27 Jun 2011-22.7068.7572.4051.930.9688 Jul 201118.1364.7867.2751.570.9243 Jul 2011-35.9764.9074.2054.320.97413 Jul 20119.7744.4945.5541.510.9709 Jul 2011-25.3548.4954.7240.300.91318 Jul 201112.2957.2058.5051.310.94111 Aug 20115.6541.0441.4232.010.97811 Aug 2011-3.4037.5137.6629.440.99112 Aug 20110.3232.8532.8525.040.96323 Aug 201155.4953.0176.7460.690.99720 Aug 20117.7756.6757.2038.050.91811 Sep 201132.1637.2049.1736.640.96921 Aug 20119.1151.0951.9038.970.97815 Sep 201121.1873.9076.8862.740.87924 Aug 201118.9356.4659.5546.520.89916 Sep 201123.2043.5049.3041.640.9699 Sep 20113.3471.6371.7155.630.91017 Sep 2011-0.5159.6959.6945.190.91430 Sep 201141.4341.0458.3143.600.989Average14.7258.7860.5946.840.945Average-6.7256.9557.3439.180.949IT_MboGRA10 Apr 2011-29.7451.9359.8448.150.910FR_PueEBF6 Apr 2011-36.4536.9351.8938.720.97810 May 20110.2920.0320.0316.500.9719 Apr 2011-4.7361.8562.0346.980.99525 Jun 20114.9732.8633.2325.140.89616 Apr 2011-42.2250.0065.4449.120.9143 Jul 201115.8267.8069.6242.000.94117 May 2011-50.6649.1070.5553.690.96824 Aug 201136.0622.4642.4837.550.87928 May 2011-4.1860.9061.0449.300.97825 Aug 201132.1122.4939.2032.690.98619 Jun 2011-37.8559.7070.6964.090.92513 Sep 201115.1526.7330.7322.440.9768 Jul 2011-14.5840.3742.9335.780.94621 Sep 201131.5724.5039.9632.220.93626 Sep 201111.5731.3133.3826.110.91726 Sep 201116.4813.2421.1417.150.91414 Sep 201123.0742.1148.0138.770.91330 Sep 201141.4341.0458.3143.600.98920 Sep 2011-6.8628.5529.3620.380.979Average16.4140.9744.1331.740.940Average-16.2952.9855.4342.290.951ES_AguSHR7 Apr 2011-1.0930.3030.3225.050.99127 Apr 2011-17.0724.5329.8924.170.9308 May 2011-8.2929.7230.8522.230.97814 May 2011-10.7624.7727.0022.460.91523 May 2011-30.7533.2945.3233.510.99713 Jul 2011-27.7833.1443.2431.190.93729 Jul 2011-4.4137.5837.8428.450.91414 Aug 201120.6835.5841.1631.220.98926 Aug 20118.1947.5248.2234.040.9377 Sep 20110.0730.0230.0222.990.993Average-7.0134.8035.5025.030.958ALL SITESAVERAGE-0.0853.5655.3639.570.95
An overview of Tair1.3m simulation accurancy.
SitePFTDayStatistical Test SitePFTDayStatistical Test BiasScatterRMSDMAENASHBiasScatterRMSDMAENASHES_LjuOLI14 Apr 20110.752.933.032.560.330IT_Ro3CRO9 Apr 20112.192.813.563.130.8879 May 20113.872.584.653.870.63111 Apr 20110.053.243.242.850.94424 Jun 2011-2.041.922.802.13-0.44818 Apr 20112.242.913.672.820.90927 Jun 20111.993.924.403.86-1.46021 Apr 20111.042.742.932.390.93819 Jul 20112.643.144.113.350.61220 Jun 20110.494.894.914.060.90328 Jul 20115.452.596.035.450.21526 Jun 20113.703.445.063.820.8534 Aug 20113.613.555.064.53-0.97124 Aug 2011n/an/an/an/an/a22 Aug 20113.352.764.343.610.69528 Aug 2011n/an/an/an/an/a25 Aug 20115.313.946.615.680.0499 Sep 2011n/an/an/an/an/a28 Sep 20113.594.956.125.49-0.19811 Sep 2011n/an/an/an/an/aAverage2.753.954.824.02-0.054Average1.012.993.151.970.905IT_ColDBF26 Jun 20115.292.335.785.310.493IT_LavEN L27 Jun 20112.191.802.832.440.3598 Jul 20111.2167.097.132.420.7573 Jul 20110.541.201.321.140.85513 Jul 20116.011.746.266.010.3969 Jul 20112.783.094.163.64-0.60718 Jul 20112.832.083.513.120.76611 Aug 20112.812.843.994.00-0.01911 Aug 20113.982.924.944.020.80612 Aug 20110.022.062.061.790.59423 Aug 2011-1.352.052.462.060.90420 Aug 20110.642.532.612.180.46911 Sep 20115.351.715.625.350.74021 Aug 20111.542.462.902.590.35315 Sep 20111.251.672.091.610.92924 Aug 20111.782.763.282.670.23616 Sep 20110.241.741.751.400.9449 Sep 20114.473.965.975.30-0.07017 Sep 20111.582.122.652.160.91530 Sep 20112.702.013.212.970.871Average3.493.745.124.080.765Average1.682.843.302.510.304IT_MboGRA10 Apr 20113.310.993.463.310.177FR_PueEBF6 Apr 20115.831.696.075.830.66210 May 20111.402.472.841.980.6699 Apr 20112.263.584.233.940.79425 Jun 20111.030.911.381.260.84516 Apr 20112.361.102.602.360.8323 Jul 20114.811.445.024.810.32017 May 20111.681.051.981.780.86624 Aug 20112.551.212.822.550.60028 May 20115.211.935.565.210.55425 Aug 20112.183.624.223.800.42519 Jun 20113.491.053.653.490.35513 Sep 20114.210.964.324.210.4658 Jul 20112.790.892.932.790.76621 Sep 20110.981.581.861.270.88314 Sep 20113.332.464.143.330.74726 Sep 20112.311.842.952.350.73920 Sep 2011-1.672.462.972.690.79630 Sep 20112.011.182.332.030.76426 Sep 20111.962.252.992.150.883Average3.343.184.613.460.589Average3.133.074.383.760.725ES_AguSHR7 Apr 20111.333.804.023.620.61027 Apr 20110.022.592.592.130.8038 May 2011-0.752.352.472.100.82114 May 20111.172.282.562.090.84423 May 2011-0.211.851.861.480.87013 Jul 20111.944.214.633.760.72229 Jul 20111.463.463.753.190.58314 Aug 20110.383.753.773.170.87126 Aug 20111.944.214.633.760.7227 Sep 20113.072.794.153.490.493Average0.723.523.592.720.734All SitesAverage2.303.334.143.220.567
An overview of Tair50m simulation accurancy.
SitePFTDayStatistical Test SitePFTDayStatistical Test BiasScatterRMSDMAENASHBiasScatterRMSDMAENASHES_LjuOLI14 Apr 20110.841.561.771.560.591IT_Ro3CRO9 Apr 20111.563.483.813.490.8749 May 20110.723.773.843.350.45711 Apr 20110.054.734.734.280.91624 Jun 20111.013.403.552.800.89318 Apr 20112.554.355.043.980.87127 Jun 20111.144.694.824.40-1.80421 Apr 20110.694.364.413.900.89919 Jul 20110.304.704.714.040.84620 Jun 20110.494.894.914.060.90328 Jul 20113.312.624.223.460.50126 Jun 2011-2.102.983.642.990.8294 Aug 20112.243.374.043.43-0.49524 Aug 2011n/an/an/an/an/a22 Aug 20111.954.625.024.120.83828 Aug 2011n/an/an/an/an/a25 Aug 20110.603.974.023.420.4279 Sep 2011n/an/an/an/an/a28 Sep 20112.724.655.394.71-0.02811 Sep 2011n/an/an/an/an/aAverage1.483.984.253.530.223Average0.724.424.482.890.882IT_ColDBF26 Jun 20114.292.334.8928.450.583IT_LavEN L27 Jun 20112.341.913.022.410.3658 Jul 20110.903.013.142.630.7973 Jul 20110.690.811.060.820.89513 Jul 20110.562.002.081.550.8459 Jul 20113.352.184.003.38-0.49418 Jul 20112.283.003.763.220.75911 Aug 20113.272.664.223.44-0.03011 Aug 20113.193.855.003.510.83112 Aug 20110.101.971.971.670.62223 Aug 2011-1.313.443.683.350.84320 Aug 20111.322.122.501.830.55411 Sep 20110.652.802.882.490.87921 Aug 20111.011.812.071.430.64415 Sep 20110.832.612.732.350.89724 Aug 20111.362.432.792.140.38716 Sep 2011-0.123.013.022.830.8869 Sep 20113.934.055.644.850.02117 Sep 20111.313.353.603.160.87630 Sep 20112.732.973.212.780.789Average1.263.363.592.950.820Average1.742.613.132.200.375IT_MboGRA10 Apr 20112.991.143.202.990.257FR_PueEBF6 Apr 20116.102.226.496.100.64610 May 20110.472.652.692.330.6129 Apr 20112.784.014.883.930.79525 Jun 20112.461.242.762.460.69516 Apr 20111.212.072.391.730.8773 Jul 20113.861.594.173.860.45417 May 20110.481.421.501.250.90624 Aug 20112.021.812.712.090.67328 May 20114.961.165.104.960.57525 Aug 20111.171.411.831.540.76719 Jun 20111.800.691.931.800.66713 Sep 20113.471.453.763.470.5598 Jul 20111.271.572.021.650.86121 Sep 20110.071.881.881.560.85714 Sep 20111.072.732.942.320.85126 Sep 20111.582.282.782.230.75220 Sep 20112.443.424.202.980.77430 Sep 20111.131.611.971.550.82026 Sep 20112.443.424.202.980.774Average1.922.132.872.410.644Average2.663.154.123.070.773ES_AguSHR7 Apr 20110.212.792.202.490.89127 Apr 2011-0.653.003.072.760.7448 May 2011-0.983.333.472.900.75414 May 20110.382.872.892.390.82223 May 2011-1.022.522.712.440.78513 Jul 2011-0.301.961.981.520.97229 Jul 20111.133.723.883.160.58714 Aug 2011-1.304.674.854.490.81726 Aug 20110.744.564.623.860.7147 Sep 20112.282.913.692.810.593Average0.033.393.392.630.768All SitesAverage1.403.293.692.810.641
The three facets of SimSphere Architecture.
Overall methodology of SimSphere
validation followed in this study.
Scatterplot comparison of SimSphere
predicted and in-situ Rnet flux.
Scatterplot comparison of SimSphere
predicted and in-situ LE flux.
Scatterplot comparison of SimSphere
predicted and in-situ H flux.
Scatterplot comparison of SimSphere
predicted and in-situ Tair1.3m.
Scatterplot comparison of SimSphere
predicted and in-situ Tair50m.