1. Introduction
Wet canopy evaporation (EWC) is defined as evaporation of the intercepted water by vegetation canopy during and following a rainy period (Stewart 1977). The global average annual precipitation is 840 mm, of which approximately 10% is equivalent to EWC based on the modeling study of the Second Global Soil Wetness Project (Dirmeyer et al. 2006). The quantity EWC has been recognized as a significant proportion of precipitation, especially in forests. Previous studies report that annual EWC ranges from 8% to 29% of total precipitation in broadleaved forests (Rowe 1983; Kim et al. 2005; Deguchi et al. 2006; Šraj et al. 2008), from 17% to 33% in coniferous forests (Johnson 1990; Valente et al. 1997; Link et al. 2004; Kim et al. 2005), and from 10% to 48% in rain forests (Asdak et al. 1998; Schellekens et al. 2000; Vernimmen et al. 2007). In the Asian monsoon climate, annual EWC in forests can also be significant because of an extensive cover of forests and frequent rainfalls.
AsiaFlux, the Asian network of regional flux tower networks, has been conducting the long-term measurements of evapotranspiration (ET) and CO2 fluxes using the eddy covariance (EC) system and has provided a series of gap-filled ET datasets. There are 18 forests sites out of 23 sites whose data are available through the AsiaFlux database (https://db.cger.nies.go.jp/asiafluxdb/). Considering the important contribution of EWC to ET, it is essential to scrutinize the role of EWC in the Asian forests under monsoon climate. For the estimation of an annual ET, typically 10%–40% of the missing data are gap filled using standardized gap-filling methods such as mean diurnal variation and/or modified lookup table (MLT) (e.g., Falge et al. 2001; Reichstein et al. 2005; Hirano et al. 2003; Li et al. 2006; Kosugi et al. 2007; Kang et al. 2009). Artificial neural networks are another method used to fill gaps in the ET data for forest ecosystems (Papale and Valentini 2003; Leuning et al. 2005). Main causes of these gaps are the malfunctions of open-path EC systems due to precipitation and inadequate environmental conditions such as weak turbulence and storm events. However, the environmental conditions used in the gap-filling statistics are mostly collected during dry or partially wet conditions when the EC systems function properly. Therefore, we question if such gap-filled ET data during the wet canopy conditions are biased toward dry to partially wet canopy because of improper consideration of EWC.
The EWC can be either directly observed by the EC method or indirectly estimated as a residual in the energy balance equation (Mizutani et al. 1997; Gash et al. 1999; van der Tol et al. 2003; Czikowsky and Fitzjarrald 2009). During the rainy periods, however, the EC system often fails to operate and results in gaps in flux measurements (Czikowsky and Fitzjarrald 2009; Kang et al. 2009). Hence, the missing observation and the failure of energy balance closure (e.g., Wilson et al. 2002) hinder the use of the residual method from the assessment of annual EWC. Alternatively, physically based EWC models have been used and validated with the observed interception rainfall in various climate and vegetation types (Rutter et al. 1975; Gash 1979; Link et al. 2004). The Rutter-type models, in particular, have been widely adopted for EWC algorithms in many hydrological models and land surface models (LSMs) (e.g., Rutter et al. 1971; Valente et al. 1997).
The purpose of this study is to ascertain the role of EWC to ET for establishing an accurate ET database by examining a possible shortcoming of the current gap-filling method. Our specific objectives are 1) to quantify the durations of wet and dry canopy spells and the operation of the open-path EC system under wet canopy conditions; 2) to characterize the magnitudes and patterns of the directly measured EWC; 3) to evaluate the various model algorithms for the estimation of EWC; 4) to scrutinize the accuracy of the current gap-filling method (i.e., MLT) by comparing with the estimation of EWC based on model algorithms, thereby quantifying the annual difference between the gap-filled and the modeled EWC; and 5) to provide a new gap-filling strategy for overcoming the weaknesses of the current gap-filling method.
To accomplish these objectives, we have implemented multilevel leaf wetness sensors into the ongoing ET measurements with typical open-path EC systems at the KoFlux Gwangneung deciduous and coniferous forests for the entire year from September 2007 to August 2008. We employed and tested a wide range of EWC algorithms [i.e., the De Groen method, the Gash sparse model, Variable Infiltration Capacity (VIC) LSM, and Noah LSM] against the measured EWC by the EC method. The algorithm that most accurately measured the magnitude and variability was used to estimate the missing EWC data under wet canopy conditions. These data were used to identify the mechanisms to generate EWC in the current gap-filling process by the MLT method, which should be implemented to minimize the potential biases in ET data from the sites in monsoon climate.
2. Materials and methods
a. Study sites
The study was conducted at two KoFlux sites: the Gwangneung deciduous forest (GDK; 37°45′25.37″N, 127°9′11.62″E, 291 MSL) and the Gwangneung coniferous forest (GCK; 37°44′54.3″N, 127°9′45.3″E, 120 MSL). Both sites are in a complex, hilly catchment (~220 ha) with a mean slope of 10°–20° and the two sites are about 1.5 km apart (Fig. 1). The 30-yr climate normal was 11.5°C for temperature and 1332 mm for precipitation (Hong et al. 2008). At the GDK site, the vegetation is dominated by an old natural forest of Quercus and Carpinus species (80–200 yr old) with a mean canopy height of ~18 m. The GCK site is located in a generally lower and flat area than the GDK site and is a plantation forest with dominant species of Abies (90–100 yr old) with a mean canopy height of ~23 m. Soil depth is 0.4–0.8 m and the soil texture is mainly silt loam at the GDK site and sandy loam at the GCK site. Further description of both sites can be found in Kim et al. (2006) and Lee et al. (2007).
Location of Gwangneung KoFlux forest catchment. The black dot indicates the deciduous forest (GDK) site, while the white dot indicates the coniferous forest (GCK) site.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Location of Gwangneung KoFlux forest catchment. The black dot indicates the deciduous forest (GDK) site, while the white dot indicates the coniferous forest (GCK) site.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Location of Gwangneung KoFlux forest catchment. The black dot indicates the deciduous forest (GDK) site, while the white dot indicates the coniferous forest (GCK) site.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
b. Eddy covariance and meteorological measurements
Eddy covariance technique was used to measure ET from a 40-m tower at both sites. Vertical and horizontal wind speeds and temperature were measured with a three-dimensional sonic anemometer (model: CSAT3, Campbell Scientific Inc., Logan, Utah) at 10 Hz for both sites. An open-path infrared gas analyzer (IRGA; model: LI-7500, LI-COR Inc., Lincoln, Nebraska) was used for both sites to measure water vapor concentration. Half-hourly eddy covariances and the associated statistics were calculated online from 10-Hz raw data and stored in the dataloggers (model: CR-5000, Campbell Scientific Inc.). Other measurements such as net radiation, air temperature, soil temperature, ground heat fluxes, and soil water content were sampled every minute, averaged over 30 min, and logged in the dataloggers (models: CR-3000 for the GDK site and CR-1000 for the GCK site, Campbell Scientific Inc.). More information can be found in Lee et al. (2007), Kang et al. (2009), and Kwon et al. (2009).
A multilevel profile system was installed at both sites to measure the vertical profile of concentrations of H2O and CO2 and air temperature and to estimate storage effects within the plant canopy using a closed-path IRGA (model: LI-6262, LI-COR Inc.) and a thermocouple (model: Type-E, OMEGA Engineering Inc.) (Hong et al. 2008; Yoo et al. 2009). The profile system, controlled by the dataloggers (model: CR-23X-TD, Campbell Scientific Inc.), was automatically calibrated on a daily basis for H2O zero and CO2 zero–span calibrations and was manually calibrated on a weekly to biweekly basis for H2O and CO2 zero–span calibrations. We used the data of the H2O concentration and air temperature at 40 m from the profile system because of their reliability with regular calibrations and robust operation in rainy conditions.
c. Data processing, quality control, and gap filling
The eddy covariance data were processed, quality controlled, and then gap filled using the standardized KoFlux protocol (Hong et al. 2009). The standardized protocol includes planar fit rotation (PFR; Wilczak et al. 2001; Yuan et al. 2010), Webb–Pearman–Leuning (WPL) correction (Webb et al. 1980), spike detection (Papale et al. 2006), and gap filling (Reichstein et al. 2005). The ET data were screened further based on self-diagnostic variables such as warning flag for CSAT3 and auto gain control (AGC) value for LI-7500. When the total counts of CSAT3 warning flag were above 1800 points (10% of 18 000 points) and the AGC value deviated from the default value (i.e., 50 or 56 for the GDK site and 63 or 69 for the GCK site) with large fluctuation over 30 min, we discarded the data, considering them to be contaminated because of water drops or other objects blocking the signals of each instrument. After the quality control, the ET data retrieval rate was 85% at the GDK site and 80% at the GCK site during the study period. Prior to the gap filling of the ET data, the meteorological data were gap filled based on their linear relationship with the auxiliary data such as air temperature observed at the site. The MLT was used for the gap filling of the ET data. Since the ET data were processed following the standardized protocol of data processing proposed by the global flux network (FLUXNET), the MLT was selected as the gap-filling method in this study (Papale et al. 2006), which is known for its easy implementation and good performance (Moffat et al. 2007). ET data were binned by net radiation (RN), air temperature (Ta), and vapor pressure deficit (VPD) over a time window of 28 days. The binning intervals were 50 W m−2 for RN, 2.5°C for Ta, and 0.5 kPa for VPD. The missing ET was replaced with the binned ET with similar meteorological conditions. More information can be found in Hong et al. (2009) and Kwon et al. (2010).
d. Plant area index measurements
The measurements of the plant area index (PAI) was conducted every 2 or 3 weeks using plant canopy analyzers (model: LAI-2000, LI-COR Inc.) under diffuse light conditions at 10 sampling points with 50 × 50 m2 grid interval at the GDK site and 7 sampling points at the GCK site. Gap fraction was also estimated using LAI-2000 at each sampling point. We applied foliage clumping factor of 1.0 for the GDK site and 1.6 for the GCK site (Gower and Norman 1991), but did not apply shoot clumping factor for both sites. The leaf out occurred in early to mid-April and grew to full size around mid-June with PAI of 5.4 at the GDK site. PAI gradually decreased and reached its minimum of less than 1 in November (data not shown). At the GCK site, the leaf out occurred in early to mid-April and PAI reached its maximum of 7.5 around late May. PAI gradually decreased and reached its minimum of 4.4 in November. The height to the crown base is about 10 m at the GDK site and about 13 m at the GCK site.
e. Measurement of canopy wetness
Canopy wetness was measured using the leaf wetness sensor (model: 237, Campbell Scientific Inc.), which is a simple resistive grid [a circuit board with interlacing gold-plated copper fingers, 71 mm (width) × 75 mm (length) × 6.4 mm (depth)]. We installed these sensors at four different heights considering the canopy structure (i.e., forest floor, base of the crown, middle of the crown, and the canopy top): 0.1, 10, 15, and 20 m at the GDK site and 0.2, 13, 19, and 24 m at the GCK site, respectively. They were mounted horizontally and the output of the sensor (ranging from 0 kΩ to infinity) was stored every 30 min in the dataloggers (models: CR-23X at the GDK site and CR-1000 at the GCK site, Campbell Scientific Inc.). The threshold of dry–wet conditions was set at 150 kΩ, indicating wet condition with 0–150 kΩ and dry condition with >150 kΩ.
We defined the term “fully wet canopy” as the conditions when precipitation is detected and the four wetness sensors are all wet. Similarly, “dry canopy” is defined as the conditions when all the four wetness sensors are dry. Everything in between fully wet and dry canopy is defined as “partially wet canopy”—that is, when at least one (typically the one at the canopy top) of the four sensors is dry. “Wet canopy” corresponds to “fully and partially wet canopy.”
At the GDK site, the transition time required for the wetness sensors to turn from wet to dry conditions was shortest at 20 m (on average, 0.4 h) because of exposure to higher wind and radiation compared to those below (Fig. 2). As expected, the transition time was longest at the forest floor (~5.1 h at 0.1 m). Similarly, at the GCK site, the transition time was shortest at 24 m (0.5 h on average) and longest at 0.2 m (~5.7 h). Transition times of longer than 10 h are noticed near the forest floor at both sites. Unlike the GDK site, the second-shortest transition time was observed in the middle of the crown (i.e., 19 m) at the GDK site, whereas it was below the crown base (at 13 m) at the GCK site. The unique structure of coniferous trees provided relatively open trunk space between the understory vegetation and the crown base (between 6 to 15 m above the ground), where the second maximum in the within-canopy wind speed profile was observed. Overall, the partially wet canopy conditions lasted on average 5 h at the GDK site and 6 h at the GCK site.
The transition time required for the wetness sensors to turn from wet to dry conditions at different heights: (left) GDK and (right) GCK. The leftmost boundary of the box plot indicates lower quartile, the solid line within the box marks the median, and the rightmost boundary of the box plot indicates upper quartile. Error bars below and above the box indicate 10% and 90%. The thick dotted line within the box marks the mean.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
The transition time required for the wetness sensors to turn from wet to dry conditions at different heights: (left) GDK and (right) GCK. The leftmost boundary of the box plot indicates lower quartile, the solid line within the box marks the median, and the rightmost boundary of the box plot indicates upper quartile. Error bars below and above the box indicate 10% and 90%. The thick dotted line within the box marks the mean.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
The transition time required for the wetness sensors to turn from wet to dry conditions at different heights: (left) GDK and (right) GCK. The leftmost boundary of the box plot indicates lower quartile, the solid line within the box marks the median, and the rightmost boundary of the box plot indicates upper quartile. Error bars below and above the box indicate 10% and 90%. The thick dotted line within the box marks the mean.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
f. Modeling of wet canopy evaporation
To calculate EWC, we examined four algorithms used in various models with a temporal scale ranging from hourly to monthly intervals. First, the De Groen algorithm is an analytical model with a daily time scale based on statistical properties of daily rainfall and interception threshold (De Groen and Savenije 2006). The second is the Gash sparse algorithm—the most widely used analytical interception model that provides a simplified solution to the Rutter model with an output based on individual precipitation events (Gash et al. 1995; Muzylo et al. 2009). The other two algorithms are the VIC and the Noah LSMs (Liang et al. 1994; Chen and Dudhia 2001). Both algorithms are conceptually similar to that of the Rutter sparse model, which provides half-hourly outputs (Valente et al. 1997).
1) A monthly interception algorithm
2) Gash sparse algorithm
3) Algorithms in VIC and Noah land surface models
g. Evaluation of model wet canopy duration
h. Error assessment
3. Results and discussion
a. Characteristics of precipitation and wet canopy conditions
In this analysis, we separated the data into the daytime (when incoming solar radiation Rg > 0 W m−2) and nighttime periods (Rg = 0 W m−2) because the mechanism of energy partitioning is different between day and night. For simplicity, we excluded the snow events from our analysis, which amounted to ~1.5% of the total precipitation.
1) GDK site
During the 1-yr study period, the precipitation added up to 1503 mm (i.e., 10% higher than the 30-yr normal). The rainfalls occurred 68 times with the highest frequency (of 14 times) in September (Fig. 3). About 60% of the rainfall events occurred between June and September, which coincided with the period of summer monsoon including typhoons. In terms of the amount, the precipitation during these four months accounted for 77% of the annual total. The duration of precipitation was in a total 642 h (i.e., ~27 days), which spread out evenly to daytime and nighttime. Although the duration of precipitation was similar, the precipitation amount during the nighttime (824 mm) was 20% greater. The duration was longest in July with 155 h, followed by September and August with ~100 h each. As expected, longer duration accompanied with larger amount of precipitation (e.g., 630 mm in July and 292 mm in August).
Monthly values of precipitation frequency, precipitation duration, precipitation amount, and wet (i.e., fully and partially wet) canopy duration at the (a) GDK and (b) GCK sites. Black bar indicates nighttime, whereas gray bar indicates daytime.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Monthly values of precipitation frequency, precipitation duration, precipitation amount, and wet (i.e., fully and partially wet) canopy duration at the (a) GDK and (b) GCK sites. Black bar indicates nighttime, whereas gray bar indicates daytime.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Monthly values of precipitation frequency, precipitation duration, precipitation amount, and wet (i.e., fully and partially wet) canopy duration at the (a) GDK and (b) GCK sites. Black bar indicates nighttime, whereas gray bar indicates daytime.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
In comparison to the duration of precipitation, that of the wet (i.e., fully and partially wet) canopy conditions was 1363 h (~57 days), which was twice as long as and also spread out almost evenly to daytime and nighttime. The characteristics and seasonality of frequency and duration of the wet canopy were similar to those of the precipitation duration.
2) GCK site
The total precipitation was 1476 mm. The monthly patterns of precipitation amount, duration, and wet canopy conditions at the GCK site were similar to those at the GDK site, except small quantitative differences in the duration of precipitation (~30 h shorter) and wet canopy conditions (~30 h longer).
3) Data collection by EC
After the precipitation stopped during the daytime, the wet canopy conditions persisted on average 11.5 (±7.5) h at the GDK site and 12.5 (±7.0) h at the GCK site with an overall range of 2–28 h. During the nighttime, it lasted about 11.5 (±6.0) h with a range of 3.5–32 h at both sites (data not shown). During those 57 days of wet canopy conditions, the percentage of the missing wet canopy evaporation (EWC) data was about 65% at both sites. The major cause of such data loss was the malfunction of the EC system, of which ~90% of the failure was associated with the open-path infrared gas analyzers. On the contrary, the 3D sonic anemometers operated reasonably well under wet conditions, causing only 10%–25% of data loss, as reported by Czikowsky and Fitzjarrald (2009).
We examined the performance of the sonic anemometer during wet conditions following the Monin–Obukhov similarity theory. The relationship between the standard deviation of the vertical wind (σw) and u* should be constant in neutral conditions and the regression between σw and u* should be linear.
When the stability was neutral [−0.1 < (z − dzero)/L < 0.1; where z is the measurement height, dzero is the zero-plane displacement, and L is the Obukhov length] following the previous study on turbulent characteristics at the Gwangneung site (Hong et al. 2008), the slope of the linear regression was 1.29 with r2 of 0.77 for the GDK site and 1.20 with r2 of 0.84 for the GCK site (data not shown). These are comparable with the typical value of 1.25 reported by Garratt (1992). These results are similar to those under dry conditions: 1.23 with r2 of 0.85 for the GDK site and 1.16 with r2 of 0.86 for the GCK site. These results demonstrate that the sonic anemometer performed well without being affected by rain during the wet canopy conditions.
Overall, the characteristics of the precipitation and wet canopy conditions were very similar between the two sites. The large amount of precipitation centered around the growing season along with considerably long durations of wet canopy conditions suggest that EWC at these two sites would be substantial, which was further examined below.
b. Wet canopy evaporation and energy partitioning
To characterize the measured wet canopy latent heat flux (λEWC) and energy partitioning (in terms of the Bowen ratio, β = SH/λE), we analyzed their mean diurnal variations and then compared against those observed under dry canopy conditions (Fig. 4). The sign convention is such that negative sign indicates the energy loss from the surface. Because the measured λE cannot be separated into wet canopy evaporation and transpiration, we excluded the λE under partially wet canopy conditions from this analysis.
Mean diurnal variation of RN, SH, and λE: (a) fully wet and (b) dry conditions at the GDK site, and (c) fully wet and (d) dry conditions at the GCK site. Negative sign convention indicates the energy loss from the surface, while positive sign convention indicates the energy addition to the surface. Partially wet canopy condition was excluded in this figure. Error bar indicates the standard deviation at a given time.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Mean diurnal variation of RN, SH, and λE: (a) fully wet and (b) dry conditions at the GDK site, and (c) fully wet and (d) dry conditions at the GCK site. Negative sign convention indicates the energy loss from the surface, while positive sign convention indicates the energy addition to the surface. Partially wet canopy condition was excluded in this figure. Error bar indicates the standard deviation at a given time.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Mean diurnal variation of RN, SH, and λE: (a) fully wet and (b) dry conditions at the GDK site, and (c) fully wet and (d) dry conditions at the GCK site. Negative sign convention indicates the energy loss from the surface, while positive sign convention indicates the energy addition to the surface. Partially wet canopy condition was excluded in this figure. Error bar indicates the standard deviation at a given time.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
1) GDK site
During the daytime, λEWC ranged from −420 to 20 W m−2. The magnitude of λEWC was on average −53 (±73) W m−2. The averaged β was 0.14, indicating that most of the available energy was partitioned to λEWC under fully wet canopy conditions. The magnitude of λEWC was comparable to λE under dry canopy conditions, ranging from −580 to 60 W m−2 with an average of −55 (±80) W m−2. The β under dry canopy conditions was ~1. During the nighttime, λEWC ranged from −220 to 30 W m−2 with an average of −12 (±25) W m−2, which was greater than λE under dry canopy conditions. The latter ranged from −110 to 60 W m−2 with an average of −3 (±12) W m−2.
2) GCK site
During the daytime, the range of λEWC at the GCK site (−380 to 20 W m−2) was similar to that of the GDK. The averaged λEWC [−57 (±70) W m−2] was smaller than that under dry canopy conditions [−85 (±99) W m−2]. The β under fully wet canopy conditions was virtually zero whereas the β was ~1 under dry canopy conditions. During the nighttime, λEWC was similar to that at the GDK.
Large magnitudes of λEWC (<−100 W m−2) were occasionally observed at nighttime during fully wet canopy conditions at both sites. Several studies have reported large λEWC at nighttime (Pearce et al. 1980; Czikowsky et al. 2006). The large magnitudes of λEWC are attributed to an increased ga and/or additional energy sources such as heat storage and sensible heat advection (Stewart 1977; Schellekens et al. 2000; Czikowsky and Fitzjarrald 2009). For instance, the GCK site had −210 to −90 W m−2 of λEWC from 2130 on 28 October to 0130 on 29 October. During these hours, the canopy was completely wet following 11 mm of the precipitation, and 
c. Comparison among the wet canopy evaporation algorithms
The seasonal patterns of the monthly EWC estimated by the four algorithms were similar and the magnitudes of the monthly EWC were also similar except EWC estimated by the algorithm of the De Groen at the GDK and GCK sites (Fig. 5). As expected, the magnitudes of the monthly EWC were closely related to wet canopy duration at both sites (Fig. 3). The algorithm of the De Groen method overestimated EWC values by more than a factor of 2 compared to the other algorithms. Such an overestimation was due to unrealistically high values of Dd (of 4.2), which should be lower and also seasonally variable. The annual EWC averaged from the other three methods was 83 ±6 mm at the GDK site and 83 ±20 mm at the GCK site. The annual EWC was 6% of the annual precipitation at both sites. For further analysis, we selected the algorithms of VIC and Noah LSMs, which provide half-hourly outputs allowing direct comparison with the measured and gap-filled EWC.
Monthly comparison of EWC by the algorithms of the De Groen (solid), the Gash sparse (white), VIC LSM (white hatched), and Noah LSM (gray hatched) at the (top) GDK and (bottom) GCK sites.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Monthly comparison of EWC by the algorithms of the De Groen (solid), the Gash sparse (white), VIC LSM (white hatched), and Noah LSM (gray hatched) at the (top) GDK and (bottom) GCK sites.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Monthly comparison of EWC by the algorithms of the De Groen (solid), the Gash sparse (white), VIC LSM (white hatched), and Noah LSM (gray hatched) at the (top) GDK and (bottom) GCK sites.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
d. Validation of the estimated wet canopy duration
The duration of wet canopy is an important determinant of λEWC in the algorithms of the LSMs, in which wet canopy is defined as the canopy when Wc > 0 mm, thereby including the conditions of partially wet (or partially dry) canopy.
The observed duration of wet canopy consisted of approximately 16% of the total duration of the field observation at both sites. The VIC LSM identified 93% of the observed duration of wet canopy (H = 0.15) at the GDK sites and 97% (H = 0.15) at the GCK site (Table 1). The Noah LSM, on the other hand, identified 77% (H = 0.12) at the GDK site and 76% (H = 0.12) at the GCK site. The different results between the two LSMs were mainly due to the different calculations of S and n [in Eq. (9)]. The S in Noah LSM was constant (=0.5) whereas S in VIC LSM was a function of PAI, varying from 0.3 to 1.1. As a result, S in Noah LSM was lower than in VIC LSM from April to October at the GDK site (and throughout the entire year at the GCK site), causing a faster rate of EWC and reduced duration of wet canopy condition. In addition, the smaller n (=0.5) in Noah LSM shortened the wet canopy duration compared to that (=⅔) of VIC LSM. Accordingly, the total duration of wet canopy with Noah LSM was 1091 h at the GDK site and 1062 h at the GCK site, which was 280–440 h less than the respective total duration of 1373 and 1502 h with VIC LSM.
Comparison of the wet canopy duration between the algorithms in VIC and Noah LSMs and the observation. Here H denotes the number of cases when the canopy was observed and estimated as wet, whereas M denotes when the canopy was observed as wet but estimated as dry; N and F express the number of cases of dry in the observation that accompanied with dry estimate and wet estimate, respectively; θ1 and k express the fraction of the correct estimate and k statistic, respectively.
The overall performance of the algorithms in the estimation of wet canopy duration was evaluated using the fraction of the correct estimates (θ1) and k statistics [see Eqs. (11) and (12)]. The values of θ1 were high (0.98 with VIC LSM and 0.96 with Noah LSM) because of high values of N from longer hours of dry canopy duration compared to those of wet canopy duration. The values of k, which incorporated the correction for the artifact of high N, were 0.92 with VIC LSM and 0.84 with Noah LSM. These k values are higher than those reported from the empirical models based on relative humidity threshold, decision tree, or fuzzy logic system (0.65–0.70 of the 90th percentile among the 15 sites) (Kim et al. 2010). We concluded that both VIC and Noah LSM algorithms performed very well. In particular, the former performed better because of its realistic leaf phenology in the parameterization of S.
e. Comparison of wet canopy evaporation
In Fig. 6, we evaluated the algorithms of the VIC and Noah LSMs by comparing the modeled λEWC with the observed at the GDK and GCK sites. The number of the data points used in the comparison with Noah LSM was smaller than that with VIC LSM for both test sites. As indicated earlier, this was due to lower values of S and n in the algorithm of Noah LSM [see Eq. (9)], resulting in quickened canopy dryness and shortened wet canopy duration. To minimize the confounding effect of the partially dry canopy in this analysis, the partially wet conditions were excluded.
Comparison of wet canopy evaporation during fully wet canopy conditions: λEWC_Obs indicates observed λEWC, while λEWC_VIC and λEWC_Noah indicate estimated λEWC from the algorithms in VIC and Noah LSMs, respectively: (a),(b) λEWC_VIC vs λEWC_Obs and (c),(d) λEWC_Noah vs λEWC_Obs at the (a),(c) GDK and (b),(d) GCK sites.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Comparison of wet canopy evaporation during fully wet canopy conditions: λEWC_Obs indicates observed λEWC, while λEWC_VIC and λEWC_Noah indicate estimated λEWC from the algorithms in VIC and Noah LSMs, respectively: (a),(b) λEWC_VIC vs λEWC_Obs and (c),(d) λEWC_Noah vs λEWC_Obs at the (a),(c) GDK and (b),(d) GCK sites.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Comparison of wet canopy evaporation during fully wet canopy conditions: λEWC_Obs indicates observed λEWC, while λEWC_VIC and λEWC_Noah indicate estimated λEWC from the algorithms in VIC and Noah LSMs, respectively: (a),(b) λEWC_VIC vs λEWC_Obs and (c),(d) λEWC_Noah vs λEWC_Obs at the (a),(c) GDK and (b),(d) GCK sites.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
During the daytime, the algorithm of Noah LSM underestimated λEWC with MAE of 28 and 24 W m−2 at the GDK and GCK sites, respectively. Although MAE and RMSE were greater because of larger magnitudes of λEWC during the daytime, the agreement with the observed λEWC was better (with higher r2 and d values) than the nighttime at both sites (Table 2 and Fig. 6). Compared to Noah LSM, the algorithm of VIC LSM showed similar results but better agreement with the observed λEWC because of more realistic leaf phenology. Potential causes of such difference may be attributed to inappropriate parameterization of S and/or unaccounted processes of small droplets produced by splashes. Based on the reliable results in estimating λEWC (Fig. 6 and Table 2), we selected the algorithm of VIC LSM for further analysis.
Statistical parameters for error assessment at the GDK and GCK sites.
f. Comparison between the MLT method and the algorithm of VIC LSM
To ascertain the accuracy of the current gap-filling method, we filled up the missing λEWC data during the fully wet canopy conditions by using 1) the MLT gap-filling method (λEWC_MLT) and 2) the algorithm of VIC LSM (λEWC_VIC). As shown in Fig. 7, λEWC_MLT consistently underestimated λEWC_VIC. The mean bias error 
Comparison of wet canopy evaporation simulated by the modified lookup table method (λEWC_MLT) and the algorithm of VIC LSM (λEWC_VIC) at the (a) GDK and (b) GCK sites. The λEWC values during fully wet canopy conditions were used in this analysis.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Comparison of wet canopy evaporation simulated by the modified lookup table method (λEWC_MLT) and the algorithm of VIC LSM (λEWC_VIC) at the (a) GDK and (b) GCK sites. The λEWC values during fully wet canopy conditions were used in this analysis.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Comparison of wet canopy evaporation simulated by the modified lookup table method (λEWC_MLT) and the algorithm of VIC LSM (λEWC_VIC) at the (a) GDK and (b) GCK sites. The λEWC values during fully wet canopy conditions were used in this analysis.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
In addition to the aerodynamic effect, we identified additional energy sources (i.e., SE and advection of sensible heat) that enhanced λEWC but were not taken into account in the MLT-based gap filling. To illustrate such a mechanism driving the λEWC at low RN, we present two cases: 1) the nighttime wet canopy on 9 April 2008 and 2) the daytime wet canopy on 16 July 2008 at GDK and GCK sites. To estimate SE, we considered the heat storage of canopy air space and biomass (e.g., Oliphant et al. 2004) and assumed under wet canopy conditions that 1) G was negligible, 2) Ta and vapor pressure were constant inside the canopy for canopy air space heat storage, and 3) Ta and bole temperature were the same for biomass heat storage (e.g., Papale et al. 2006). Based on the study of Lim et al. (2003), we estimated the biomass to be 26.1 kg m−2 at the GDK site and 31.4 kg m−2 at the GCK site. We used the specific heat of vegetation of 3340 J kg−1 K−1 (Wilson and Baldocchi 2000). In Fig. 8, the λEWC_VIC on both days (with a range of −120–0 W m−2 at nighttime and −260–0 W m−2 at daytime) displayed the mirrored patterns of the sum of the other energy budget components. When RN and VPD were low, however, λEWC_MLT was negligible at nighttime and fluctuated between −40 and 0 W m−2 at daytime. The available energy for λEWC was provided predominantly by the advection of sensible heat (64 ±43 W m−2 at the GDK and 36 ±29 W m−2 at the GCK at nighttime and 42 ±44 and 25 ±38 W m−2 at daytime, respectively). The contribution of heat storage was relatively small with daytime average of 10 ±39 W m−2 at the GDK site and 8 ±36 W m−2 at the GCK site. The contribution at nighttime was even smaller with an average of 2 ±24 and −3 ±37 W m−2, respectively.
Diurnal variation of RN, λE, SH, the sum of three energy components (=RN + SE + SH, where SE is heat storage), and wet canopy evaporation simulated by the modified lookup table method (λEWC_MLT) and the algorithm of VIC LSM (λEWC_VIC) at the GDK and the GCK sites for the dates shown. Shaded area represents the period of wet canopy condition.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Diurnal variation of RN, λE, SH, the sum of three energy components (=RN + SE + SH, where SE is heat storage), and wet canopy evaporation simulated by the modified lookup table method (λEWC_MLT) and the algorithm of VIC LSM (λEWC_VIC) at the GDK and the GCK sites for the dates shown. Shaded area represents the period of wet canopy condition.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Diurnal variation of RN, λE, SH, the sum of three energy components (=RN + SE + SH, where SE is heat storage), and wet canopy evaporation simulated by the modified lookup table method (λEWC_MLT) and the algorithm of VIC LSM (λEWC_VIC) at the GDK and the GCK sites for the dates shown. Shaded area represents the period of wet canopy condition.
Citation: Journal of Hydrometeorology 13, 3; 10.1175/JHM-D-11-07.1
Table 3 shows the annually integrated EWC_MLT and EWC_VIC under fully wet canopy conditions. The annual EWC_MLT at the GDK and GCK sites were respectively 24.0 and 24.7 mm whereas the annual EWC_VIC was much greater with 57.8 and 47.8 mm, respectively. The resulting discrepancies would be site specific and vary considerablely according to canopy structure, frequency and intensity of precipitation, and other meteorological conditions. Overall, the above results suggest the necessity of a separate gap-filling procedure during wet canopy conditions.
Comparison of annual estimation of EWC at the GDK and GCK sites under fully wet canopy condition. The EWC_MLT represents EWC using the MLT method, while EWC_VIC represents EWC using the algorithm of VIC LSM as a gap-filling method; DWC indicates wet canopy duration. Unit of EWC is (mm yr−1).
Prior to suggesting new gap-filling strategies below, it should be noted that so far we have examined fully wet canopy conditions only. However, the duration of partially wet canopy conditions consisted of ¼ of total duration of wet canopy conditions. The annually integrated ETWC_MLT (including both EWC and transpiration) and EWC_VIC under partially wet canopy conditions were 32.7 and 24.4 mm at the GDK site and 54.0 and 38.5 mm at the GCK site, respectively. The comparisons contrast with those under fully wet canopy conditions because of the contribution of transpiration under partially wet conditions (Table 4). Neglecting such contribution would result in biased EWC (e.g., Ohta et al. 2008).
Similar to Table 3 for partially wet canopy condition. ETWC_MLT represents ET using the MLT method as a gap-filling method.
Below, we propose two different gap-filling strategies based on the availability of canopy wetness measurements and canopy wetness conditions.
g. New gap-filling strategies for EWC
1) When the measurements of multilevel leaf wetness are available
As a first step, fill in all the missing gaps using the gap-filling method (e.g., MLT) in which only the data from dry canopy conditions (i.e., when all wetness sensors are dry) are used. Then, for wet canopy conditions (i.e., when at least one of wetness sensors is wet), replace the gap-filled data (from the first step) with the sum of EWC_VIC and the gap-filled data multiplied by 1 − (Wc/S)n (i.e., contribution from transpiration) [see Eqs. (4), (9), and (10)].
2) When the measurements of multilevel leaf wetness are absent
First, as a measure of canopy wetness, calculate the intercepted canopy water [Wc; see Eq. (10)]. Then, follow the same procedures as described above. Fill in all the missing gaps using the gap-filling method (e.g., MLT) in which only the data from dry canopy conditions (i.e., when Wc = 0). Then, for wet canopy conditions (i.e., when Wc > 0), replace the gap-filled data with the sum of EWC_VIC and the gap-filled data multiplied by 1 − (Wc/S)n.
As the final check, we applied the above two gap-filling strategies for the entire periods of wet canopy duration (~57 days) for which ETWC_MLT was 57 mm at the GDK site and 79 mm at the GCK site. The two new gap-filling strategies yielded virtually identical results: 94 mm (~65% increase) at the GDK site and 110 mm (~40% increase) at the GCK site.
4. Summary and conclusions
Based on the direct measurements of wet canopy evaporation by eddy covariance and canopy wetness by multilevel wetness sensors, we have examined the role of EWC in ET from temperate deciduous and coniferous forests in monsoon Asia. The major findings in our study are 1) for the entire year of observation, the duration of precipitation was 27 days in a deciduous forest and 26 days in a coniferous forest in Gwangneung, Korea. Wet canopy duration was ~57 days for both sites, of which 35% was successfully measured by the open-path EC system; 2) the magnitudes of the measured EWC were comparable at daytime and greater at nighttime in comparison to those of ET under dry canopy conditions. Particularly at nighttime, sizable λEWC (−210 to −90 W m−2) was frequently observed because of enhanced aerodynamic effect (i.e., high ga) and additional energy sources from sensible heat advection and heat storage; 3) among the four algorithms tested for the estimation of λEWC, VIC LSM showed the best agreement in terms of wet canopy duration (k ≈ 0.92), magnitudes, and patterns of λEWC (d = 0.81 ~ 0.95) in comparison against the direct measurements; 4) the half-hourly values of λEWC estimated by a traditional MLT-based gap-filling consistently and significantly underestimated those estimated from the algorithm of VIC LSM by 40%–65% because of the failure of considering aerodynamic coupling, advection of sensible heat, and heat storage; and finally 5) we proposed two different gap-filling strategies based on the availability of canopy wetness measurements and canopy wetness conditions.
Despite the potentially significant contribution of EWC to ET, little attention has been given in the process of gap filling of EWC in the monsoon regions. For example, among the current 23 sites whose data are available in the AsiaFlux database, more than half of the sites employ open-path EC systems and apply the MLT-based gap filling for ET. Those datasets deserve further scrutiny regarding the potential biases in EWC. The application of the proposed new gap-filling strategies would improve the reliability of the gap-filled ET database. Alternatively, a better way to resolve the shortcomings of the current gap-filling method may be the direct measurement of EWC by an application of aerodynamic-variance method combined with the use of closed-path EC system, which is currently in progress (e.g., Dias et al. 2009; Yoo 2010).
Acknowledgments
This study is supported by grants (code: 1-8-3) from the Sustainable Water Resources Research Center for 21st Century Frontier Research Program, the Long-Term Ecological Study and Monitoring of Forest Ecosystem Project of Korea Forest Research Institute, and the A3 Foresight Program from the National Research Foundation. We thank the two anonymous reviewers for their constructive comments and suggestions. We also thank the PIs who provided valuable information on the gap-filling method from the flux sites in the AsiaFlux database. We extend our thanks to Dongho Lee, Sujin Kim, Namyi Chae, Jinkyu Hong, Jaeill Yoo, Bindu Malla Thakuri, Juyeol Yun, and Jewoo Hong for their support.
REFERENCES
Asdak, C., Jarvis P. G. , van Gardingen P. , and Fraser A. , 1998: Rainfall interception loss in unlogged and logged forest areas of Central Kalimantan, Indonesia. J. Hydrol., 206, 237–244.
Chen, F., and Dudhia J. , 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585.
Czikowsky, M. J., and Fitzjarrald D. R. , 2009: Detecting rainfall interception in an Amazonian rain forest with eddy flux measurements. J. Hydrol., 377, 92–105.
Czikowsky, M. J., Fitzjarrald D. R. , Sakai R. K. , Moraes O. L. L. , Acevedo O. C. , and Medeiros L. E. , 2006: Direct observation of the evaporation of intercepted water over an old-growth forest in the eastern Amazon region. Extended Abstracts, 27th Conf. on Agricultural and Forest Meteorology, San Diego, CA, Amer. Meteor. Soc., P2.1. [Available online http://ams.confex.com/ams/BLTAgFBioA/techprogram/paper_111122.htm.]
De Groen, M. M., and Savenije H. H. G. , 2006: A monthly interception equation based on the statistical characteristics of daily rainfall. Water Resour. Res., 42, W12417, doi:10.1029/2006WR005013.
Deguchi, A., Hattori S. , and Park H. T. , 2006: The influence of seasonal changes in canopy structure on interception loss: Application of the revised Gash model. J. Hydrol., 318, 80–102.
Dias, L. N., Hong J. , Leclerc M. , Black T. A. , Nesic Z. , and Krishnam P. , 2009: A simple method of estimating sclar fluxes over forests. Bound.-Layer Meteor., 132, 401–414.
Dietterich, T. G., 2000: An experimental comparison of three methods for constructing ensembles of decision trees: Bagging, boosting, and randomization. Mach. Learn., 40, 139–157.
Dirmeyer, P. A., Gao X. , Zhao M. , Guo Z. , Oki T. , and Hanasaki N. , 2006: GSWP-2: Multimodel analysis and implications for our perception of the land surface. Bull. Amer. Meteor. Soc., 87, 1381–1397.
Falge, E., and Coauthors, 2001: Gap filling strategies for long term energy flux data sets. Agric. For. Meteor., 107, 71–77.
Garratt, J. R., 1992: The Atmospheric Boundary Layer. Cambridge University Press, 316 pp.
Gash, J. H. C., 1979: An analytical model of rainfall interception in forests. Quart. J. Roy. Meteor. Soc., 105, 43–55.
Gash, J. H. C., Wright I. R. , and Lloyd C. R. , 1980: Comparative estimates of interception loss from three coniferous forests in Great Britain. J. Hydrol., 48, 89–105.
Gash, J. H. C., Lloyd C. R. , and Lachaud G. , 1995: Estimating sparse forest rainfall interception with an analytical model. J. Hydrol., 170, 79–86.
Gash, J. H. C., Valente F. , and David J. S. , 1999: Estimates and measurements of evaporation from wet, sparse pine forest in Portugal. Agric. For. Meteor., 94, 149–158.
Gower, S. T., and Norman J. M. , 1991: Rapid estimation of leaf area index in conifer and broad-leaf plantations. Ecology, 72, 1896–1900.
Hirano, T., and Coauthors, 2003: CO2 and water vapor exchange of a larch forest in northern Japan. Tellus, 55B, 244–257.
Hong, J., Kim J. , Lee D. , and Lim J.-H. , 2008: Estimation of the storage and advection effects on H2O and CO2 exchanges in a hilly KoFlux forest catchment. Water Resour. Res., 44, W01426, doi:10.1029/2007WR006408.
Hong, J., Kwon H. , Lim J.-H. , Byun Y.-H. , Lee J. , and Kim J. , 2009: Standardization of KoFlux eddy-covariance data processing (in Korean with English abstract). Korean J. Agric. For. Meteor., 11, 19–26.
Johnson, R. C., 1990: The interception, throughfall and stemflow in a forest in Highland Scotland and the comparison with other upland forests in the U.K. J. Hydrol., 118, 281–287.
Kang, M., Park S. , Kwon H. , Choi H. T. , Choi Y.-J. , and Kim J. , 2009: Evapotranspiration from a deciduous forest in a complex terrain and a heterogeneous farmland under monsoon climate. Asia-Pac. J. Atmos. Sci., 45, 175–191.
Kim, J., and Verma S. B. , 1990: Components of surface energy balance in a temperate grassland ecosystem. Bound.-Layer Meteor., 51, 401–417.
Kim, J., and Coauthors, 2006: HydroKorea and CarboKorea: Cross-scale studies of ecohydrology and biogeochemistry in a heterogeneous and complex forest catchment of Korea. Ecol. Res., 21, 881–889.
Kim, K. H., Jun J. H. , Yoo J. Y. , and Jeong Y. H. , 2005: Throughfall, stemflow and interception loss of the natural old-growth deciduous and planted young coniferous in Gwangneung and the rehabilitated young mixed forest in Yangju, Gyeonggido(I)—With a special reference on the results of measurement (in Korean with English abstract). J. Korean For. Soc., 94, 488–495.
Kim, K. S., and Coauthors, 2010: Spatial portability of numerical models of leaf wetness duration based on empirical approaches. Agric. For. Meteor., 150, 871–880.
Kosugi, Y., Takanashi S. , Tanaka H. , Ohkubo S. , Tani M. , Yano M. , and Katayama T. , 2007: Evapotranspiration over a Japanese cypress forest. I. Eddy covariance fluxes and surface conductance characteristics for 3 years. J. Hydrol., 337, 269–283.
Kwon, H., Park T.-Y. , Hong J. , Lim J.-H. , and Kim J. , 2009: Seasonality of net ecosystem carbon exchange in two major plant functional types in Korea. Asia-Pac. J. Atmos. Sci., 45, 149–163.
Kwon, H., Kim J. , Hong J. , and Lim J.-H. , 2010: Influence of the Asian Monsoon on net ecosystem carbon exchange in two major ecosystems in Korea. Biogeosciences, 7, 1493–1504.
Lee, D., and Coauthors, 2007: Lessons from cross-scale studies of water and carbon cycles in the Gwangneung forest catchment in a complex landscape of Monsoon Korea. Korean J. Agric. For. Meteor., 9, 149–160.
Lee, D., Kim J. , Lee K. S. , and Kim S. , 2010: Partitioning of catchment water budget and its implications for ecosystem carbon exchange. Biogeosciences, 7, 1903–1914.
Leuning, R., Cleugh H. A. , Zegelin S. J. , and Hughes D. , 2005: Carbon and water fluxes over a temperate eucalyptus forest and a tropical wet/dry savanna in Australia: Measurements and comparison with MODIS remote sensing estimates. Agric. For. Meteor., 129, 151–173.
Li, S.-G., Eugster W. , Asanuma J. , Kotani A. , Davaa G. , Oyunbaatar D. , and Sugita M. , 2006: Energy partitioning and its biophysical controls above a grazing steppe in central Mongolia. Agric. For. Meteor., 137, 89–106.
Liang, X., Lettenmaier D. P. , Wood E. F. , and Burges S. J. , 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res., 99, 14 415–14 428.
Lim, J.-H., Shin J. H. , Jin G. Z. , Chun J. H. , and Oh J. S. , 2003: Forest stand structure, site characteristics and carbon budget of the Kwangneung Natural Forest in Korea. Korean J. Agric. For. Meteor., 5, 101–109.
Link, T. E., Unsworth M. , and Marks D. , 2004: The dynamics of rainfall interception by a seasonal temperate rainforest. Agric. For. Meteor., 124, 171–191.
Mizutani, K., Yamanoi K. , Ikeda T. , and Watanabe T. , 1997: Applicability of the eddy correlation method to measure sensible heat transfer to forest under rainfall conditions. Agric. For. Meteor., 86, 193–203.
Moffat, A. M., and Coauthors, 2007: Comprehensive comparison of gap-filling techniques for eddy covariance net carbon fluxes. Agric. For. Meteor., 147, 209–232.
Monteith, J. L., 1965: Evaporation and environment. The State and Movement of Water in Living Organisms: Symposia of the Society for Experimental Biology, G. E. Fogg, Ed., Cambridge University Press, 205–234.
Muzylo, A., Llorens P. , Valente F. , Keizer J. J. , Domingo F. , and Gash J. H. C. , 2009: A review of rainfall interception modelling. J. Hydrol., 370, 191–206.
Ohta, T., and Coauthors, 2008: Interannual variation of water balance and summer evapotranspiration in an eastern Siberian larch forest over a 7-year period (1998–2006). Agric. For. Meteor., 148, 1941–1953.
Oliphant, A. J., and Coauthors, 2004: Heat storage and energy balance fluxes for a temperate deciduous forest. Agric. For. Meteor., 126, 185–201.
Papale, D., and Valentini R. , 2003: A new assessment of European forests carbon exchanges by eddy fluxes and artificial neural network spatialization. Global Change Biol., 9, 525–535.
Papale, D., and Coauthors, 2006: Towards a standardized processing of Net Ecosystem Exchange measured with eddy covariance technique: Algorithms and uncertainty estimation. Biogeosciences, 3, 571–583.
Pearce, A. J., Rowe L. K. , and Stewart J. B. , 1980: Nighttime, wet canopy evaporation rates and the water balance of an evergreen mixed forest. Water Resour. Res., 16, 955–959.
Reichstein, M., and Coauthors, 2005: On the separation of net ecosystem exchange into assimilation and ecosystem respiration: Review and improved algorithm. Global Change Biol., 11, 1424–1439.
Rowe, L. K., 1983: Rainfall interception by an evergreen beech forest, Nelson, New Zealand. J. Hydrol., 66, 143–158.
Rutter, A. J., Kershaw K. A. , Robins P. C. , and Morton A. J. , 1971: A predictive model of rainfall interception in forests, I. Derivation of the model from observations in a stand of Corsican pine. Agric. Meteor., 9, 367–384.
Rutter, A. J., Morton A. J. , and Robins P. C. , 1975: A predictive model of rainfall interception in forests, II. Generalization of the model and comparison with observations in some coniferous and hardwood stands. J. Appl. Ecol., 12, 367–380.
Schellekens, J., Bruijnzeel L. A. , Scatena F. N. , Bink N. J. , and Holwerda F. , 2000: Evaporation from a tropical rain forest, Luquillo Experimental Forest, eastern Puerto Rico. Water Resour. Res., 36, 2183–2196.
Šraj, M., Brilly M. , and Mikos M. , 2008: Rainfall interception by two deciduous Mediterranean forests of contrasting stature in Slovenia. Agric. For. Meteor., 148, 121–134.
Stewart, J. B., 1977: Evaporation from the wet canopy of a pine forest. Water Resour. Res., 13, 915–921.
Thom, A. S., 1972: Momentum, mass, and heat exchange of vegetation. Quart. J. Roy. Meteor. Soc., 98, 124–134.
Valente, F., David J. S. , and Gash J. H. C. , 1997: Modelling interception loss for two sparse eucalypt and pine forests in central Portugal using reformulated Rutter and Gash analytical models. J. Hydrol., 190, 141–162.
van der Tol, C., Gash J. H. C. , Grant S. J. , McNeil D. D. , and Robinson M. , 2003: Average wet canopy evaporation for a Sitka spruce forest derived using the eddy correlation-energy balance technique. J. Hydrol., 276, 12–19.
Vernimmen, R. R. E., Bruijnzeel L. A. , Romdoni A. , and Proctor J. , 2007: Rainfall interception in three contrasting lowland forest types in Central Kalimantan, Indonesia. J. Hydrol., 340, 217–232.
Webb, E. K., Pearman G. I. , and Leuning R. , 1980: Correction of flux measurements for density effects due to heat and water vapor transfer. Quart. J. Roy. Meteor. Soc., 106, 85–100.
Wilczak, J. M., Oncley S. P. , and Stage S. A. , 2001: Sonic anemometer tilt correction algorithms. Bound.-Layer Meteor., 99, 127–150.
Willmott, C. J., 1982: Some comments on the evaluation of model performance. Bull. Amer. Meteor. Soc., 63, 1309–1313.
Willmott, C. J., and Matsuura K. , 2005: Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance. Climate Res., 30, 79–82.
Wilson, K. B., and Baldocchi D. D. , 2000: Seasonal and interannual variability of energy fluxes over a broadleaved temperate deciduous forest in North America. Agric. For. Meteor., 100, 1–18.
Wilson, K. B., and Coauthors, 2002: Energy balance closure at FLUXNET sites. Agric. For. Meteor., 113, 223–243.
Yoo, J., 2010: Estimation of CO2 flux in the Gwangneung coniferous forest using aerodyanmic variance method (in Korean with English abstract). M.S. thesis, Dept. of Atmospheric Sciences, Yonsei University, 126 pp.
Yoo, J., Lee D. , Hong J. , and Kim J. , 2009: Principles and applications of multi-level H2O/CO2 profile measurement system (in Korean with English abstract). Korean J. Agric. For. Meteor., 11, 27–38.
Yuan, R., Kang M. , Park S. , Hong J. , Lee D. , and Kim J. , 2010: Expansion of the planar-fit method to estimate flux over complex terrain. Meteor. Atmos. Phys., 110, 123–133.