Progress in water and energy flux studies in Asia: A review focused on eddy covariance measurements

The eddy covariance (EC) technique-based observation system allows for researchers to determine latent and sensible heat fluxes, which are key components of the surface energy balance. The number of water and energy flux studies in Asia has increased as the number of flux measurement sites and the length of the observation periods have grown. To retrace the footprints of the AsiaFlux network and predict future research directions, we reviewed the progress in water and energy flux studies in Asia from the 1990s to the present day. This included studies on continuous evapotranspiration (ET) and surface energy balance measurements in various ecosystems, from the tropics to the polar regions. We also reviewed comparative experiments between the EC technique and other observation techniques including the use of a lysimeter or scintillometer, data processing techniques, connections between carbon and water fluxes, and multi-site syntheses. This paper discusses three remaining challenges that are hindering the derivation of scientific knowledge for ET and the surface energy balance, namely: the non-closure of the surface energy budget, imperfect compatibility between openand closed-path gas analyzers, and difficulty in partitioning ET into evaporation and transpiration. If we leverage the advantages of the EC technique (i.e., high sampling rates of ≥ 10 Hz and continuous measurement capabilities), standardized methods for correcting and partitioning can be developed in the near future.


Introduction
Latent and sensible heat fluxes are the main targets of the application of an eddy covariance EC techniquebased observation system. These fluxes are key components of the surface energy balance. In particular, latent heat flux is critically important, as it can be converted to evapotranspiration ET , which is also a primary component of the surface water balance. Sensible heat flux is related to convection and determines mass and heat transfer to the free atmosphere through the planetary boundary layer. If an open-path type gas analyzer is used to measure trace gas fluxes e.g., carbon dioxide, methane , it is imperative to measure both fluxes to enable the application of air density fluctuation corrections e.g., Webb et al., 1980 . For these reasons, both latent and sensible heat flux measurements have been conducted for most of the observing sites around the world, including in Asia.
The number of water and energy flux studies in Asia has increased along with both the number of flux measurement sites and the length of observation periods. The 20 th anniversary workshop of AsiaFlux i.e., the continental network of flux tower measurements in Asia, http://www.asiaflux.net was held in Takayama, Japan from October 2 to October 5, 2019. Although the history of AsiaFlux is relatively short compared with similar networks in Europe and America, the number of AsiaFlux sites increased to 110 as of January 2020. During this period, the number of publications related to water and energy flux studies in Asia had risen to over 530, largely because more multi-year observation data are now available for synthetic analyses and validations of various models and satellite algorithms Fig. 1 .
To date, several regional networks have been established to monitor carbon and water cycles in terrestrial ecosystems. AsiaFlux is a regional research network in Asia, and it was established in 1999 to study the exchanges of carbon, water, and energy between terrestrial ecosystems and the atmosphere; this network shares data with several national networks such as ChinaFLUX, JapanFlux, and KoFlux. Several papers have provided introductions to such networks. Notably, Mizoguchi et al. 2009 reviewed the state of tower flux observation sites in Asia including AsiaFlux. Yu et al. 2006 introduced ChinaFLUX which has been led by a government knowledge innovation program to support ecosystem integration and local representation, for innovation, research, and forecasting. Lee et al. 2014 established the Lake Taihu Eddy Flux Network, the first lake EC network, and it was launched to monitor the temporal and spatial patterns of lake air fluxes in Lake Taihu, China. These networks are playing increasingly important roles in Earth and environmental sciences. Such network-based studies are providing more opportunities to improve and generalize environmental knowledge through the integration of distributed observations.
To retrace the footprints of the AsiaFlux network and predict future research directions, we reviewed the progress in water and energy flux studies in Asia from the 1990s to the present. In section 2, we classify publications into a number of categories. These publications included observations, technique inter-comparisons, data processing studies, evaluations of the connection between carbon and water fluxes, multi-site syntheses, model-data fusion studies, satellite remote sensing analyses, and finally, other publications that did not fit into any of the previous categories. Among those previous studies, we mainly focused on EC flux measurements in this review. In section 3, we discuss the observations of continuous ET and surface energy balance measurements in various ecosystems e.g., forest, cropland, grassland, urban from the tropics to the polar regions. In section 4, we discuss comparative experiments between EC and other observation techniques, such as lysimeters or scintillometers. In section 5, we discuss data processing techniques such as flux calculation or gap-filling. In section 6, we discuss the connections between carbon and water fluxes. In section 7, we discuss multi-site synthesis methods, such as inter-comparison and upscaling. Finally, in section 8, we outline and discuss the remaining challenges that remain in the field of water and energy flux studies in Asia.

Literature review
To identify what research has been conducted in which field, we classified publications according to the following subjects.
Observations: A study primarily based on observation, including documentation and analyses of the diurnal/seasonal/ inter-annual variations of various land cover types and climates.
Inter-comparison with other techniques: A study primarily based on inter-comparison between EC and other flux observation techniques, such as lysimeter and scintillometer evaluations.
Data processing: A study primarily based on data processing techniques, including gap-filling and partitioning of ET into evaporation and transpiration.
Connections between carbon and water fluxes: A study primarily based on connections between carbon and water fluxes, such as quantifying the efficiency of water use of various land cover types and climates.
Multi-site synthesis: A study primarily based on multi-site observations, such as inter-comparison and upscaling using an observation network.
Model-data fusion: A study primarily based on analytical or numerical modeling of the surface energy balance or hydrology, such as calibration and validation of the land surface-hydrology model.
Satellite remote sensing: A study primarily based on satellite remote sensing, such as ET mapping using satellite imagery.
Other: A study that does not belong to the above categories, such as aerodynamic parameters of boundary layer meteorology.
From 1998 to 2019, the percentage of studies classified as observation 39 was the largest among the water and energy flux studies in Asia Fig. 2 . Those classified as model-data fusion 19 made up the second largest portion, followed by satellite remote sensing 13 . The total number of publications has more than doubled since 2012 Fig. 1 , and such increases occurred in the whole field of research, with the exception of inter-comparison with other techniques. Such overall trends support the postulation that the sharp increase since 2012 was caused by more multi-year observation data becoming available for the whole research field. It should be noted that the boundaries between categories have been becoming ambiguous. In such cases, we judged the category subjectively after checking the journal in which the article was published and reading the abstract. Hereafter, we have reviewed each category, with the exception of the model-data fusion category, the satellite remote sensing category, and the Other category, which were reviewed by Ito in this special issue and Kobayashi in this special issue , respectively.

Observations
We identified a number of studies that reported long-term e.g., annual, whole growing season ET observations from 131 sites across Asia. These applied the EC technique for measuring water and energy fluxes in a wide range of regions from Siberia to Southeast Asia 10 -80 N, 60 -80 E . The EC towers are distributed in tropical, arid, temperate, and boreal climate zones. Figure 3a shows the relationship between the annual mean air temperature and annual precipitation for the 109 sites. The annual mean air temperature ranges from -4 C to 28 C and the annual precipitation ranges from below 30 mm to

Forests
Forest ecosystems exert a significant influence on regional and global water cycling Baldocchi et al., 1988 . The ET from forest ecosystems constitutes nearly one half of the total terrestrial ET Oki and Kanae, 2006 . Forests are believed to function as a "green dam" that 1 prevents flooding and soil erosion via buffering actions that reduce the rate of run-off by intercepting and evaporating precipitation, and 2 conserves the moisture contents in the surface soil by controlling the transpiration rate. For this reason, many long-term field experiments have been performed to measure the exchange of water and energy between forest ecosystems and the atmosphere in various areas of woody PFTs and climates.
The annual ET rate ranged from 330 mm year -1 to 1,600 mm year -1 in 32 forest regions in Asia Table 1 . At the biome level, EBF had the highest mean annual ET rate, followed by ENF, DBF, MF, and DNF Fig. 4 . The annual ET rate gradually decreased from tropical to temperate and boreal zones. The daily ET rate in tropical forests was 2.8 -6 mm day -1 and showed little seasonal variation Igarashi et al., 2015;Takanashi et al., 2010 , and the annual water loss via ET was > 1100 mm year -1 Hirano et al., 2015;Kosugi et al., 2012;Kumagai et al., 2005;Kume et al., 2011 . For the other climatic zones, the peak daily rate was comparable, but strong seasonal variations resulted in much less water loss via ET 330 -1030 mm year -1 . It should be noted that there has been a lack of observations in the arid regions compared to the other regions even though large-scale air temperature C and annual total precipitation mm . Tower locations are indicated by different markers with colors representing the vegetation type based on the International Geosphere Biosphere Program IGBP definitions ENF = evergreen needleleaf forests; EBF = evergreen broadleaf forests; DNF = deciduous needleleaf forests; DBF = deciduous broadleaf forests; MF = mixed forests; SH = shrublands; GRA = grasslands; WET = permanent wetlands; CRO = croplands; Barren = deserts; WAT = water bodies . Gray dots represent the annual mean temperature and total precipitation from the CRU TS v4.03 gridded climate dataset over the entire Asian region Harris et al., 2020 . Temperature and precipitation grid cells included in this figure were averaged with 0.5 resolution from 1989 to 2018. b The relationship between the aridity index and evaporation index of observations across Asia and the difference from the Budyko curve. Each black point represents a grouping of means based on the aridity index at 0.5 intervals, and bars are one standard deviation. The observations from the wetlands were excluded in this analysis. The blue dotted line expresses the limitation by available energy PET < P; PET: potential evapotranspiration, P: precipitation or water PET > P . Gray dots represent the aridity index and evaporation index from the CRU TS v4.03 gridded precipitation dataset Harris et al., 2020 andMOD16A2 ET andPET data Mu et al., 2011 over the entire Asian region. Annual precipitation, ET, and PET grid cells used in this figure are averaged at a 0.5 resolution over the observation period, or averaged from 2000 to 2014. afforestation/reforestation areas in vulnerable arid/semi-arid regions has resulted in a higher demand for ET measurements Cao et al., 2016;Chen et al., 2010;Cho et al., 2019 .

Cropland
Agricultural intensification over the past 50 years, along with green revolution trends, have resulted in an approximate doubling in agricultural productivity. A single plant pumps 200 -500 g of water from soil to leaves to photosynthesize one gram of sugar. This means that water scarcity due to the doubling of crop productivity and the subsequent increase in water consumption remains problematic. In Asia, more than 70 of freshwater resources are used to irrigate agricultural lands Dubois, 2011 , particularly rice paddies. Moreover, crop irrigation is expected to continue increasing in response to the expansion of irrigated areas Siebert et al., 2015 andfuture global warming Wada et al., 2014 . In this context, information on ET for agricultural ecosystems is critical, as such data can allow for better water management, e.g., through adjustments to the timings, amounts, and types of irrigation.
The ET rate ranged from 210 mm to 1,370 mm over the growing season in the 34 croplands in Asia Table 2 . Among the types of crops, cacao had the highest mean growing season ET rate, followed by rice, cotton, maize, vineyards, and wheat Fig. 5 . The highest daily ET rate occurred in the cotton fields and maize fields 7.8 mm day -1 , Jiang et al., 2014;Yang et al., 2016 , followed by the rice paddies 6.57 mm day -1 , Hossen et al., 2011 , wheat Gao et al., 2019 . The growing season mean ET rate of the croplands was highest in the tropical zone 1,215 mm and was within a comparable range in the other climatic zones 400 -650 mm .
The crop coefficient K c method is an approach that has been widely used for the estimation of water consumption by crops, because of its simplicity and robustness. The K c is the ratio of the crop evapotranspiration ET c , mm day -1 to the reference crop evapotranspiration ET o , mm day -1 , and it represents the ET of plants under growing conditions Allen et al., 1998 . Fig. 4. Annual ET mm of each a plant functional type and b climatic zone for the 32 forest sites in Asia. Each box represents the quartile below Q1 and above Q3 the median value, and dots indicate outliers, which are defined as observations more than 1.5 times the inter-quartile range away from the top or bottom of the box.

Fig. 5.
Growing season ET mm of each a crop type and b climatic zone for the 34 cropland sites in Asia. Each box represents the quartile below Q1 and above Q3 the median value, and dots indicate outliers, which are defined as observations more than 1.5 times the inter-quartile range away from the top or bottom of the box. Table 2. Locations and brief descriptions of the eddy covariance-based evapotranspiration ET, mm in the cropland sites reviewed in this study 1 . LAI MAX is the maximum leaf area index m 2 m -2 , P is the mean annual/growing season precipitation mm , T a is the mean annual/growing season temperature in C , K c is the measured crop coefficients of whole growing season, and EBR is the mean energy balance ratio unitless . System types are shown as an open-path eddy covariance OPEC or a closed-path eddy covariance CPEC .
The Food and Agricultural Organization FAO of the United Nations recommended calculation of the ET c FAO56 by multiplying the ET o value by K c as a relatively simple method for assessments Allen et al., 1998 . However, the FAO56 approach can overestimate ET by more than 20 as a result of various factors such as crop variety differences, planting density, and quality of the input dataset Allen, 2000 . In addition, the K c values can vary significantly depending on crop characteristics e.g., leaf area, height, growth stages, and leaf physiological properties and field management strategies e.g., irrigation control, mulching , or environmental conditions, and so data need to be adjusted to reflect the actual conditions Gharsallah et al., 2013;Hunsaker et al., 2003;Katerji and Rana, 2006 . Table 2 also shows the seasonally averaged K c values for various crops. As expected, there was a difference in the K c values between the measured values, and values reported by the FAO Allen et al., 1998 . For example, the K c of cotton and tomato under mulch and drip irrigation was substantially decreased compared to open field conditions. Similarly, in the case of maize, plastic mulch was shown to have a beneficial effect on improving water use Gong et al., 2017a;Li et al., 2008c . Therefore, to accurately estimate K c , it is still prudent to measure the amount of ET directly and continuously as a reference for calibrating and updating the value of K c .

Grassland
Grassland ecosystems are the most dominant ecosystem type throughout the Northern Hemisphere, and these account for approximately 32 of natural global vegetation Parton et al., 1995 . Grasslands not only provide livestock products and plant resources O'Mara, 2012 , but also a wide variety of critical ecosystem services, such as soil erosion reductions, carbon storage, and wildlife habitat FAO, 2010;Fu et al., 2011 . Despite their importance, grasslands are an endangered biome, that is being threatened by land conversion practices, agricultural intensification, fire suppression activities, and abandonment. In addition, grasslands are declining in response to warming from climate change, changed patterns of precipitation, and other trends Guo et al., 2017 . The annual ET rate ranged from 160 to 630 mm year -1 in 26 grassland areas across Asia Table 3 . The mean annual ET rate for meadows was 200 mm higher than that for steppes Fig. 6 . The annual ET rate for the different climate regions was lowest in the arid region, followed by the boreal, and polar regions. The daily ET value during the growing season was 1 -1.5 mm day -1 , and it was less than 0.5 mm day -1 during the non-growing season in the arid area. In the boreal zone, the daily ET rate varied from 1.4 to 2.9 mm day -1 during the growing season and was lower than 0.5 mm day -1 during the non-growing season. The daily ET rate in the polar area was 0.9 -1.3 mm day -1 throughout the whole year. The grasslands have short growing seasons and intense rainfall in the summer regardless of climate region, and the available water during the intensive rainy period is used for vegetation growth.

Others
Measurements of ET also have been conducted for the other types of land cover. The annual ET rate from the other ecosystems ranged from 70 to 3,030 mm year -1 over Asia Table 4 . Depending on the land cover type, the daily ET rate showed large differences, which ranged from less than 0.2 mm day -1 in the desert Kimura et al., 2016 Fig. 7 . In the desert, as expected, the ET rate was extremely low compared to that at the other sites because of the scarcity of water. However, the wetland ecosystem contained open water and much vegetation, and thus, it showed higher ET rates and large variations. Fig. 6. Annual ET mm of each a grass type and b climatic zone for the 26 grassland sites in Asia. Each box represents the quartile below Q1 and above Q3 the median value, and dots indicate outliers, which are defined as observations more than 1.5 times the inter-quartile range away from the top or bottom of the box.

Relationship between ET and precipitation
We identified a trend in which the observed annual ET generally increased with annual precipitation over various ecosystems and climates in Asia Fig. 8 . It is well-known that ET dynamics are complex because ET depends on various controlling factors such as the radiation, temperature, vapor pressure deficit, soil water content, and leaf area index. From the synthetic analysis using the data in this review Tables 1 -5 , it was difficult to find general relationships between the ET and controlling factors, except for precipitation, which could explain with statistical significance the annual ET trends over various ecosystems and climates in Asia. The likely reasons are as follows: 1 primary limiting factors for ET are site-specific, and 2 the Asian monsoon with intensive rainy spells e.g., "Meiyu" in China, "Baiu" in Japan, "Changma" in Korea changes the controlling factors overall. In the same context, the observed annual ET except in wetlands generally followed the Budyko curve Fig. 3b .

Inter-comparison with other techniques
The EC technique is a micrometeorological measurement method used to monitor the vertical turbulent transport of Table 3. Locations and brief descriptions of the eddy covariance-based evapotranspiration ET, mm in the grassland sites reviewed in this study 1 . LAI MAX is the maximum leaf area index m 2 m -2 , P is the mean annual/growing season precipitation mm , T a is the mean annual/growing season temperature in C , and EBR is the mean energy balance ratio unitless . System types are shown as an open-path eddy covariance OPEC or a closed-path eddy covariance CPEC . mass and energy between the surface and atmosphere, and it requires data from both a fast response ≥ 10 Hz sonic anemometer-thermometer SAT and an infrared gas analyzer IRGA placed on an observation tower. Based on a mass conservation equation, the value of ET can be expressed as follows e.g., Baldocchi et al., 1988;Hong et al., 2008 : Table 4. Locations and brief descriptions of the eddy covariance-based evapotranspiration ET, mm in the other land cover sites reviewed in this study 1 . LAI MAX is the maximum leaf area index m 2 m -2 , P is the mean annual/growing season precipitation mm , T a is the mean annual/growing season temperature in C , and EBR is the mean energy balance ratio unitless . System types are shown as an open-path eddy covariance OPEC or a closed-path eddy covariance CPEC . where q is the water vapor H 2 O concentration; u, v, and w represent the velocity components in the longitudinal x , lateral y , and vertical z direction, respectively; h is the measurement height; an overbar denotes Reynolds averaging; a prime e.g., w` denotes the deviation from the mean; and t is time. Term I i.e., eddy flux represents the flux via vertical turbulence, term II i.e., storage flux is the flux stored below the measurement height, term III i.e., vertical advective flux is the flux advected by the mean vertical flow in the presence of a vertical H 2 O gradient, and term IV i.e., horizontal advective flux represents the fluxes transported by the horizontal mean flow and turbulence in the presence of a horizontal H 2 O gradient beneath the height of the measurement. Assuming that the site is flat and homogeneous, and under well-developed turbulent conditions III ≈ IV ≈ 0 , ET can be quantified as the sum of terms I and II. The ET measured by the EC technique typically represents the sum of the eddy flux and the storage flux, or the eddy flux only, in cases where the storage flux can be considered negligible on timescales longer than the daily timescale e.g., Moon et al., 2015 . In the same context, the EC technique can also provide the sensible heat flux if the air temperature is measured by using a SAT instead of with the H 2 O concentration. Often, the EC technique is used to quantify latent and sensible heat fluxes in conjunction with other methods because it has both advantages and disadvantages. The EC technique can provide both fluxes over a relatively large area < 1 km 2 continuously without artificial disturbance, but the experimental field must be large enough; additionally, the EC technique only can provide a spatially averaged value. In this section, we have reviewed studies that were primarily based on the inter-comparison between the EC technique and other flux observation techniques such as those employing a lysimeter and scintillometer. We review the differences in ET fluxes between a closed-path EC system and an open-path EC system in section 8.2.
Lysimeter and pan evaporation techniques are traditional methods that measure the evaporative water loss in the soil or from an evaporation pan. Liu et al. 2009 found that the ET from a winter wheat and summer maize rotation agricultural system in the northwestern Shandong Plain, China, as determined by the EC technique was systematically underestimated by approximately 20 in comparison to that measured by the weighing lysimeter method. Ding et al. 2010 also reported that the ET of a maize field in Northwest China as measured by the EC technique was underestimated by 21.8 during the daytime and by 30.2 during the nighttime in comparison to values measured by a large-scale weighing lysimeter. After adjustments of the daytime EC data by using the Bowen ratio forced energy balance closure method, and adjustments of the nighttime EC data by using the filtering/interpolation method, the differences between the EC technique and the lysimeter method decreased to 4.8 during the daytime and 10.3 during the nighttime. The remaining discrepancy after the adjustments further decreased to 3.2 after discarding overestimated ET data measured by the lysimeter during periods of irrigation and heavy rainfall events. Zuo et al. 2016 showed that the actual evaporation measured by the EC technique and the pan evaporation technique in the arid region of Northwest China presented a clear asymmetrical complementary relationship due to the significant non-uniformity of heat and moisture between the pan water surface and the Fig. 7. Annual ET mm of the 30 sites located at the deserts, lakes, shrublands, and wetlands in Asia. Each box represents the quartile below Q1 and above Q3 the median value, and dots indicate outliers, which are defined as observations more than 1.5 times the inter-quartile range away from the top or bottom of the box. surrounding land surface. Latent and sensible heat fluxes can be measured by other micrometeorological methods. The Bowen ratio-energy balance BREB method is based on the surface energy conservation equation and the gradient diffusion equation e.g., Zhu et al., 2003 . It estimates latent LE and sensible heat H fluxes by measuring the vertical gradients of temperature ∆T and water vapor pressure ∆e to determine the Bowen ratio = H/LE ≈ γ∆T/∆e, where γ is the psychometric constant and the other surface energy components i.e., net radiation and ground heat flux that allow for the quantification of available energy = net radiationground heat flux = LE + H . Inter-comparison experiments between the BREB and EC methods have been conducted in various ecosystems such as heterogeneous grasslands Zhu et al., 2003, forests Shi et al., 2008Wu et al., 2005 , andoases Zhang et al., 2011a in China. The discrepancies between the EC and BREB methods depended on water vapor gradient Wu et al., 2005, vapor pressure deficit Shi et al., 2008 , resolution of slow-response sensors, and atmospheric stability Zhang et al., 2011a . There were a few studies that compared the EC technique with other micrometeorological techniques such as the variational method Yang et al., 2006 and the flux variance, and surface renewal methods Zhao et al., 2010 . Large aperture scintillometers LAS also can be used to measure turbulent characteristics, including H, at regional scales. An LAS transmits near-infrared light over the optical path from the transmitter to the receiver and measures fluctuations of the beam irradiance, which can be used to estimate the refractive index of air caused by turbulent eddies Asanuma and Iemoto, 2007 . With an LAS, one can adjust the field of view of the measurement by up to several kilometers by locating the transmitter and the receiver, whereas the footprint of ECs ~1 km 2 depends on the wind direction and atmospheric stability, which cannot be controlled. Most inter-comparison experiments between an LAS and the EC method were conducted in large areas displaying landscape heterogeneity e.g., Liu et al., 2011; . The discrepancies between the results of the two methods were mainly caused by surface energy imbalances Liu et al., 2011 and different fields of view over heterogeneous surfaces  , whereas the measurements agreed well over homogeneous surfaces Xu et al., 2013 . The ET measurements from ECs were also validated by comparing measurements from the other methods in terms of various spatial and temporal scales. The daily ET rate measured by ECs in the oasis was in good agreement with that estimated by the water balance WB method based on soil moisture measurements Jnad et al., 2006 . Kosugi and Katsuyama 2007 compared daily ET rates measured by ECs in a Japanese cypress forest, while using a correction for the surface energy balance closure to those estimated by the WB method based on precipitation and runoff measurements catchment ET = precipitationrunoff for the inter-validation of both measurements. Li et al. 2008b reported that the EC, WB, and BREB methods provided similar estimates of total ET from a vineyard in the arid desert region of Northwest China.
Multi-scale ET measurements using multiple techniques, such as leaf chambers at the leaf scale , sap flow meters at the plant scale , ECs at the field scale , and the WB at the catchment scale , were also conducted in a cotton field Zhang et al., 2014 , a planted coniferous forest Shimizu et al., 2015 , anda sub-humid mountainous forest Tie et al., 2018 .

Data processing
Data processing consists of the following three main steps: flux calculation, quality assurance and quality control QA/QC , and gap-filling and partitioning. Among the previous studies on flux calculations, Asanuma et al. 2005 introduced the advanced band-pass covariance technique for frequency extrapolation with LE measurements collected by using a relatively slow-response hygrometer. Multiple methods were tested on sites with various topographic and vegetation conditions Zheng et al., 2015;Zhu et al., 2005 , including double rotation rotating the coordinate to set v = w = 0, Wesely et al., 1970 , triple rotation rotating the coordinate to set McMillen, 1988 , and planar fit rotation rotating based on the measured mean wind vector during the entire experimental period as well as a fitted plane obtained by using multiple-linear regression for constructing a stable coordinate frame, Wilczak et al., 2001 methods. It should be noted that most studies found that there was better surface energy balance closure or greater LE after applying the suitable coordination rotation method for the specific site. Yuan et al. 2007 and 2011 introduced a sector-wise planar fit rotation i.e., applying planar fit rotation with each sector of the wind direction to consider the effect of the rolling topography around the flux tower.
The QA/QC and gap-filling steps are essential for constructing high-quality time series data and quantifying diurnal, seasonal, and annual budgets for comparisons with modeling or remote-sensing results. Mano et al. 2007 tested several QC methods for open-path EC flux data, including steady state tests and integral turbulent characteristic tests, and results showed that a higher quality of flux data led to better surface energy balance closure. Many traditional gap-filling techniques, such as mean diurnal variation, nonlinear regression, and marginal distribution sampling MDS, one of the standard gap-filling methods in the global flux network, FLUXNET, Reichstein et al., 2005 were evaluated for sites with various climate conditions and land cover types Du et al., 2014;Park et al., 2015. In particular, Kang et al. 2012 showed that the gap-filled ET data, derived by using MDS under wet canopy conditions, were underestimated because the data used in the gap-filling methods were mostly collected during dry or partially wet canopy conditions; MDS also failed in regard to the consideration of the aerodynamic coupling, advection of sensible heat, and heat storage. Because MDS performed poorly for long-period flux data gaps i.e., gaps longer than a month because of the absence of marginally distributed data around gaps, Kang et al. 2019a suggested that researchers apply a data-driven approach using machine learning and remote-sensing data to apply gap-filling for the long gaps.
In addition to the above research, there have been a considerable number of studies aimed at improving the data quality and usability. He et al. 2010 reported on the random sampling errors for the flux data from six EC sites in ChinaFLUX. Instrument heating corrections for an open-path gas analyzer Burba et al., 2008 where shown to have an insignificant effect on the ET and surface energy balance closure, contrary to the findings for the carbon dioxide CO 2 flux Zhu et al., 2012 . Ono andMaruyama 2015 developed an onsite computation scheme for EC fluxes in real-time assessments. Liu et al. 2016b introduced flux calculations and QA/QC procedures for an EC system installed on a 325 m meteorology tower in an urban area. The studies associated with correcting the surface energy balance closure and partitioning ET into evaporation and transpiration are discussed in sections 8.1 and 8.3.

Connections between carbon and water fluxes
Carbon and water fluxes are key aspects of the functioning of an ecosystem. The ratio of carbon gain to water loss, known as water use efficiency WUE , is an important physiological parameter linking carbon and water cycles. The WUE has been defined in various ways because the spatiotemporal scales and measurement methods used are research-specific. Considering the original definition of the WUE i.e., the ratio of CO 2 flux to H 2 O flux and the spatiotemporal scale of EC measurements, the ecosystem-level WUE and canopy-level WUE can be defined as the ratio of net ecosystem production NEP to ET = NEP/ET , and the ratio of net primary production NPP to transpiration, respectively Kang et al., 2018 . Because of the difficulties associated with the partitioning of the ET into evaporation and transpiration, and in estimating NPP from ECs, most studies based on ECs have reported the ratio of gross primary production GPP to ET = GPP/ET for the canopy-level WUE, with the exception of a few studies e.g., Huang et al., 2010;Kang et al., 2018 . The assessments of the WUE using NEP/ET were conducted mainly in arid regions suffering from water scarcity. For the croplands in arid regions, NEP/ET during the growing season ranged from 0.54 to 1.68 g [ C ] kg -1 [ H 2 O ] i.e., values corresponding to pear orchards in the arid region and maize in arid cropland , where the unit "g [ C ] kg -1 [ H 2 O ] " refers to grams of carbon per kilogram of water Such WUE values are largely regulated by environmental conditions such as the amount of radiation and water. Tong et al. 2014 , based on a 5-year experimental dataset, showed that the GPP/ET was approximately 30 higher under cloudy skies as result of the increase in the proportion of diffuse radiation. Ma et al. 2019 found that drought reduced the GPP/ET in a young plantation, whereas Liu et al. 2017 found that drought enhanced the GPP/ET in an old-growth sub-tropical forest.
These findings suggest that the resilience of ecosystem functions to drought might be system-specific.

Multi-site synthesis
In the early stages of multi-site synthesis, the main objective of most studies was to simply compare the water and energy fluxes among sites with different or similar land cover to obtain a better understanding of the surface energy partitioning process. For example, Kang et al. 2009 quantified the ET from deciduous forest and farmland under a monsoon climate and documented its temporal variations and control mechanisms by using multi-year observations. There have been other studies that have shown differences in water and heat exchanges due to the distinctiveness in surface properties, or meteorological conditions, for two different PFTs e.g., an alpine meadow and banana plantation, Ding et al., 2017; or maize farmland and reed wetland, Li et al., 2009. Zhang et al. 2018b compared the water budget of two typical agricultural ecosystems i.e., winter wheatsummer maize and pear orchard to compute the sustainable usage of groundwater in the North China Plain. In these previous studies, the point was raised that the spatiotemporal variability for water and energy fluxes may depend on climate, PFT, and ecophysiology.
Climate has a significant impact on the seasonal and inter-annual variations of ET via various environmental factors. In Southeast Asia peatland, Hirano et al. 2015 found that drainage and fires caused by low groundwater levels in El Niño years decreased the ET and increased the annual discharge. The rubber trees, a major economic tree crop in tropical areas, exhibited limited water use as a result of their strict regulation of stomatal conductance under seasonal water stress due to the monsoon changes and El Niño -Southern Oscillation ENSO changes Giambelluca et al., 2016;Kumagai et al., 2015 showed that the surface energy partitioning of four sites in the Tibetan Plateau was clearly influenced by the Asian summer monsoon i.e., H was dominant in the pre-monsoon period, whereas LE was greater during the monsoon season . Xiao et al. 2013 synthesized EC fluxes and micrometeorological data from 22 flux sites across China to investigate the variations in water and carbon fluxes, including the WUE, and they found that these processes were being controlled by the annual temperature, precipitation, and growing season length with increasing latitude.
In dry grasslands under water-limited conditions, the available energy, precipitation, and soil water content were found to be the primary factors driving the inter-annual variability of water and energy fluxes during the growing season Liu et al., 2010;Wilske et al., 2010;Yang et al., 2019b . Therefore, many studies have been conducted to improve our understanding of the interactions and coupling mechanisms among the energy, soil, water, and vegetation.  showed that the main factors controlling the daily LE were net radiation in normal years and soil water content in the dry season. Yang et al. 2019b argued that leaf area index LAI is not suitable for estimating the ET of semi-arid natural vegetation, even though it is useful for identifying the physiological constraints on ET. Wilske et al. 2010 found that, among the inputs of the FAO-Penman-Monteith model, the soil water potential and available energy were the main drivers in semiarid shrub-dominated ecosystems.
Because the number of EC flux towers is relatively limited in terms of the spatial coverage because of cost constraints and operational difficulties, it is critical to assess and correct the spatial representation of the EC network before synthetic analyses. An optimized flux network distribution will more accurately represent major ecosystems and promote the integration of fluxes to improve the accuracy of upscaling water and energy fluxes from local tower observations to regional scales. Wang et al. 2013 assessed the spatial distribution of the existing 85 EC flux sites in China by using a multivariate geographic clustering approach and recommended the addition of EC observations numbering up to 100 -150 in total to represent the entire ecoregions of China. In addition, Zheng et al. 2016 constructed the variation of actual ET in China by synthesizing the ecosystem-level EC observation data from 61 sites and argued that additional observation sites are needed for parameterization or validation of global ET products.
Combining observational EC data with satellite remote sensing data is an effective approach for overcoming the limitations posed by a lack of spatial EC observations. Satellite remote sensing is useful for estimating key surface biophysical variables, such as the fraction of photosynthetically active radiation FPAR , LAI, and land surface temperature for an unobserved area. The scaling-up of the observed ET by using remotely sensed data has facilitated ET estimations at the regional and continental level. Wu et al. 2012 estimated ET for water use reduction management in the Hai Basin 320,000 km 2 of North China, within an area that has experienced serious over-exploitation of ground water, and this was accomplished by using new algorithms for ET calculations with remotely sensed data ETWatch ; the results were validated by in situ in Asia. LAI MAX is the maximum leaf area index m 2 m -2 , P is the mean annual/growing season precipitation mm , T a is the mean annual/growing season temperature in C , ET is the mean annual/growing season evapotranspiration mm . NEP is the mean annual/ growing season net ecosystem production g C m -2 , GPP is the mean annual/growing season gross primary production g C m -2 , and EBR is the mean energy balance ratio unitless . System types are shown as an open-path eddy covariance OPEC or a closed-path eddy covariance CPEC . WUE is defined as the NEP/ET or GPP/ET. Vegetation type is defined based on the International Geosphere Biosphere Program IGBP definitions as described in Fig. 3 measurements i.e., lysimeter, EC system, LAS, and water balance calculations . Liu et al. 2016a analyzed the differences between upscaling methods and proposed a combined method for the acquisition of ground-truth ET data at the satellite pixel scale. Similarly, Xu et al. 2018b evaluated the upscaling performance of several machine learning methods e.g., artificial neural network, cubist, deep belief network, random forest, and support vector machine , which were the most popular ones for upscaling from tower-based ET observations to large scales. As noted, the difficulty in matching the entire EC tower sampling area footprint with the satellite pixel scale/model grid scale over a heterogeneous land surface is largely caused by the inherent variability of the EC tower footprint. Xu et al. 2017a proposed a flux aggregation method for determining the area-averaged EC flux with a high-resolution land-cover map to improve the representativeness of EC towers. In spite of some limitations, the current flux integration scheme is expected to provide a richer understanding of the mechanisms controlling water and energy fluxes on a large scale.

Identification of remaining challenges
After navigating a number of publications related to water and energy flux studies in Asia, we noted that there still remain challenges that are hindering the derivation of scientific knowledge on the surface energy balance, namely, 1 the non-closure of the surface energy budget, 2 the imperfection of compatibility between open-and closed-path gas analyzers, and 3 difficulty in partitioning ET into evaporation and transpiration. To identify fruitful directions for future research, we retrace the footprints and summarize the three remaining challenges below.

Non-closure of the surface energy budget
A discrepancy between the terms of available energy estimated by slow-response sensors and an EC system is unavoidable at most flux sites in Asia. Generally, the energy balance has been examined by using the energy balance ratio EBR defined as follows Wilson et al., 2002 : where R net is the net radiation, G is the ground heat flux, and S is the energy storage. Even though EBR should be 1.0 based on the conservation of energy principle the first law of thermodynamics , an actual EBR of less than or greater than 1.0 often is obtained. The extent of non-closure of the surface energy balance with EC measurements and how to handle this phenomenon is still an open question. This is a well-known issue that has plagued the EC flux community for decades. For a comprehensive analysis and evaluation of the surface energy balance closure, field experiments over homogeneous and/or heterogeneous land surfaces have been conducted in Asia Kim et al., 2014;Li et al., 2005;Xin et al., 2018;Xu et al., 2017b .
The non-closure of the surface energy budget can result from factors such as a failure to satisfy the fundamental assumptions of the EC technique, a mismatch of the flux footprint, sampling error, instrument biases, incorrect accounting for storage terms, influence of longwave eddies, and advection effects e.g., Foken, 2008;Leuning et al., 2012;Wilson et al., 2002 . Considering that much of the landscape in Asia is not well suited for the ideal application of the EC technique i.e., the land is not flat and homogeneous , we can expect to encounter such surface energy imbalance problems in the Asian region. The EBR from the studies in this review ranged from 0.5 to 0.99 Tables 1-5 with a median of 0.79. As expected, the median EBRs for the forest sites 0.77 and other sites e.g., wetland, 0.76 were less than those for the cropland sites 0.80 and grassland sites 0.84 , which were typically located on flat and/or homogeneous surfaces Fig. 9 .
A correction for the surface energy balance closure was often applied to some sites in Asia. Most corrections assume that the measured R net -G -S is the true value e.g., Allen, 2008;Pan et al., 2017 . Among the correction methods, the Bowen ratio forced energy balance closure method, while assuming that the measured Bowen ratio by the EC technique is preserved, has been widely used in Asia e.g., Ding et al., 2010;Kosugi and Katsuyama, 2007;Kumagai et al., 2005;Tsuruta et al., 2016 . It is also based on the fact that water vapor and heat are transferred by eddies simultaneously, and thus, there are similarities that allow these processes to be compared. It should be noted that the discrepancies between EC and other methods, such as the BREB and WB methods, decreased after applying a correction for the surface energy balance closure e.g., Ding et al., 2010;Kosugi and Katsuyama, 2007 . Application of a correction for the surface energy balance closure as a standardized procedure is still under debate. Such correction was not applied to most of the flux sites in Asia so far. Leuning et al. 2012 strongly criticized the implicit application of a correction for the surface energy balance closure. They investigated the possible sources of a lack of energy balance closure and found that these were related to 1 too short of an averaging time to capture the low-frequency contributions, 2 a high frequency contribution loss due to the instrument path length, sensor separation, and tube attenuation, 3 the error of radiation measurements over sloping terrain, and 4 the error of the storage term in soil, air, and biomass. They showed that most of the lack of energy could be taken into account in this way. Whether or not to apply this to the CO 2 flux is another issue since CO 2 also shows similarities with water vapor and heat. We expect that international collaborative efforts, such as the Energy Balance Residual Correction initiative Mauder et al., 2020 , will offer a solid argument on this subject in the near future.

Imperfections in the compatibility between open-and closed-path gas analyzers
There are two types of gas analyzers, namely, closed-path and open-path gas analyzers. A closed-path analyzer has an internal sample cell optical path for analyzing the gas concentration that is flushed by sampled air; in open-path sensors, the sample cell is in the open air Munger et al., 2012 . For a closed-path system, the fluctuation of gas concentrations in the high frequency domain is attenuated while the air is drawn in through a tube. Such attenuation depends on measurement conditions such as the flow rate, tube diameter, wind speed, and atmospheric stability. For recovering the loss of fluctuation, a frequency response correction should be applied Aubinet et al., 1999;Moncrieff et al., 1997 . The ET measured by a closed-path system can still be underestimated after applying a frequency response correction for tube attenuation.  reported that the ET measured by a closed-path system with attenuation correction frequently underestimated the ET measured by an open-path system under wet canopy conditions with high relative humidity RH . The RH does not affect the CO 2 flux but does affect the H 2 O flux significantly e.g., Fratini et al., 2012;Ibrom et al., 2007 . The attenuation effect is similar to low-pass filtering, such that the attenuation domain expands with the increasing RH from high frequency to medium frequency. It is worth noting that the ET measured by a closed-path system, after applying the attenuation correction that considers RH, still can be considerably underestimated compared to that measured by an open-path system Fratini et al., 2012;Kang et al., 2019b . Such imperfection in the compatibility between open-and closed-path gas analyzers requires care during synthesizing multi-site measurements and in applying the Bowen ratio forced energy balance closure method to fluxes measured by a closed-path system.

Difficulty in partitioning the ET into evaporation and
transpiration Partitioning of ET into evaporation and transpiration is essential for an improved understanding of ET dynamics. The EC technique measures net fluxes, i.e., the net ecosystem exchange of CO 2 , which consists of ecosystem respiration and GPP, and ET, which consists of plant transpiration T , soil evaporation E S , and wet canopy evaporation E WC , intercepted rainfall . The partitioning of ET into T, E S and E WC is required to understand how ET is affected by environmental changes and how the water cycle is connected to the carbon cycle in an ecosystem, since each component is controlled by different mechanisms and processes e.g, Kang et al., 2018. In particular, Savenije 2004 argued that ET is an outdated terminology that hinders our separate consideration of different evaporative processes in terms of the time scale, time of occurrence, physical characteristics, climatic feedbacks, and isotope compositions.
Practical difficulties still remain for the widespread adoption of ET partitioning. Because of the importance of this subject, there have been a considerable number of previous studies in Asia that partitioned ET by using other supplementary measurements, such as oxygen stable isotopes Liu et al., 2018;Wei et al., 2015;Xu et al., 2016;Zhang et al., 2011c , sap flow Liu andMan, 2017;Zhao et al., 2015;, lysimeter data Gong et al., 2017aLiu et al., 2018;Yang et al., 2018;Zhao et al., 2015; , and supplementary EC system data inside the canopy Kang et al., 2018 . However, such approaches based on other supplementary measurements are costly and difficult to apply to previous data. Therefore, many studies based on a modeling approach using two-source models e.g., Shuttleworth-Wallace model have also been conducted in Asia Hu et al., 2009;Tian et al., 2016;Wang and Yamanaka, 2014;Xu et al., 2016;Yang et al., 2018;Zhao et al., 2015 . However, the modeling approach also requires supplementary measurements to validate the results of ET partitioning. For partitioning the net ecosystem exchange of CO 2 into ecosystem respiration and GPP, most methods are stand-alone and do not require additional data, except for the data from flux towers e.g., Reichstein et al., 2005;Saigusa et al., 2013 . This approach has been applied almost everywhere in the world as a standardized procedure.
Similarly, to activate the partitioning of ET as is done with the CO 2 flux, it will be necessary to develop a stand-alone method for ET partitioning, which can minimize additional but necessary information for application and validation purposes. It should be noted that Wang et al. 2016b reported on the partitioning results of ET from steppe ecosystems in Inner Mongolia following use of the partitioning method based on the flux-variance similarity developed by Scanlon and Kustas 2010 . Their partitioning method is reliant upon the fact that the measured high-frequency time series of CO 2 and H 2 O concentrations are a result of stomatal processes photosynthesis and transpiration, in which CO 2 and H 2 O concentrations are negatively correlated and non-stomatal processes respiration and direct evaporation, in which CO 2 and H 2 O concentrations are positively correlated and thus requires one parameter only, the leaf-level water use efficiency. Most of the ET partitioning studies have focused on the partitioning of ET into E S or direct evaporation and T. In this context, it is noteworthy that Kang et al. 2018 developed the gap-filling and partitioning method for forest ecosystems based on a simplified rainfall interception model, which can estimate E WC using the inputs and parameters from a flux tower measurement and be optimized using available ET data from an EC system under wet canopy conditions. If we continue to capitalize on the advantages and possibilities of the EC technique being capable of a high sampling rate of ≥ 10 Hz and continuous observations, a standardized ET partitioning method will be feasible in the near future.