Application of the Flux-Variance Technique for Evapotranspiration Estimates in Three Types of Agricultural Structures

Irrigation of protected crops requires sound knowledge of evapotranspiration. Previous studies have established that the eddycovariance (EC) technique is suitable for whole canopy evapotranspiration measurements in large agricultural screenhouses. Nevertheless, the eddy-covariance technique remains difficult to apply in the farm due to costs, operational complexity, and postprocessing of data, thereby inviting alternative techniques to be developed. +e subject of this paper is the evaluation of a turbulent transport technique, the flux variance (FV), whose instrumentation needs and operational demands are not as elaborate as the EC, to estimate evapotranspiration within large agricultural structures. Measurements were carried out in three types of agricultural structures: (i) a banana plantation in a light-shading (8%) screenhouse (S1), (ii) a pepper crop in an insectproof (50-mesh) screenhouse (S2), and (iii) a tomato crop in a naturally ventilated greenhouse with a plastic roof and 50-mesh screened sidewalls (S3). Quality control analysis of the EC data showed that turbulence development and flow stationarity conditions in the three structures were suitable for flux measurements. However, within the insect-proof screenhouse (below the screen) and the plastic-covered greenhouse, R2 of the energy balance closure was poor; hence, the alternative simple method could not be used. Results showed that the FV technique was suitable for reliable estimates of ET in shading and insect-proof screenhouses with R2 of the regressions between FV latent heat flux and latent heat flux deduced from energy balance closure of 0.99 and 0.92 during validation for S1 and S2, respectively.


Introduction
In recent years, the area of vegetables and orchards grown in protected cultivation systems is constantly increasing.ese include, among other structures, naturally ventilated greenhouses [1], insect-proof screenhouses [2], and shading screenhouses [3].
ese structures are naturally ventilated and hence have significant interaction with the external environment.e advantages and limitations of such protected cultivation systems are well documented in the literature [4,5].
Protected crops are exposed to microclimatic conditions that are significantly different from those in the open field.Hence, the interaction between protected crops and their microenvironment has been the topic of much research during the past years (e.g., [6][7][8]).One effect of covering the crops is in modifying the exchange of energy, mass, and momentum between the plants and their environment.is modification may affect the evapotranspiration; that is, the water vapor flux from the canopy to the atmosphere, which, in turn, will affect the irrigation demands.e possibility of water saving through reduced transpiration and irrigation demands initiated a number of research studies focused on evapotranspiration measurements and estimates, mainly in screenhouses [3,[9][10][11][12][13].
e most common method for direct measurements of evapotranspiration and other scalar fluxes is the eddy covariance [14].
e method was originally developed and mostly used for flux measurements over open surfaces like forests, natural, or agricultural fields and open water bodies.Due to its high capabilities in reliable measurements of whole canopy evapotranspiration, in recent years, its performance was also examined in protected environments like screenhouses.
Results obtained in various screenhouses [2,3,9,12,13] illustrated the reliability of the eddy-covariance technique within such protected environments.In all these studies, the EC system was deployed above the plants and below the screen, at a height which is smaller than twice the canopy height.Although conditions at such a height apparently do not meet the common requirements for flux measurements [15], turbulence analysis and flux results supported the use of the EC method at such heights within screenhouses.For example, Tanny et al. [3] evaluated the suitability of the eddy-covariance technique to directly measure evapotranspiration in a large banana screenhouse with almost mature plants.Results were promising: they found 94% closure of the energy balance and daily evapotranspiration values, in agreement with the irrigation applied by the grower.Even though their EC system was deployed relatively close to the canopy top, the spectral energy density decayed with the frequency in a rate close to −5/3, suggesting that turbulence properties resembled the flow in the inertial subrange of steady-boundary layers.Tanny et al. [12] extended these results by measuring turbulent fluxes simultaneously with two EC systems installed at two heights above the crop and below the screen within a large banana screenhouse.Similar friction velocities were measured at the two levels, validating the constant-flux layer assumption within the air gap between the canopy top and the horizontal screen.
Due to the high cost of sensors and complex operation and data analysis, the EC method is inaccessible for day-today use by growers as a tool for irrigation management.To assist growers in improving irrigation management, a family of simplified methods was developed in recent years that are capable of indirectly estimating the canopy sensible heat flux and extracting evapotranspiration as a residual of the energy balance closure.One method of this family is the fluxvariance (FV) method, which is derived from the MOST principle that any scalar variance normalized by the scalar flux depends on atmospheric stability only.Using this theory, it can be shown that, under unstable conditions, the sensible heat flux is proportional to σ 3/2 T , where σ T is the standard deviation of air temperature measured at high frequency (∼10 Hz) above the canopy.Hence, the method can be applied using fast response single-point measurement of air temperature [16] and auxiliary, relatively simple measurements of net radiation and soil heat flux.
e flux-variance (FV) method, which is the topic of the present paper, has been applied in several studies in open fields.ere are three technical aspects involved with applying the FV method: measurement height, sampling frequency, and the available fetch.Measurement height: no study in the literature identified the optimal height where the measurement of sensible heat flux, H, is in best agreement with a reference value.Most literature studies show measurements that were done at a single height in the surface layer, larger than 1.1 h c (e.g., [16]), that is, within and above the roughness sublayer which is about 2 h c .Measurement frequency: most studies reported sampling frequencies between 0.1 and 20 Hz [17][18][19]; however, no report compared the method's performance at different frequencies to identify the optimal one.Fetch: literature studies were conducted under height/fetch ratio in the range 1 : 90-1 : 200 [17,18,20].
Several studies examined the value of C T , the similarity constant associated with the FV method (see Section 2) and the correlation between EC and FV sensible heat fluxes.e value of C T was in the range 0.9-1.1, and the coefficient of correlation with EC sensible heat flux was in the range 0.72-0.98[17][18][19][20].
e main goal of the present study was to examine the FV technique for crops cultivated in three modified environments that are very common in regions of mild winter climates like the Mediterranean basin [4].e structures examined are a tomato greenhouse with impermeable plastic roof and screened sidewall openings for natural ventilation, an insect-proof screenhouses with dense net that blocks insect invasion, in which pepper was grown, and a banana screenhouse that protects the crop from hail, high wind speed, and supraoptimal solar radiation.e ultimate goal is to provide guidelines on the optimal use of the FV technique in estimating ET for irrigation management in such structures.Hence, sensible heat flux estimates using FV are used for extracting evapotranspiration from the energy balance closure, and results are compared with ET measurements.

Theory
e detailed theory of the flux-variance method is given by Wesson et al. [16]. is section provides only a brief outline with major equations.
e Monin-Obukhov similarity theory (MOST) implies that any nondimensional turbulence statistics depends on the atmospheric stability only, ζ � (z − d)/L, where z is the measurement height, d is the zero-plane displacement height, and L is the Obukhov length defined as follows: is the friction velocity, u ′ , v ′ , and w ′ are fluctuations in longitudinal, transversal, and vertical velocity components, respectively, T is air temperature, k is von-Karman's constant, g is the gravitational acceleration, and 〈w ′ T ′ 〉 is the covariance between vertical velocity and temperature fluctuations which represents the mean kinematic sensible heat flux.Based on MOST, the air temperature standard deviation can be expressed as follows: where T * is the temperature scaling parameter given by T * � 〈w ′ T ′ 〉/u * .As shown by Albertson et al. [21] and 2 International Journal of Agronomy Wesson et al. [16] under unstable conditions, the temperature standard deviation, σ T , can be approximated by the following equation: where C T 0.99 is a similarity constant [22].From ( 2) and (3), it can be shown that where ρ is the air density and c p is the air speci c heat at constant pressure.Hence, the sensible heat ux can be estimated by calculating the temperature standard deviation obtained from a single-point measurement of air temperature at high-sampling frequency.Note that (4) can only predict positive sensible heat uxes under unstable conditions since σ T ≥ 0. An approximate expression for stable conditions was also suggested [16], but analysis of such conditions was outside the scope of the present study.
To estimate evapotranspiration, the energy balance closure equation is used as LE + H Rn − G, where LE is the latent heat ux (evapotranspiration), H is the sensible heat ux, Rn is the net radiation, and G is the soil heat ux.e FV latent heat ux is extracted by where H FV is the ux variance sensible heat ux calculated by (4).Since in the present study, the measured energy balance was not perfectly closed (see Section 4), a "closed" LE EB was estimated, derived by forcing the energy balance closure [23] using the eddy-covariance sensible heat ux, LE EB Rn − G − H EC .Finally, the ux-variance latent heat ux, LE FV , is validated against the "closed" latent heat ux, LE EB .

Sites, Crops, and Structures.
e study is based on three comprehensive eld campaigns carried out in three di erent agricultural structures each with a di erent crop.Details on each of the campaigns are given below (see also [24] for details on S1 and S2).  1).Screenhouse cover was a woven screen, with nominal shading of 8% (manufacturer's data), and a rectangular hole of 2.3 mm × 3 mm, made of clear, round polyethylene mono laments 0.3 mm in diameter (Polysack Plastic Industries Inc., Israel).Banana, Grand Nain AAA, was planted during May 2011, in groups of four, separated 4.5 m between rows and 3.5 m between plants in a row.During the experiment, plant height was 4.3 m and Leaf area index (LAI) was 1.4.Plants were irrigated following regional recommendations for screenhouse banana.Soil comprised 44% clay, 26% sand, and 30% silt.Dry and volumetric soil heat capacities were Screenhouse dimensions were 100 × 110 m 2 and 3.8 ± 0.1 m in height, with the longer side oriented east-west (Figure 1).Screenhouse was covered by a white insect-proof 50-mesh screen woven of round polyethylene mono laments, 0.23 mm in diameter, with hole size of 0.46 mm (Ginegar Inc., Israel).Pepper 1715 (Hish-Till, Inc.Israel) was planted on the 15th of September 2012, upon rows 0.60 m wide, separated 1 m apart, with plant density of 3.5 plants•m −2 .During measurements, plant height was constant at 1.6 ± 0.1 m.Plants were planted in compost channels, which were dug into the local Reg soil, and irrigated following regional recommendations.
e Reg soil comprised 50% clay, 25% sand, and 25% silt.e compost comprised 90% clay, 5% sand, and 5% silt [26].us, the dry and volumetric soil heat capacities were International Journal of Agronomy following regional recommendations for greenhouse tomato.Before plantation, the farmer added compost to the soil (60 m 3 •Ha −1 ).Soil comprised 39% clay, 29% sand, 29% silt, and 3% organic matter.Hence, dry and volumetric soil heat capacities were 1 presents the instruments, their heights above soil, z, and normalized height, Z S , in the three campaigns.For S2, the table shows the heights of EC system and net radiometer above (S2a) and below (S2b) the screen.Figure 1 depicts the geometry of each structure and the position of the EC tower and FV miniature thermocouples in each of the structures.

Sensors and Data Acquisition. Table
Miniature thermocouples were placed on the same tower of the EC system, at 9 heights ranging from 1 to 5.8 m and from 1.9 to 5.4 m, in S1 and S2, respectively (−1.34 ≤ Z S ≤ 1.15 and 0.14 ≤ Z S ≤ 1.83, resp.), and at 5 heights ranging from 1.5 to 3.8 m in S3 (0.35 ≤ Z S ≤ 0.89).All high-frequency data were recorded on a CR3000 data logger (Campbell Scientific Inc., USA).

Eddy Covariance.
In S1, the eddy-covariance (EC) system was deployed below the screen at a height of 4.5 m on one of the screenhouse-supporting poles (Figure 1). is position allowed a minimum fetch of about 200 m for the prevailing wind direction in the site.e height of the EC system was between 2.1 and 2.5 m above plant zero-plane displacement height, as estimated according to Stanhill [27] assuming neutral stability.Hence, the height/fetch ratio of the EC system was 1/95-1/80, suitable for surface flux measurements based on footprint models developed for open canopies [28,29].A similar EC setup in screenhouses that are similar to S1 had already been reported to be valid by Tanny et al. [3,12].In S2, the EC system was deployed on the southeast side of the screenhouse at a height of 5 m, which was 1.3 m above the 50-mesh screen (campaign S2a).In a screenhouse similar to S2, the zero-plane displacement of a canopy-screenhouse system was, for neutral stability, about 0.8 of the screen height [30].In S2, the prevailing wind was north-northwest, and resulting fetch was about 100 m. us, the height/fetch ratio at S2 was about 1/50, which is larger than in S1 but still suitable for surface flux measurements according to simplified footprint models [28].In S2, an internal EC system was also deployed below the screen (designated as S2b) and above the plants, at a height of 3 m above the ground.
ree additional ultrasonic anemometers (CSAT3) were deployed in S2 at heights 1.15, 1.7, and 2.5 m above the ground and below the screen.
In S3, the EC system was deployed on the southeast region of the greenhouse at a height of 2.5 m, which was 2 m below the gable.
e zero-plane displacement height of a canopy-greenhouse system was estimated, for neutral stability, to be about 0.67 of the plant height.e dominant wind was north-northwest where the fetch was about 100 m. us, the height/fetch ratio at S3 was 1/75-1/60, which is still suitable for surface flux measurement according to simplified footprint models [28].
Data regarding footprint distribution in the three campaigns are given in Table 4 (Section 4.1), which describes data quality analysis.

Flux Variance.
For FV analysis, air temperature was measured at 10 Hz by fine-wire miniature thermocouples.Radiation load on thermocouples can introduce small errors; however, due to the small size of the thermocouple junctions and low-radiation load (due to the screen or plastic cover), the errors are assumed negligible [31].
In S1, T-type fTcs (fine-wire thermocouples) were selfconstructed in our lab using fine wires of 50 or 76 µm in diameter.When installed in the field, the junction and additional 0.02 m of wire were exposed to the air.In S2 and S3, commercial T-type fTcs (model COCO-002, Omega Engineering Inc., UK), 50 μm in diameter, were installed.Length of exposed wire varied between 0.01 and 0.02 m.In all campaigns, the TC junction was directed upwind and extension wires of 24 gauge SLE thermocouples were used to connect the fTcs to the data logger.

Additional Measurements.
In all campaigns, additional measurements (Table 1) were conducted to enable energy balance closure and data analysis.ese included net radiation, soil heat flux, and soil temperature.Soil heat flux and storage were measured and calculated following Tanny et al. [3].Data of these sensors were recorded on a CR23X data logger (Campbell Scientific Inc., USA).

Data Processing.
Eddy-covariance calculations of sensible and latent heat fluxes were done using the EddyPro software (LI-COR, Inc.) which incorporates all necessary raw data filtering and corrections.Quality control of the EC 4 International Journal of Agronomy data was implemented using the common criteria for testing the developed turbulence and steady-state conditions [15].e developed turbulence criterion is quantified by the value of the integral turbulence characteristic, ITCσ [15]: where σ x is the standard deviation of the variable x and x * is its dynamical parameter (e.g., u * for the velocity).e steady-state test is estimated by where x ′ is the scalar of interest, the subscript SI represents the covariance calculated during a short time interval (e.g., 5 min), and the subscript WI is the covariance calculated during the whole averaging interval (30 min).Using this approach each half hour is categorized according to one of three quality levels from high to low.Highest quality data (classes 1-3) can be used for fundamental research and development of parametrizations (ITCσ, RNx < 0.5), intermediate quality data (classes 4-6) can be used for general use like routine flux measurements (0.51 < ITCσ, RNx < 2.5), and lowest quality data (classes 7-9) can be used for orientation only or rejected [15].Since in the present study the EC data were used as a reference for flux measurements of sensible and latent heat fluxes, data up to class 6, representing the intermediate category of "routine flux measurements," were chosen for analysis.
For the FV data analysis, sensible heat flux was calculated by (4) based on the half-hourly standard deviation of the temperature measured by each fTc.All measurements by the miniature fTcs were conducted at 10 Hz; nevertheless, data were also analyzed at lower frequencies of 1, 2, and 5 Hz.Lower frequency analysis was done by skipping the corresponding data at the temperature time series.is analysis was performed to examine the possibility of measurements at sampling rates lower than 10 Hz, with the goal of using low-cost data acquisition systems in future application of the FV technique.
Data of each campaign were divided into two periods: calibration and validation, as described in Table 2. is division was chosen so as to simulate future application of the FV method, where simultaneous FV and EC data from the initial period are used for calibrating the system, while, during the consecutive period, the FV is operated independently.Hence, during calibration the corresponding value of C T was extracted from regressions of H FV from (4), against direct eddy-covariance measurements of sensible heat flux, H EC .e value of C T thus obtained was then used along with σ T from the temperature measurements in the validation phase for estimating H FV using (4).
Data were also categorized according to the atmospheric stability conditions within each of the structures using the value of L, the Obukhov length scale (1), to indicate the stability level.e criteria based on Webb [32], summarized in Table 3, were adopted in the present analysis.

Quality Control of EC Data.
Figure 2 shows the percentage of EC data points within each range of quality levels: 1-3 (research), 4-6 (flux data), and 7-9 (orientation).Figures 2(a) and 2(b) show the results for the steadystate test of the covariances, w ′ T ′ and w ′ q ′ , that are proportional to the sensible and latent heat fluxes, respectively.Figures 2(c) and 2(d) show the results of developed turbulence tests for the horizontal and vertical velocity components, u and w, and Figures 2(e) and 2(f) for the temperature and absolute humidity, T and q, respectively.In the analysis, data from S2 were presented separately for measurements below (S2b) and above (S2a) the screen.Figure 2 shows that, for most variables, the percentage of data points with flags 7-9 is relatively low. is indicates that most data points in the three structures met the quality criteria of developed turbulence and steady state.
Table 4 summarizes the data quality results.Data points under neutral and stable conditions were excluded from further analysis in this paper.Hence, results in subsequent sections of this paper are based on data that met the three quality criteria: unstable conditions, quality flag <7, and footprint peak within the structure under study.
One of the basic assumptions of the eddy-covariance technique is that measurements are done within the constant-flux layer, namely, the vertical flux of the measured scalar is constant with height. is assumption was examined in campaign S2 where 4 ultrasonic anemometers were installed below the screen and one above it.Figure 3 shows the vertical distribution of mean friction velocity, which represents the vertical flux of horizontal momentum, measured during S2. Figure 3 clearly shows that under the screen and above the plants, at z (Z S ) � 1.7 (0.05), 2.45 (0.39), and 3 m (0.64), the friction velocity is nearly constant with height, supporting the constant-flux layer assumption.Above the screen (Z S � 1.55), the friction velocity is much higher.is is presumably due to the screen being a sink of momentum due to drag.A similar change in friction velocity across a shading screen was observed by Tanny et al. [13].International Journal of Agronomy (7)-( 10) below present the linear ts for the energy balance closure analyses in Figure 4. e results show that, for S1, energy balance slope was low (0.41); however, the high R 2 indicates close correlate between consumed (LE + H) and available (Rn − G) energy.For S2a (pepper screenhouse above the screen), both slope and R 2 were reasonable.For S2b (pepper screenhouse below the screen) and S3, low closure slopes and very low values of R 2 were obtained.Columns from left to right represent campaign code, year, total number of available data points, number of data points under stable conditions, number of data points under neutral conditions, number of data points under unstable conditions, number of data points with quality ags 1-6, and percentage of data points whose ux footprint peak is within the structure.

Energy Balance Closure.
6 International Journal of Agronomy Due to the poor R 2 values of the energy balance closure of S2b and S3, the subsequent analyses of this paper were conducted only for campaigns S1 and S2a.

FV Analysis Based on Temperature Signal of the Ultrasonic Anemometer.
e FV method was initially applied by analyzing the temperature signal measured by the ultrasonic anemometers (CSAT3, Campbell Scienti c Inc., USA). e sensible heat ux was estimated by (4) where during the calibration, the constant C T was estimated from the slope of the regression between H FV and H EC .For validation, the constant C T obtained during calibration was used to estimate H FV that was eventually compared with H EC .Figure 5 shows the regressions obtained during calibration and validation in the two campaigns, S1 and S2a.
e statistical variables associated with the regressions in Figure 5 are given in Table 5. e results in both Figure 5 and Table 5 show that good correlation was obtained between the two uxes in campaigns S1 and S2a.

FV Analysis Based on Temperature Signal of the Miniature
ermocouples.In order to develop a simple measurement system for estimating the sensible heat ux by the FV technique, analysis was carried out for the data of the thermocouples installed at the di erent heights.For each sensor, a regression was calculated between H FV and H EC , and an optimal C T value was determined, which guaranteed a regression slope closest to one.Figure 6 shows vertical distributions of R 2 and C T as a function of normalized height during calibration and validation in the two campaigns.
e results in Figure 6 show that, in S1 (Figure 6(a)), the value of R 2 decreased with height from within the canopy and up to the canopy top, and then, it increases towards the screen level.A slight reduction in R 2 is observed above the screen.In S2 (Figure 6(c)), R 2 was nearly constant just above the canopy and increased sharply near the screen.Maximum R 2 during calibration was obtained in S1 at z 6 m (Z S 1.41) and in S2 at z 5 m (Z S 1.54).In both campaigns, the C T value generally increased with height and reached a maximum value just above the screen.
Table 6 shows statistical variables and C T values for the calibration and validation periods in campaigns S1 and S2a, for the miniature thermocouple that provided the best R 2 in each campaign, as deduced from Figure 6.
Farmers, who will apply the FV method, are not expected to use high frequency (10 Hz) measurements that require expensive data loggers, as done in this study.More probably, simple data acquisition systems with lower sampling frequency would be more practical for such purposes.
erefore, an attempt was made to examine the performance of the FV technique under sampling frequencies lower than 10 Hz.Hence, the standard deviation of the temperature signal was analyzed at frequencies of 5, 2, 1, 0.5, 0.1, and 0.0166 Hz. e analysis was done for the fTc at z 5 m height (Z S 1.54) in S2a which provided the best R 2 at 10 Hz data analysis (Figure 6). Figure 7 shows relations between characteristics of regressions between H FV and H EC , when H FV is calculated by each of the lower frequencies, to that obtained at 10 Hz.
It is shown that ratios between all regression parameters, namely, R 2 (Figure 7(a)), slope (Figure 7(b)), and intercept (Figure 7(c)) are una ected by reducing the data analysis frequency down to 0.5 Hz (i.e., one sample every 2 s!). is result implies that data acquisition systems of relatively low sampling rates (and presumably lower price) can be used in future implementation of the FV technique for day-to-day use by farmers.

Latent Heat Flux: FV Method.
e ux-variance latent heat ux was estimated in each campaign, S1 and S2a, assuming a closed energy balance closure and using H FV .Figure 8 presents the regressions between latent heat ux estimated by the FV technique, LE FV , and LE EB deduced by forcing energy balance closure using H EC .
e corresponding coe cients of the regressions are given in Table 7. Results show high capability of the FV method in estimating the latent heat ux in the two campaigns with deviations of up to 8% (calibration period of S2a) between measured and estimated latent heat ux.

Discussion
is is the rst study that examines the applicability of the ux-variance method in protected environments like screenhouses and greenhouses.e ux-variance method is based on MOST (Section 2), which assumes steady state and well-developed turbulence conditions.In campaign S1 (the banana screenhouse), best performance of the FV method, with largest R 2 , was realized within the canopy and above the screen (Figure 6(a)).e performance was lowest just below the canopy top.In a banana plantation, the leaf area density is low near the ground, increases with height, and then decreases again at plants' top [8].e regions of high R 2 (Figure 6(a)) are commensurate with regions where leaf area   International Journal of Agronomy density is low, presumably since, in these regions, MOST is less violated by the high density of canopy elements.In the pepper screenhouse (Figure 6(c)), R 2 which represents the performance of the FV method increases with height from plants' tops towards the screen and reaches its maximum above the screen.is distribution may also be related to the fact that MOST validity is higher above the screen where conditions of well-developed turbulence are more favorable.FV data are based on the temperature signal of the ultrasonic anemometer.e results correspond to the graphs in Figure 5.
International Journal of Agronomy e three structures examined in this study are very popular among growers in mild winter climates where water scarcity is constantly increasing, and modi ed environments are widely used by growers to increase water saving.Hence, whole canopy evapotranspiration measurements are essential for improving the irrigation management.Although the use of the EC technique has already been established in various types of screenhouses (e.g., [2,3,13]), to the best of our knowledge, this is the rst study which examined the use of the EC method to measure turbulent uxes in a naturally ventilated greenhouse (S3) with an impermeable plastic roof.erefore, one of the goals of this study was to determine and compare the quality of data used for the EC analysis in the di erent structures.
Screenhouses allow relatively intense interaction between inside and outside through the permeable roof, as was already presented by Tanny et al. [3] for a banana crop under a lightshading screen.On the contrary, the greenhouse has an impermeable roof, which totally blocks any such interaction, as well as the vertical exchange of mass between the plants and the outside atmosphere.However, the greenhouse investigated in S3 had screened sidewalls that allowed natural ventilation by horizontal wind ow over the tomato plants.is wind ow facilitated the development of the turbulent canopy ow as was illustrated by the reasonable results of the quality tests of ITCσ and RNx in the greenhouse (Table 4), also for the temperature, humidity, and vertical velocity uctuations (Figure 2).On the other hand, the impermeable plastic roof restricted the vertical ux of momentum, mass, and heat, thereby inducing signi cant advective e ects in the greenhouse, which signi cantly deteriorated the energy balance closure and R 2 (10).In a similar manner, the dense insect-proof screen in campaign S2b also restricted the vertical uxes, which resulted with a poor energy balance closure slope and R 2 values (9). is observation suggests that, in greenhouses, the validity of the two common data quality criteria of developed turbulence and steady state is insu cient to guarantee reliable EC ux measurements.Conditions of developed

Conclusions
A eld experiment was carried out in 3 di erent naturally ventilated agricultural structures, two screenhouses, and a greenhouse to study the suitability of a simple turbulent transport technique, namely, the ux variance, in estimating sensible and latent heat uxes.e major conclusions of this study are as follows: (i) Quality control analysis of the EC method showed that conditions in the shading and insectproof screenhouses (above the screen) were reasonable for ux measurements.However, in the plastic-covered greenhouse and in the insect-proof screenhouse below the screen, energy balance closure and its R 2 were poor.erefore, the FV method could not be employed in these campaigns; (ii) e FV technique was suitable for reliable estimates of H and LE in shading and insect-proof screenhouses, provided the miniature temperature sensors are positioned above the screen; and (iii) data analysis frequency of the temperature signal used in the FV analysis could be reduced down to 0.5 Hz with no e ect on the statistical parameters of the regressions between sensible heat ux measured by EC and estimated by FV.Hence, for the future application of this method, simple and low-cost temperature sensors and data acquisition systems can be used e ectively.Air density (kg•m −3 ) σ T : Standard deviation of air temperature ( °C) EC: Eddy covariance fTc: Fine-wire thermocouple FV: Flux variance S1: Measurements campaign in the banana screenhouse with 8% shading screen S2: Measurements campaign in the pepper screenhouse covered with a 50-mesh insect-proof screen S2a: Measurements above the screen during S2 S2b: Measurements below the screen during S2 S3: Measurements campaign in the tomato greenhouse with a plastic roof.

Figure 4
presents the regressions of half-hourly values of consumed energy (LE + H) versus available energy (Rn − G) in each of the campaigns.Equations

Figure 3 :Figure 5 :
Figure 3: Vertical distribution of the half-hourly friction velocity averaged during campaign S2.Horizontal lines represent the standard error.Canopy height is Z S 0. Screen height is Z S 1.

Figure 6 :
Figure 6: Vertical distributions of R 2 during calibration (black line) and validation (grey line) in the two campaigns, S1 (a) and S2 (c), respectively.(b) and (d) are the corresponding distributions of C T for the validation conditions.In S2, calibration and validation were against reference measurements of H EC above the screen.Z S 0, canopy top; Z S 1 screen height.

Figure 7 :
Figure 7: Ratio between the regression R 2 , slope, and the di erence between intercept, between H FV and H EC , of data analysis at each frequency to those obtained at 10 Hz.(a) Coe cient of determination; (b) slope; (c) Y-axis intercept di erence.

Figure 8 :
Figure 8: e latent heat ux estimated based on FV technique, LE FV , versus the one estimated by forcing the energy balance closure, LE EB .S1 banana screenhouse: (a) calibration; (b) validation.S2 pepper screenhouse (above screen): (c) calibration; (d) validation.Statistical data of these graphs are presented in Table7.

Table 1 :
Instruments, their heights, and normalized heights in the 3 campaigns.

Table 2 :
Calibration and validation periods of the three campaigns.

Table 3 :
Stability regimes and corresponding values of L.

Table 4 :
Summary of data quality control.

Table 5 :
Statistical variables and C T value for the regressions between sensible heat ux measured by the EC and estimated by FV.

Table 6 :
Statistical variables and C T value for the regressions between sensible heat uxes measured by the EC and FV methods.

Table 7 :
Statistical variables for the regressions between latent heat uxes deduced from energy balance and estimated by the FV method (Figure8).Results are presented for the thermocouple positioned at the measurement height that provided the best performance in each campaign.