Introduction
In recent years distributed temperature sensing (DTS)
technology has quickly improved . The precision and spatial
resolution now allow its widespread use in hydrological and atmospheric
sciences , from measuring groundwater flow
and seepage into streams to soil
moisture , soil heat flux , and wind
speed . First introduced by , DTS can also
be used for measuring the Bowen ratio, to estimate the evaporation flux. A
dry and wet stretch of the same fibre optic cable are installed vertically to
obtain the so-called dry- and wet-bulb temperature gradient, respectively.
This method mitigates some problems of the conventional Bowen ratio, since
usually at least two different sensors are used to measure the temperature
and vapour pressure gradients, of which each has its own independent error
. The DTS-based Bowen ratio does not suffer from
this drawback, by having a large amount of data points over the height (up to
8 per metre) with only a single sensor. It also has a resolution of 0.06 K
for 1 min averages (Silixa machine calibration), and will be more accurate
when measuring over a longer time period, allowing for very small temperature
gradients to be measured.
In addition to estimating the latent and sensible heat flux, the measurements
can also be used to get a better understanding of the processes taking place
in complex ecosystems, such as forests. A vertical temperature and humidity
profile is available in high resolution and precision, both above, inside,
and under the canopy. DTS can also estimate different components of the
energy balance, such as the heat storage in the air column, and the soil heat
flux . Finally, it can be used to increase our
understanding of the energy exchange between the canopy and undergrowth
layers by looking at the air temperature gradient under the canopy.
This paper elaborates on the method of , by considering more
energy balance components like the latent and sensible heat storage in the
air column, including a data-quality system, and using the potential air
temperature. The performance of the method is tested in a mixed forest in the
Netherlands by looking at the accuracy of the DTS-measured air temperature
and wet-bulb temperature, compared to reference temperature and humidity
sensors. It appears that solar radiation can have a significant influence on
the cable temperature, which can be mitigated by providing artificial shadow.
Lastly, the fluxes resulting from the method are compared to an eddy
covariance (EC) system, and the sources of differences between the methods
are shown.
Forest type distribution within 500 m of the tower site at Speulderbos Forest, the Netherlands.
Materials and methods
Theory
The Bowen ratio energy balance method (BREB) combines the energy balance with
the Bowen ratio . The energy balance can be described by
RN+A=ρλE+H+GS+dQdt,
where RN is the net radiation (W m-2), ρλE
the latent heat flux (W m-2), H the sensible heat flux (W m-2),
GS the soil heat flux (W m-2),
and dQdt is the change of energy storage in the
system (W m-2). A represents a net advection of energy
into the system (W m-2), but is assumed to be 0. The energy
flux associated with photosynthesis (GP) was not measured, and is
therefore not included in the equation. The Bowen ratio (β) is the
ratio of the sensible heat flux to the latent heat flux and can be
approximated using the air temperature gradient and the vapour pressure
difference over the height
β=HρλE≈γΔTaΔea,
where γ is the psychrometric constant (kPa K-1) (see
Eq. ), ΔTa the difference in air temperature between two
heights (K) and Δea the difference in actual vapour pressure between
the two heights (kPa). However, when gradients are very small, the adiabatic
lapse rate cannot be neglected . Therefore the potential
temperature should be used instead:
β=HρλE=cpλ∂Θ/∂z∂q/∂z=γ∂Θ/∂z∂ea/∂z,
where cp is the specific heat of air (MJ kg-1) (see Eq. ),
λ the latent heat of vaporization (2.45 MJ kg-1 K-1), Θ the potential
temperature (K), q the specific humidity (kg kg-1) (see
Eq. ) and z the height above the ground (m). The
potential temperature gradient can be approximated by the right-hand side of
Eq. (), as the ratio ΘTa is
nearly 1 :
∂Θ∂z=ΘTa(∂Ta∂z+Γ)≈∂Ta∂z+Γ,
where Ta is the air temperature (K), and Γ is the adiabatic lapse
rate (typically around 0.01 K m-1). The numerical
implementations of Eqs. () and () are explained in
Sect. . Under dry and unsaturated
conditions the lapse rate is equal to
Γ=gcp,
where g is the gravitational acceleration (9.81 m s-2).
The specific heat capacity of air is determined by
cp=1.004+1.84q
and the specific humidity by
q=εeaP,
where ε is the ratio of molecular mass of water vapour to dry air
(0.622), and P the atmospheric pressure (kPa). The actual vapour pressure
is determined by
ea(Ta)=es(Tw)-γ(Ta-Tw),
where Tw is the wet-bulb temperature (K), and es the saturation vapour
pressure (kPa) given by
es(Tw)=0.61⋅exp(19.9⋅Tw273+Tw).
The psychrometer constant is related to the air pressure and ventilation of
the psychrometer . If sufficiently ventilated,
the psychrometric constant is defined by
γ=cpPελ=0.665×10-3⋅P.
As the air pressure also varies over height, the measurements have to be
corrected for elevation using the following approximation p. 8:
P(z)=P0⋅exp(-z/7290)
with P0 being the pressure at sea level (kPa). By combining the Bowen
ratio (Eq. 3) with the energy balance (Eq. 1),
the latent heat flux and sensible heat flux can be
determined:
H=RN-GS-dQdt1+1β,ρλE=RN-GS-dQdt1+β.
The storage component in the energy balance has multiple parts, ranging from
the storage of heat in the soil, to the storage of heat in the form of water
vapour in the air column:
dQdt=dQHdt+dQEdt.
The changes in storage of heat and water vapour in the air column below the
height at which the energy fluxes (RN, H and ρλE) are
measured are represented by dQHdt and
dQEdt respectively (W m-2).
The change in biomass heat storage (dQBdt) was
not measured, and is therefore not included in this equation.
dQHdt and dQEdt
are defined as
dQHdt=∫0zρacpdTadtdz,dQEdt=∫0zρaλdqdtdz.
Schematic overview of the measurement setup at the tower.
Study site
The measurements were carried out at the Speulderbos mixed forest
(52∘15′4′′N, 5∘41′24′′E), on a tower located within a patch
of Douglas fir trees (Pseudotsuga menziesii (Mirb.) Franco) of
2.5 ha in Garderen, the Netherlands (Fig. ). The surrounding area is
characterized by the presence of broadleaved and coniferous tree species,
distributed in blocks around the tower site . Within a 500 m radius
it is possible to find native tree species such as beech (Fagus sylvatica L.),
pedunculate oak (Quercus robur L.) and Scots pine (Pinus sylvestris L.),
as well as the introduced species hemlock (Tsuga heterophylla (Rafinesque)
Sargent) and Japanese larch (Larix kaempferi (Lambert) Carriére)
. Canopy heights differ between cover types depending on species
and growing stage. Some coniferous canopies like the Douglas fir have a canopy height
between 26 and 30 m, while the broadleaved stands can reach up to 30 m height for old-growth beech trees , or heights under 10 m for smaller pedunculate oak trees.
The study site has an oceanic climate (Cfb) under the Köpen classification
system, with a yearly average temperature of 9.8 ∘C and an average
precipitation of 910 mm yr-1 . The
topography is slightly undulating with smooth height differences , a
well-drained soil, and a groundwater table below 40 m depth . The
soil texture ranges from fine sand to sandy loam .
Setup
The temperature of fibre optic cables is measured using the DTS technique
. In the setup, two cables with different diameters were
used. The first cable has a diameter of 6 mm and has both a dry and a wetted
stretch. To wet the cable it was wrapped in cloth, and water was supplied to
it continuously. A second cable with a diameter of 3 mm was used to study the
effects of solar radiation, as a thinner cable will warm up less
. However, this method added additional uncertainties due
to the required extrapolation and the 3 mm cable was not used in this study.
(While correlation with reference sensors improved, the uncertainty of
extrapolation caused extra noise in the Bowen ratio calculations). Both
cables were connected to the same DTS machine (in single-ended mode) and
calibrated in a calibration bath (see Fig. )
The DTS machine used was the Silixa Ultima , which has a
sampling resolution of 12.5 cm, measurement resolution of 35 cm, and a
measurement standard deviation of 0.06 K at a 1 min time resolution.
The fibre optic cable with a diameter of 6 mm was secured at the top of the
tower, with the dry stretch hanging 1.2 m away from the tower, and the wet
stretch 0.25 m away. The cable with a diameter of 3 mm was secured next to
the dry 6 mm cable. The response times of the cables are of the order of 2–3 min
for the 6 mm cable, and 20 to 40 s for the 3 mm cable. The
cables were secured at multiple locations distributed over the height (in and
above the canopy; see Fig. ), using loops (with a
diameter of 5 cm) to prevent direct contact with the support structure. For
both cables a stretch of 10 m at both the start and end was placed in a
calibration bath, an enclosed styrofoam box filled with water, along with two
Pt100 temperature probes that were connected to the DTS machine. An air
bubbler was installed in the styrofoam box to ensure a homogeneous
temperature distribution. The cables were shielded from direct solar
radiation using screen gauze secured onto PVC rings; see Fig. .
Only the southern 180∘ of the cables was
shielded, to allow for sufficient ventilation. The screen gauze had holes 1.5 mm
wide, and the mesh material had a diameter of 0.3 mm. Two layers of the
gauze were used. Each segment of shield was 2 m long, and was secured to the
tower with a horizontal beam. Due to the angle of the incident sunlight the
gauze was able to block most direct sunlight, except during the early
morning. To supply the wet cable with water, a reservoir was installed near
the top of the tower, along with a pump. The pump speed was set to 1500 mL h-1
during sunny days without rainfall, and to 800 mL h-1 on other days, which was enough to keep the cable wet
over the entire height, while keeping the influence of relatively warm water
at the top of the cable at a minimum. As water supplied at the top has a
higher temperature than the wet-bulb temperature, the top 2 m of wet
cable data was excluded from the data analysis to allow the slowly flowing
water to reach the wet-bulb temperature.
A net radiometer (Kipp & Zonen CNR4) was located on the top of the tower (48 m),
measuring both incoming and outgoing short- and longwave radiation. One-minute averages were logged. On the tower six humidity and temperature
sensors were located over the height, at 4, 16, 24, 32, 36 and 46 m a.g.l.
(above ground level). The lower four were Rotronic HC2-S3C03 sensors (with
active ventilation), and the top two were Campbell CS216 sensors with passive
ventilation. The sensors were inter-calibrated to the sensor at 24 m.
The temperature and humidity was logged at 1 min averages.
At the top of the tower an EC system was installed to measure
the sensible and latent heat fluxes. It consisted of a Campbell CSAT3 sonic
anemometer and a LI-COR Biosciences LI7500 gas analyser connected to a CR5000
Campbell data logger, to which the data were logged at 20 Hz.
Schematic of one 2 m segment of the solar screen construction.
Two cup anemometers (Onset S-WSB-M003) were used to measure the wind speed,
one at the top of the tower (48 m), and one below the canopy (4 m). The data
from the lower anemometer lacks the resolution to properly measure the low
wind speeds below the canopy, which are at times too low to be registered.
One-minute average wind speeds, along with the maximum gust speeds, were
logged.
The biomass heat storage change and the photosynthesis energy flux were not
measured. The biomass heat storage change is estimated to have a maximum of
45 W m-2, and the photosynthesis energy flux is estimated
to be of the order of 5 W m-2 . For the soil heat flux, the soil temperature was measured at
different depths (1, 3, 4, 8, 20, 50 cm). Soil moisture was measured using
Campbell Sci. Inc. CS616 water content reflectometers. Thermal conductivity
was fitted to soil heat flux measurements done at 8 cm. The soil heat flux
was then determined using the harmonics method .
Data processing
The DTS machine was set to measure the cable temperature at 1 min
averaging intervals. For the comparison with reference temperature sensors,
these 1 min resolution data are used. To compare the wet-bulb temperature
measured by the fibre optic cable to the reference sensors, the reference
wet-bulb temperature is iteratively derived from the reference air
temperature and relative humidity. For the purpose of calculating the Bowen
ratio, the temperature and actual vapour pressure are averaged over time for
15 min time periods. For DTS Bowen ratio calculations, the temperatures
between 38.5 and 44 m are used. This area is shaded from the sun by the
screen gauze, and at the top of the stretch the new water on the wet cable
has reached the wet-bulb temperature.
When calculating the gradients for the Bowen ratio, the 15 min average
temperature and vapour pressure are fit to the natural logarithm of the
height, in the following form:
Ta,fit=a⋅ln(z)+b.
A logarithmic shape of the profiles was assumed based on Monin–Obukhov
similarity theory. A linear fit was also looked at, but it resulted in a
minimal difference in the resulting fit. From the fits the temperature
difference over height is then calculated:
∂Θ∂z≈∂Ta∂z+Γ(z)≈ΔTa,fitΔz+Γ(z)≈Ta,fit(z=44)-Ta,fit(z=38.5)44-38.5+Γ(z¯=41.25),∂ea∂z≈Δea,fitΔz=ea,fit(z=44)-ea,fit(z=38.5)44-38.5,
where ΔTa,fit is the difference in air temperature (K) of the
fitted temperature curve, between the top and bottom of the height range used
for the Bowen ratio. Δea,fit is the difference in vapour pressure
(kPa) of the fitted vapour pressure curve between those heights. Δz
is the difference in height (m). The coefficients of determination of the
regressions of the temperature and vapour pressure, rTa,z and
rea,z, can be used for determining the goodness of fit. A high
(positive or negative) regression means that the logarithmic slope (of the
15 min average) is very well defined.
To calculate the air column storage terms dQHdt
and dQEdt (Eqs. and ), the
DTS-measured temperature and vapour pressure are used, except for the centre
of the canopy where DTS data are not accurate due to the sunlight and lack of
screens in the canopy. The temperature and specific humidity are integrated
over the height from 0 to 41 m, up to the height of the Bowen ratio
measurements.
As quality control scheme for the DTS-measured Bowen ratio, two flags are used. The
first flag tests the correlation coefficient of the actual vapour pressure
over height, for which we chose a lower limit of 0.20 (Eq. ). We
do not consider rTa,z of the air temperature gradient as it is always
higher than rea,z (as the uncertainty in ea is higher due to the
propagation of errors in Ta and Tw). The second flag is for the case
where the Bowen ratio approaches -1, which causes the uncertainty in the BREB
fluxes to be very high, as the denominator of Eqs. () and () approaches 0 :
Flag1:rea,z2>0.20,Flag2:β<-1.1orβ>-0.9.
If flag 1 is true, the outcome of the Bowen ratio calculation is considered
reliable. The other data points are removed from further analysis. If flag 2
is also true, then the Bowen ratio can be used for calculating the
atmospheric heat fluxes.
After processing the EC data using LI-COR's
EddyPro® software ,
several quality flags are available. The quality flag system used is from
, ranging from 0 (best) to 2 (worst). The EC
fluxes with a quality flag of 0 or 1 are used in this research.
To summarize, the method of this paper differs in a few points from
. The fit of the Bowen ratio temperature and vapour pressure
profiles is done separately, to get the correct ratio, as ∂T∂z/∂ea∂z≠∂T∂ea. More energy balance storage terms are taken into account,
namely the latent and specific heat storage in the air column. The potential
temperature is used instead of the air temperature, to correct for the lapse
rate. The local air pressure is taken into account in the calculations, as it
has an influence on the psychrometric constant, specific heat capacity and
specific humidity. Lastly, a system for simple quality flags is introduced to
allow for simple objective quality control.
Meteorological conditions during the days that both DTS
and EC data were available. From top to bottom: wind
speed at the top of the tower, wind speed at the bottom of the tower,
wind direction at the top of the tower, and the measured energy fluxes
(green: net radiation; red: soil heat flux; black: energy storage change
dQdt).
Results and discussion
Meteorological conditions
For the comparison of the DTS temperature with the reference temperature data
(Sect. ), the days 10–22 August 2016 are used.
For a good comparison between DTS and EC, both devices should work properly.
Due to several technical problems with data collection, only 11 days within
the measurement campaign have both EC and DTS data available,
namely 10, 12–14, 19–22, and 28–30 August 2016. On the other days data are
missing in either the EC or the DTS. The meteorological
conditions of these days are shown in Fig. . All days were
partially clouded, or completely clouded. The wind direction was mainly west
and northeast. Above the canopy the wind speed varied between 2 and 6 m s-1,
while under the canopy the wind speed was often too
low to be measured with the cup anemometer (under 0.4 m s-1).
Temperature validation
In Fig. the comparison between the 6 mm DTS
cable and the reference sensor is shown. For the above-canopy comparison, the
46 m reference sensor is compared to the cable temperatures at 44 m height,
as the temperatures at the top are unreliable due to influence from the sun
and the warm water from the reservoir. Below the canopy the dry cable
temperature correlates perfectly with the reference sensor temperature
(Fig. e). In and above the canopy incoming solar
radiation warms up the fibre optic cable
(Fig. a, c), which causes an error at 34 m, where no screen
was installed. This error is a deviation of up to 3 K from the reference
sensor temperature (for 1 min temperature averages). The comparison at 34 m
also has an offset – this is a constant error of about 1 K, due to the
reference temperature sensor drift and inter-calibration problems. The
addition of screens above the canopy largely reduces the error from solar
radiation to under 1 K, leading to a very good agreement between the two
sensor types (Fig. a).
Below the canopy the wet cable temperature is in good agreement with the
reference wet-bulb temperature (Fig. f), even
though wind speeds were often low. This shows that the wet cable gives a good
estimate of the wet-bulb temperature. At 34 m, where no screens were placed,
the error in the wet-bulb temperature is larger than the error in the air
temperature. Deviations of up to 4 K occur in the measurement period. The
shielded top part of the wet cable performs much better
(Fig. b), and errors are small (under 1 K).
Comparison between the 6 mm DTS cable and reference
temperatures. Grey line shows 1:1 correlation. Data from 10–23 August 2016. (a) Dry cable at 44 m and reference air temperature at
46 m;
the cable is shielded by the screen. (b) Wet cable at 44 m and
reference wet-bulb temperature at 46 m; the cable is shielded by the
screen. (c) Dry cable and reference air temperature at 34 m; the cable is
exposed to direct sunlight. (d) Wet cable and reference wet-bulb
temperature at 34 m; the fibre optic cable is exposed to direct sunlight. (e) Dry
cable and reference air temperature at 16 m, under the canopy
so less direct sunlight hits the fibre optic cable. (f) Wet cable and
reference wet-bulb temperature at 16 m, under the canopy so less direct
sunlight hits the fibre optic cable. Shown are the linear correlation
coefficients: the coefficient of determination (r2), the slope (s)
and the intercept (i).
Bowen ratio verification
The Bowen ratio resulting from the BR-DTS method (βDTS) is
compared to the EC Bowen ratio (βEC), at a
15 min averaging interval. In Fig. the correlation
between the EC Bowen ratio estimate and the BR-DTS is shown. It
shows a grouping around the 1 : 1 line, and a good correlation (r2=0.59).
The EC Bowen ratio was only calculated for fluxes with an
absolute value larger than 10 W m-2, as the uncertainty of
the EC Bowen ratio is very high when the fluxes are small. Even
the negative (night-time) values seemed to be accurate, since they passed the
quality control flags. However, both EC and BR-DTS have problems
measuring the night-time Bowen ratio. For EC this is due to the
lower friction velocity at night , while for the BR-DTS
method the gradients are very small due to the small fluxes.
Correlation between the DTS-measured (βDTS)
and EC-measured (βEC) Bowen ratios. Daytime
data are between 07:00 and 18:00. Data from 10, 12–14, 19–22, and
28–30 August. R2=0.59. RMSE = 0.81. n=319 data points.
Tukey mean-difference plot comparing BDTS
and BEC. With μ=-3.4 W m-2,
RMSE=76 W m-2, n=741 data points
(15 min averages). Data from 10, 12–14, 19–22, and 28–30 August 2016.
One drawback of the DTS-based Bowen ratio is the assumption that the eddy
diffusivity of heat and water vapour are the same. In reality these eddy
diffusivities can be dissimilar . This can cause an error
(both a bias and extra noise) in the Bowen ratio as measured by the
temperature and vapour pressure gradients compared to the EC
Bowen ratio. Another source of differences between βDTS and
βEC is that the two are measured at different heights.
During the measurement period the 80 % fetch of the EC system was between 200
and 300 m. By applying the findings of , the Bowen ratio
80% equilibrium ratio would be reached at a fetch-to-height ratio of 20 to
40. This corresponds to a distance of 350 to 700 m. The fetch of the Bowen
ratio will therefore not be equal to the EC fetch, which could
cause some differences in measured fluxes.
Energy balance closure
A known problem in measuring fluxes is that the energy balance often does not
close well. This is caused by differences in fetch between the used devices,
device inaccuracies, and possibly problems with the EC method
. Part of the difference between the BR-DTS method and the
EC method may be explained by this energy balance closure
problem. EC measurements have a fetch which does not include
the area close to the flux tower. The available energy in the BR-DTS method
depends on measurements of net radiation, ground heat flux and heat storage
change (dQdt) close to the tower. Heterogeneity
in the fetch may cause differences between the two methods. In addition, the
biomass heat storage change (dQBdt)
was not measured for the BR-DTS method, and was assumed to be 0 W m-2.
The photosynthesis energy flux (GP) was also
assumed to be 0 W m-2.
Tukey mean-difference plot comparing
ρλEEC and ρλEDTS.
With μ=18.7 W m-2, RMSE=90 W m-2
(15 min averages). Data from 10, 12–14, 19–22, and 28–30 August 2016.
Tukey mean-difference plot comparing HEC and
HDTS. With μ=-10.6 W m-2,
RMSE=82 W m-2. (15 min averages).
Data from 10, 12–14, 19–22, and 28–30 August 2016.
Plot comparing the BR-DTS and EC measured sensible (H)
and latent (ρλE) heat fluxes over time (15 min averages).
Data from 10, 12–14, 19–22, and 28–30 August 2016.
To investigate the energy balance closure for the two methods, we summed up
the available fluxes in the following equations, where
dQdt is the storage term from Eq. ():
BDTS=RN-GS-dQdt,BEC=HEC+ρλEEC,
where BDTS is the energy available for heat fluxes in the
BR-DTS method (W m-2) and BEC is the sum of
the EC-measured heat fluxes (W m-2).
To compare the two measurement methods, a Tukey mean-difference (or
Bland–Altman) plot was made (Fig. )
. The mean of the two measurement methods is plotted
against the difference between them. The mean difference (μ) between
BDTS and BEC is a 3.4 W m-2
underestimation by the BR-DTS method. At low fluxes (below 100 W m-2),
the BR-DTS method measures less energy available for
fluxes compared to EC. At high fluxes (over 400 W m-2)
the opposite is visible. One possible reason for this
is that the biomass heat flux
(dQBdt) was not measured, which
causes an underestimation of the available energy in BDTS
during the night, and an overestimation during the day.
Energy fluxes
Figures and show the mean
difference plots comparing the latent and sensible heat fluxes of the EC method to the BR-DTS method. The BR-DTS fluxes are calculated
above the canopy, using only temperature data from the shielded cables. The
Tukey mean-difference plot for the latent heat flux shows no large bias when
comparing the BR-DTS method to EC, with the mean difference
being a 18.7 W m-2 overestimation by the BR-DTS method
(Fig. ).
The Tukey mean-difference plot comparing the sensible heat flux
(Fig. ) shows a strong negative bias for negative fluxes,
resulting from the negative bias in the energy balance comparison
(Fig. ). At positive fluxes there seems to be a positive
bias (HDTS>HEC). The mean difference is small,
being a 10.6 W m-2 underestimation by the BR-DTS method.
Figure shows the time series of the BR-DTS and EC
measured heat fluxes. The daytime flux estimates correspond well, and follow
the same trends. The night-time BR-DTS estimates of the sensible heat flux
are more negative than the EC estimates, one possible reason being the energy
balance differences discussed before. On many days, during the early morning
and start of the evening, the BR-DTS has missing values, which is mainly due
to the inversion of the gradient, as the temperature gradients changes from
negative (stable conditions) to positive (unstable conditions) and vice
versa. This inversion causes uncertainty, which is filtered out by the
quality control flags.
Conclusions and recommendations
This technical note investigates the use of the BR-DTS method above a forest
canopy, and introduces a number of improvements on the method as presented by
. The performance is investigated by comparing the measured
DTS cable temperatures to reference sensors, looking at energy balance
closure, and comparing the measured Bowen ratio, sensible heat flux and
latent heat flux to EC measurements.
When comparing the fibre optic cable temperature to reference sensors, it
can be seen that the wet-bulb and air temperatures can be well represented. Under
the canopy, where the cables are shaded from direct sunlight, the DTS cable
and reference sensors are in near-perfect agreement. However, above the
canopy direct sunlight may cause a large error, up to 3 K. This error can be
largely mitigated by placing screens to block the sunlight, reducing the
error to less than 1 K. Hence screens are effective and should also be placed
in the canopy.
The Bowen ratio measured by DTS correlates well with EC
estimates (r2=0.59). A simple quality control method, using the goodness
of fit of the vapour pressure gradient, also works well, and filters out most
outliers and errors. The small gradients above the forest canopy are hard to
measure accurately, which increases the uncertainty during days where fluxes
(and thus gradients) are small. The Bowen ratio assumption that the eddy
diffusivities of heat and vapour are equal was not studied, but can be a
source of differences between the BR-DTS and EC methods. The
difference in fetch for the two methods can also be a cause for differences.
The energy balance closure between the BR-DTS method and EC is
in good agreement, with the mean difference being a 3.4 W m-2
underestimation by the BR-DTS method, and an
uncertainty of RMSE=76 W m-2. However, the
BR-DTS method estimates a more negative amount of available energy during
night-time, and a more positive amount during daytime compared to EC. One cause could be the lack of biomass heat storage change
measurements, which is of the order of 45 W m-2. Another
source for the difference is that the energy balance components of the BR-DTS
method are generally point measurements, while EC and the Bowen
ratio both have a large fetch. As a result, heterogeneity can cause large
differences in the available energy for latent and sensible heat fluxes.
When comparing the latent heat flux of the two methods, they are in
agreement, although the uncertainty is high (RMSE=90 W m-2).
The BR-DTS method slightly overestimates the latent
heat flux, with a mean difference of 18.7 W m-2. The
results for the sensible heat flux are similar, with an uncertainty of
RMSE=82 W m-2, and the BR-DTS method
underestimating the sensible heat flux by 10.6 W m-2.
However, the underestimation mainly takes place during night-time, which can
be caused by differences in available energy.
While the average profiles can be useful and valuable, extra information
could be gained by opting for a smaller diameter fibre optic cable, and
measuring at a high frequency (1 Hz). This could give new insights into
surface interactions and could show convective cells transporting heat
upwards.
A way to improve the performance of the BR-DTS method is to find an
independent estimate for the sensible heat flux (H), to avoid the
uncertainties in the energy balance components (RN,
dQdt). Through the universal functions of the
Monin–Obukhov similarity theory estimates of the sensible heat flux can be
made. This could be done either by measuring the wind speed over height
using DTS or by applying the
flux-variance method . The Bowen ratio can then be used to
calculate the latent heat flux.