Dual‐scale transport of sensible heat and water vapor over a short canopy under unstable conditions

Abstract

[1] This study examines the nature of vertical transport of sensible heat and water vapor density due to mesoscale motions and how they are observed with surface eddy covariance systems. For this purpose, turbulence data measured above a flat and irrigated rice paddy field under unstable conditions were analyzed. It was shown that the larger‐scale component in the measured time series represents mesoscale motions. Wavelet cospectra of the scalar fluxes revealed the presence of vertical transport at horizontal scales larger than the cospectral gap. A time series and the eddy covariance flux were decomposed into a turbulence and a mesoscale component using the identified cospectral gap. The mesoscale transport component of sensible heat and water vapor were found to be related to the wind direction and therefore characteristic of the upwind surface conditions at this scale. In contrast, the turbulence flux component exhibits universal attributes resulting from the local surface characteristics.

1. Introduction

[2] Long‐term measurements of the surface energy components have become more feasible recently, owing to device improvements, and knowledge on the scalar exchange rates over numerous land cover is being accumulated [e.g., Malhi et al., 2002; Saigusa et al., 2005]. The monitoring, however, has revealed that the scalar exchange rates measured with the eddy covariance method are consistently smaller than expected from water and energy budget equations. This underestimation has been reported at almost all of the measurement sites regardless of the type of the land cover and external conditions [e.g., Mahrt, 1998; Twine et al., 2000; Wilson et al., 2002]. It has been argued by some that this underestimation can be caused by such factors, as instrumental instabilities, and a biased selection of the measurement location [e.g., Desjardins et al., 1997]. Others [e.g., Mahrt, 1998] attributed this “closure problem” also to the loss of flux components both at scales smaller than the sensor path length and at scales larger than the averaging time. This attenuation of the measured flux at the smallest and largest scale ends is inevitable in surface flux estimations with the eddy covariance method. Since this method integrates fluxes over a finite frequency interval defined by the averaging time and the sampling interval, this is a difficulty inherent in the method. For the flux loss at the smallest scales, correction methods have been proposed [e.g., Massman, 2000] with which the attenuation can be recovered to some extent.

[3] The issue of the flux components at the largest scales, in contrast, has remained unresolved: indeed, the questions to what extent the transport by larger‐scale motions should be included in the flux calculation, and what is the relevant physical mechanisms are left unanswered so far. Sakai et al. [2001], for example, attempted to include the eddy motions at the scale of the boundary layer depth into the flux calculation, and they argued that the effect of such motions on the scalar transport should not be ignored. Sun et al. [1996] and Von Randow et al. [2002], on the other hand, discussed the contribution of mesoscale motions to the surface fluxes. Both of the authors claimed that the mesoscale flux can be up to 20 to 30% of the total flux. In the present paper, mesoscale refers to scales larger than the spectral gap [e.g., Smedman and Högström, 1975; Vickers and Mahrt, 2003], while the smaller scales are called turbulence. It should be noted that, though the common interest of aforementioned investigators is the larger‐scale transports, the actual scales that they dealt with range from the larger turbulence to the mesoscale motions. This scale ambiguity partly stems from the difficulty in determining the exact location of the spectral gap, which is not always obvious [e.g., Williams et al., 1996]. In addition, characteristics of the mesoscale motions seem to vary with measurement sites, and they depend on the local conditions at the measurement sites [e.g., Mahrt et al., 2001]. Hence universal attributes of mesoscale motions are difficult to identify.

[4] Recently, several studies have tried to incorporate flux components associated with these larger‐scale motions into the computed fluxes with procedures, hereafter called the “modified” eddy covariance methods. These modified eddy covariance methods are different from the original eddy covariance method in two respects, namely the operations used in the Reynolds decomposition, hereafter simply called the averaging operations, and the length of the averaging time. One example is by Sakai et al. [2001], who tried to incorporate the flux transported with larger turbulence by applying filtering operations to turbulence records. Rannik and Vesala [1999] also applied three kinds of the averaging operation, though their intention is to exclude the flux carried by the mesoscale motions. As for the averaging time, Finnigan et al. [2003] and Malhi et al. [2004] demonstrated that extending the averaging time up to several hours resulted in increases in the computed surface fluxes and improved the energy budget closure for some of the sites analyzed. Von Randow et al. [2002] used the averaging time up to 4 hours in order to incorporate mesoscale flux. In the practical aspect of these studies, the scale ambiguity is also apparent: It was not clear what scale, either mesoscale or turbulence, the increase in the computed surface fluxes with the modified techniques belongs to.

[5] It should be noted that all of these previous works used observations taken over a tall vegetation. Higher above the ground, the larger the dominant scales of the turbulence eddies are. Therefore the scale separation between the turbulence and the mesoscale is less manifested [e.g., Caughey, 1977]. Probably, this might have been the primary limitation in these works.

[6] The purpose of this study is to examine the role of the larger‐scale motions on the vertical transport of sensible heat and water vapor under unstable conditions. Investigations with the “modified”eddy covariance techniques to incorporate these larger‐scale flux are also made. In contrast to the previous works, this study analyzes measurements over a short canopy at the height of several meters. Moreover, the wavelet transform, which was recently shown to clearly identify the cospectrum gap [Vickers and Mahrt, 2003], is used to decompose the total flux into the turbulence and the mesoscale.

[7] Section 2 describes the data analyzed. The modified eddy covariance methods are applied to the data in section 3. The cospectral properties of these computed surface fluxes and the characteristic length scales of the turbulence and mesoscale motions are discussed in section 4. The physical mechanism of the larger‐scale motions is examined in Section 5, and this is followed by the conclusions in section 6.

2. Data

[8] The data used in the present analysis were collected at the Mase paddy flux site in a rural area of Tsukuba city in Japan [Han et al., 2005; Miyata et al., 2005; Saito et al., 2005]. The site is located in an artificially irrigated flat rice paddy field that extends over 1.5 km from north to south and over 1 km from east to west. The measurements have been conducted since 1999 as a part of the AsiaFlux project. The analyzed data were collected during the intensive observation period between 1 June and 7 September 2004. The rice was transplanted on 2 May and harvested on 9 September 2004. The field was flooded from 17 April to 27 August except for a temporary drainage period in the middle of the growing season. The plants grew to a height of 1.2 m, and LAI (leaf area index) reached a maximum value of about 6 at the end of July. The displacement height, d, was calculated by using d = 0.75 h [Kaimal and Finnigan, 1994], where h is the plant height. In what follows, z is the height above d. The prevailing wind direction during the study was southeasterly, and the surface within 500 m from the measurement tower is almost homogeneous in all directions.

[9] The eddy covariance instruments, consisting of an open path infrared gas analyzer (LI‐7500, Li‐cor) and a sonic anemometer thermometer (DA‐600, Kaijo), were mounted on the top of the micrometeorological tower at 6.3 m above the ground. The instantaneous values of the three components of wind velocity, temperature and water vapor density were sampled at 10 Hz using a 16‐bit digital data logger (DR‐M3, Teac). Since this study focuses on unstable conditions, measurements between 0800 and 1600 in Japan standard time (JST) were chosen, and the original time series were split into 2‐hour runs. Any of automatic procedures for quality control [e.g., Vickers and Mahrt, 1997], as well as the correction procedures had not been applied to the data. Instead, each run was checked manually, and runs with a spike and with an obvious error were discarded. As a result, 97 of 2‐hour runs were selected, and this is the primary data set in this study. When longer time series are needed in the analyses, two contiguous runs were connected, while the 2‐hour run was split into four for shorter time series. These form the second data set with 32 4‐hour runs and the third data set with 388 half‐hour runs.

3. Effect of Surface Flux Calculation

[10] Prior to the analyses of the larger‐scale transport, simple versions of the modified eddy covariance methods were applied to the data. This was aimed to see whether or not the large‐scale transport is, in fact, incorporated by these methods, and, if it is, to investigate which scale the incorporated flux component belongs to. First, the eddy covariance computations with three different averaging operations, block averaging, linear detrending and running mean, were applied to the 30‐min data set. The time constant of these “filtering” operations was chosen to be 30 min, that is the same as the record length. Details on the individual operations are given by, for example, Kristensen [1998] and Moncrieff et al. [2004]. The results were compared with each other, and no significant differences between the averaging operations were found [Saito, 2007]. This is in contrast to previous work [e.g., Rannik and Vesala, 1999; Sakai et al., 2001] in which filtering operations were efficient. This disagreement is probably due to the difference in the height of the observations; all of these previous works used the data over tall vegetation, while this work is over the short vegetation. Particularly in this work, the timescale of these filters roughly corresponds to the cospectral gap where the flux component is small, as shown later in this paper.

[11] Next the influence of averaging time on the eddy covariance fluxes is investigated. Three averaging time lengths, 30 min, 2 hours and 4 hours were used to calculate covariances, and the results were mutually compared. Individual points of 30‐min surface fluxes in Figure 1 are the average value of the four or eight contiguous 30‐min surface fluxes in the corresponding 2‐hour and 4‐hour periods, respectively. In what follows, raw covariances, and , are used in the presentation of the results in place of sensible and latent heat fluxes. This is to avoid the uncertainty in the mean values such as the air density, the specific heat, and the heat of vaporization. In Figure 1, calculated using 2‐hour averaging time, _{2 hr}, are, on average, 8% larger than those with 30 min, _{30 min}. For some runs, _{2 hr} is almost twice as large as _{30 min}. As the averaging time extends to 4 hours, increases further, and the ratio of _{4 hr} to _{30 min} is about 1.3 on average. Sun et al. [1996] claimed that the contribution of mesoscale motions to the total flux can be as large as 20% of the turbulent flux, but that it can be negligibly small when an average over many records is taken, as the transport by mesoscale motions can be either upward or downward. The current results also indicate that the motions with timescale between 30‐min and 4‐hour transport the sensible heat either upward or downward, however that they have a overall tendency to be upward.

Figure 1

Comparisons between the computed (a) and (b) with 30‐min, 2‐hour, and 4‐hour averaging times. The solid line represents y = x, and the dashed and dotted lines are the regression line forced through the origin between 30 min and 2 hours and between 30 min and 4 hours, respectively.

[12] Figure 1 also indicates that the averaging time length, which affects , do not make a large contribution to up to 4 hours as a whole. This is consistent with Finnigan et al. [2003], who reported that computed and show different dependencies on the length of the averaging time. Individual points in Figure 1b indicate that differences between the different averaging times, which is the contribution of larger‐scale motions to the vapor transport, are smaller than 0.02 g m⁻² s⁻¹, and that the scatter is equally distributed on both sides of the line y = x. Therefore the slope of regression lines are about 1 for both 2 hours and 4 hours, and averaging over whole runs would reduce the relative importance of larger‐scale motions in the captured . This is also consistent with the findings of Sun et al. [1996] as described earlier. A note should be made that the current results are characteristic to the data over a short canopy, and cannot be directly applied to a tall canopy.

4. Wavelet Cospectra

[13] In section 3, it was shown that surface fluxes computed with the eddy covariance method can be quite sensitive to the length of the averaging time. In this section, the scale dependence of the surface fluxes is examined in the wave number domain by using the wavelet transform. The wavelet transform is superior to analyze nonstationary or intermittent motions [e.g., Katul et al., 1994], since it has a good localization property, both in the time and the frequency domains. The present study uses the orthogonal wavelet transform with the Haar basis [e.g., Daubechies, 1992; Katul and Parlange, 1994]. The Haar basis ψ (x) is given by

The wavelet coefficients Wt_m,n of a square integrable function f(x) are obtained as

where ψ (x) = 2^−m/2ψ (2^−mx − n), m is scale and n is time location.

[14] The wavelet transform was applied to the 2‐hour and 4‐hour data sets, and wavelet cospectra [Katul et al., 2001] were computed from the wavelet coefficients. Since the Haar wavelet only allows the data number of an integer power of 2, the cospectra are calculated at the discrete time, namely, 2¹ × 0.1 s, 2² × 0.1 s,…, 2¹⁶ × 0.1 s (≈109.2 min.) for 2‐hour data set and up to 2¹⁷ × 0.1 s (≈218.5 min.) for 4‐hour data set. These cospectra were then normalized with the corresponding covariance that was calculated as an average of the 4 or 8 covariances, each of which were for the quarter or 1/8 segments of the data, respectively.

[15] Figure 2 presents normalized wavelet cospectra, Cow and Co_wρv, bin‐averaged over all runs. Taylor's hypothesis was used to convert the time to the wave number as κ = 2 π/λ, where is the mean wind speed and λ is the timescale. Figure 2 shows clearly that cospectral gaps exist in both Cow(κ) and Co_wρv(κ) at around κz ∼ 10⁻². This dimensionless wave number is associated with a timescale of around 20 min under the average conditions at the site. Therefore the abscissa in Figure 1 is nearly equal to the turbulent component of the surface fluxes. This will be revisited later in this text.

Figure 2

Bin‐averaged wavelet cospectrum of (a) and (b) . The open and solid circles are 2‐hour and 4‐hour data sets, respectively. The wavelet cospectra are normalized with the corresponding covariance. See the text for details. The error bars represent a standard deviation from the mean.

[16] An important finding in Figure 2 is that considerable amounts of the scalar transport by mesoscale motions exist at wave numbers smaller than the cospectral gap. In contrast to the fact that Cow(κ) and Co_wρv(κ) agree with each other in the range of the turbulence scales, these normalized wavelet cospectra depart from each other in the mesoscale range, and the standard deviation of each wave number bin increases toward the smallest wave number, especially for Cow(κ). Moreover, in the mesoscale, the bin‐averaged Cow(κ) increases with decreasing wave numbers until around 10⁻³ (≈2 hours), while Co_wρv(κ) approaches zero. Though the scatter is large, t tests with the significance level of 0.05 indicate that the bin average of Cow(κ) at κz = 1 × 10⁻³ and 2 × 10⁻³ is significantly larger than zero for both 2‐hour and 4‐hour data sets. A note should be also made that the standard deviations of are much smaller than those of Cow in the mesoscale range. These tendencies of the normalized wavelet cospectra explain the results given in Figure 1, where _{2 hr} and _{4 hr} sometimes exceed largely _{30 min} even though _{2 hr} and _{4 hr} are not significantly different from _{30 min}.

[17] Considerable amounts of Cow(κ) at mesoscale for both 2‐hour and 4‐hour data sets prompt one to speculate that there may be vertical transport of sensible heat at the mesoscale induced by mesoscale motions, which are characterized by timescale of a few hours. The heat and scalar transport at the smaller wave numbers, or more specifically in the mesoscale range, has been argued already by Mahrt et al. [1994], Sun et al. [1996], Von Randow et al. [2002], Finnigan et al. [2003], and Strunin and Hiyama [2005]. Note that most of these studies are based on observations over taller vegetation or by aircraft. The present study, in contrast, presents observational evidence that there may be vertical transport of heat and scalar in the mesoscale range even over a shorter canopy, where the vertical motions tend to be suppressed in the presence of the surface. This will be investigated further in section 5.

[18] In Figure 2, the cospectral gap in the bin‐averaged cospectra is located at κz ∼ 10⁻², as mentioned earlier. The location of this cospectral gap, however, varies from run to run, and thus the cospectral gap was determined for each run. The gap location was identified as the wave number where the cospectrum changes its sign or that of the local minimum of the cospectrum at wave numbers smaller than the cospectral peak. In these determinations, was used (and not ) due to its better manifestation of the cospectral gap location compared with . The distribution of the cospectral gap location expressed as a dimensionless wave number is shown in Figure 3. The location of the cospectral gaps ranges from κz = 3 × 10⁻³ to 1 × 10⁻¹, whereas 80% of the total cospectral gaps lie at normalized wave numbers between 5 × 10⁻³ and 4 × 10⁻².

Figure 3

Frequency distribution of the cospectral gap location expressed with the dimensionless wave number, κz.

5. Mesoscale Motions

[19] The above analyses suggest that computed fluxes can be made to include the additional transport of sensible heat and water vapor by using longer averaging times, and that these additional flux components are located in the mesoscale range. In order to investigate further the nature of the sensible heat and water vapor transport at mesoscale and how they are related to the actual atmospheric motions, more detailed analyses were made with 2‐hour data set. For this purpose, the total flux was decomposed into two components, namely one in the turbulence region and one in the mesoscale by using the cospectral gap identified in section 4. These components are hereafter termed turbulence flux and mesoscale flux, and are indicated by the subscripts, _T and _M, respectively. Each of them was calculated by integrating the wavelet cospectrum over the corresponding wave number region. It should be noted that the “mesoscale flux” computed herein is a part of the vertical transport in mesoscale, and Figure 2, in fact, suggests that there can be transport at the scale longer than 2 hour.

[20] The diurnal variations of and in the dual ranges, namely, turbulence and mesoscale, are shown in Figure 4 for 4 clear days, namely, 5, 14, and 16 June and 3 July 2004. While the turbulence fluxes follow the typical diurnal pattern, the mesoscale fluxes do not, and the diurnal variation of the mesoscale fluxes is much smaller than the turbulence. This is surely the cause of the vast variation of the cospectra at mesoscale in Figure 2, as these cospectra were normalized with the 30‐min covariance, that is almost equal to _T or _T. This invalidates scaling of the mesoscale fluxes with their turbulence counterpart, and suggests relatively independent relationship between the fluxes in these two scale ranges. Most striking is that, in contrast to the erratic scatter in mesoscale of the cospectra (Figure 2), the mesoscale fluxes shown in Figure 4 are, in fact, “regulated” to some extent. A common pattern in the diurnal variation between _M and _M, such as a larger absolute value in the late morning, suggests a common transporting mechanism for the sensible heat and water vapor. The day‐to‐day variation of the mesoscale fluxes is at the same order of magnitude with its diurnal variation.

Figure 4

Diurnal variations of (a) _T and _M and (b) _T and _M on 5, 14, and 16 June and 3 July 2004. The open and solid circles represent the fluxes in the turbulence range and in the mesoscale range, respectively.

[21] In Figure 5, “Bowen ratio” in the dual ranges are compared. For this purpose, a kind of “Bowen ratio” is defined herein as β ≡ /. This value in mesoscale has large uncertainty due to the variation of the mesoscale fluxes around zero, as shown in Figure 4. Therefore, instead of plotting β, _M is plotted against _M, each normalized by its corresponding turbulence component. The solid line in Figure 5 indicates _M/_T = _M/_T, which, in turn, is β_M = β_T. Similarly, the dotted lines in Figure 5 indicate different values of β_n = β_M/β_T. The negative values of β_n in Figure 5 correspond to β_M < 0, which occurs in the case that _M and _M are of opposite sign, this is described later in this section. The sign of β_T is always positive. It is obvious in Figure 5 that the values of β are quite different in the two wave number regions, and that in most of the cases β_T is much smaller than the absolute value of β_M. The land cover around the site at these scales plainly explains this difference. The turbulence flux is controlled mainly by the local surface condition, indeed, evapotranspiration from a rice paddy field is normally much larger than the sensible heat flux, and therefore β_T < 1. In contrast, at the horizontal scales associated with the mesoscale flux, approximately of the order of 1 to 10 km, the land cover contains also drier surfaces, such as urbanized areas and uplands, and therefore the absolute value of β_M can be expected to be significantly larger than β_T. This immediately suggests ∣β_n∣ = ∣β_M/β_T∣ >1, and is generally consistent with the observations in Figure 5. In other words, the mesoscale fluxes of the sensible heat and water vapor are associated more with surfaces outside the paddy area.

Figure 5

_M plotted against _M, each of which is normalized by the corresponding turbulence flux, namely, _T and _T, respectively. The solid and the dotted lines indicate different values of β_n = β_M/β_T. See the text for details.

[22] The nature of mesoscale fluxes was further investigated in Figure 6, which presents quadrant relationships between _M and _M. The data points are divided into two groups based on the mean wind direction, namely the data with the wind direction between 60° and 180° (clockwise from the north) and those between 180° and 360°. Hereafter, these two groups are referred to as “southeasterly wind” and “nonsoutheasterly wind”, respectively. There are no data for the wind direction between 0° and 60°. The runs with variable wind direction during 2 hours were excluded from the analysis.

Figure 6

Relationship between _M and _M. The open and solid circles represent the data with the southeasterly wind and the nonsoutheasterly wind, respectively.

[23] In Figure 6, nearly all of the runs with southeasterly wind are plotted in either the first or the third quadrant, while those with nonsoutheasterly wind tend to be in the 2nd or the 4th quadrant. In other words, when the wind is from the southeast, _M and _M tend to have the same sign. For nonsoutheasterly wind, on the other hand, _M and _M tend to be of opposite sign with only a few exceptions.

[24] To explore this point further, time series of vertical velocity, w, temperature, θ, and water vapor density, ρ_v, were reconstructed with the inverse wavelet transform from the wavelet coefficients in the mesoscale range. Two runs, namely, 26 June, 1400–1600, and 14 June, 1000–1200, were selected as typical examples for the southeasterly wind group and nonsoutheasterly wind group, respectively. The average wind directions in these runs were 79° and 255°, respectively. The reconstructed time series, each of which is normalized by its standard deviation, are shown in Figure 7.

Figure 7

Reconstructed time series of vertical velocity, w, temperature, θ, and water vapor density, ρ_v, at mesoscale typical to the runs (a) with southeasterly wind and (b) with nonsoutheasterly wind. All variables are normalized by their standard deviation. The data periods are 1400–1600, 26 June for southeasterly wind and 1000–1200, 14 June 2004 for nonsoutheasterly wind.

[25] In Figure 7a with southeasterly wind, the reconstructed time series of θ and ρ_v at mesoscale show a monotonic increase with time, while w at mesoscale shows a slightly decreasing trend. This is in contrast with the nonsoutheasterly wind group given in Figure 7b, where ρ_v exhibits a decreasing trend. All of these features with the reconstructed time series are commonly found among each of the wind direction groups, except that w trace shows either increasing or decreasing trend.

[26] The monotonic increase of θ and ρ_v in the southeasterly wind group gives a positive correlation of w‐θ and w‐ρ_v when w has an increasing trend, while a negative w‐θ and w‐ρ_v correlation is produced with a decreasing w. This explains the fact in Figure 6 that most of the points of southeasterly wind group are in the first and third quadrant and therefore that _M and _M have the same sign predominance. Similarly, the decreasing ρ_v with nonsoutheasterly wind explains the points with this wind direction tend to be in the 2nd and 4th quadrant in Figure 6.

[27] These findings in Figures 6 and 7 probably are associated with contrasting land covers in the upwind regions. In the southeastern direction from the observation site, there are relatively moist surfaces including lakes and wetlands, and the Pacific Ocean is located in the northeastern to the southeastern direction at about 40 to 50 km away from the site. On the other hand, the upwind regions of the nonsoutheasterly wind are basically terrestrial, and include relatively drier surfaces, such as suburban Tokyo, than those with southeasterly wind. Therefore relatively warmer and drier advection is expected with the nonsoutheasterly wind than with the southeasterly wind. This is consistent with the observations in Figure 7 that ρ_v has an increasing trend with the southeasterly wind. Since an increasing ρ_v indicates a moistening boundary layer [Mahrt, 1991], they are caused by the moist advection with the southeasterly wind. The drying boundary layer associated with the decreasing ρ_v and the dry advection from nonsoutheasterly wind would seem to be also mutually related.

[28] Another possible connection between the observed mesoscale fluxes and the wind direction may be the influence of the wind direction on mesoscale circulations. The landscape around the observation site is characterized by floodplains along several tributaries of the Tone river and the uplands among them. These tributaries flow in parallel from north to south with intervals of about 5 to 10 km. The floodplains are covered mainly by paddy fields, while residential areas are scattered in the uplands. The altitude difference between the lowlands and the uplands is typically only several meters. Figure 8 shows the diurnal variations of the surface temperature at the Mase paddy field as well as a nearby grassland, on 14 June 2004. The grassland data were obtained at Terrestrial Environment Research Center, University of Tsukuba, that is located at about 10 km to the northeast from the paddy site. The surface temperatures differ by about 10°C between these two land covers. These marked difference of surface temperature could possibly have induced mesoscale circulations [e.g., Pielke, 1984; Segal et al., 1988; Avissar and Chen, 1993], which are often recognized in satellite pictures as lines of convective cumuli above the uplands. The strong directionality of the landscape raises the possibility that the mesoscale circulations over this region and therefore the mesoscale fluxes induced by them may be controlled by the wind direction.

Figure 8

Diurnal variations of radiation temperature on the surfaces at Mase paddy flux site (Paddy) and the experimental grassland in Terrestrial Environment Research Center, University of Tsukuba (Grass), on 14 June 2004.

[29] Since observational evidence is limited, the exact physical mechanism of the observed characteristics of the mesoscale fluxes was not fully identified. The observations presented herein, however, are probably evidence that the observed mesoscale fluxes are connected to the land surface characteristics at the mesoscale and therefore atmospheric motions associated with it. This suggests that the observed mesoscale fluxes are not an artifact but are, at least, relevant to the real phenomena.

6. Conclusions

[30] This study analyzed turbulence data collected over a rice paddy field under unstable conditions. The evidences strongly supports the idea that large‐scale components in the measured time series, which have been traditionally removed in the eddy covariance calculation, in fact, represent the vertical transport of scalars associated with mesoscale motions. Cospectral gaps were identified for each 2‐hour run using wavelet cospectra of sensible heat flux, whereupon eddy covariance fluxes and measured time series were decomposed into mesoscale and turbulence components by using the identified cospectral gap. The Bowen ratio at the mesoscale was shown to be quite different from that of the turbulence. Decomposed time series of temperature at the mesoscale exhibited a monotonically increasing trend, while the water vapor density traces at the mesoscale monotonically increased or decreased depending on the wind direction. Though the evidence is limited, these facts indicate that the observed vertical transport of sensible heat and water vapor at the mesoscale is related to larger surface areas than the turbulence flux.

[31] The results presented in this study indicate that the mesoscale motions play a role in the land‐atmosphere exchange of heat and other scalars, and they imply that a fraction of mesoscale vertical transport can be measured with a tower observation as was done here. Interesting facts found in this paper are that some effects of mesoscale motions are sensed even at only a few meters above the ground.

Acknowledgments

[32] The authors are grateful to M. Mano, H. Nagai, T. Yamada, K. Ono, Y. Kobayashi, and G. H. Han for their support of the experiment and useful comments and to A. Higuchi, who provided the support data for this study. This research was financially supported, in part, by the Ministry of Education, Culture, Sport, Science and Technology (MEXT), Japan, through Grant‐in‐Aid for Scientific Research (C) (17560450) and Asahi Breweries Foundation through its academic grant.