Estimation of water use efficiency in the terrestrial ecosystems of the Yangtze River Delta region, China

Understanding the characteristics of water use efficiency (WUE) could enhance our understanding of the integrated response of terrestrial ecosystems to climate change. In this study, the vegetation photosynthesis model (VPM) and the revised remote sensing-Penman Monteith (RRS-PM) model were used to characterize the spatial-temporal dynamics of gross primary productivity (GPP) and evapotranspiration (ET) in the Yangtze River Delta (YRD) region. Our predicted GPP and ET results were in the range of those reported by other models. To be specific, the RRS-PM model could explain 85% of the variability in ET at the Anji site, China. The root mean square error between the simulation value and the measured ET was 8.543 mm month -1 . For the evergreen broadleaf forest, deciduous needle leaf forest and deciduous broadleaf forest, GPP decreased and ET increased, thus the WUE = (GPP/ET) inevitably decreased. The increase in WUE in grassland was due to the decreased ET, for which the rate of decrease was greater than for GPP. For the spatial distribution of WUE, the maximum value occurred in the southern and northeastern parts of the YRD region, while the minimum value was distributed in the middle and northwest. This is likely due to the different land-cover types. In the south, there is more forest, while in the north the land-cover type is almost entirely cropland. In our study, GPP increased with increasing LAI across all of the plant function types. (ET), photosynthesis


IntRoDuCtIon
Terrestrial ecosystem gross primary productivity (GPP) is a quantitative description of the amount of carbon fixed by the vegetation and is at the beginning of the carbon biogeochemical cycle. In the study of the carbon cycle, GPP has been an important indicator (Zhao and Running 2010). Freshwater resources are unevenly distributed around the world and in areas of water shortage regional development is restricted, even threatening normal life for some populations. The water cycle is the basis of replenishment of freshwater resources in nature, and it is only by fully understanding each process in the water cycle, that crises in the supply of fresh water are likely to be resolved. Studies have shown that more than half of all precipitation falling on land surfaces returns to the atmosphere by the terrestrial ecosystem evapotranspiration (ET). Thus, the terrestrial ecosystem ET plays an important role in the global water cycle (Mu et al. 2011).
The carbon and water cycles are closely related in the terrestrial ecosystem because the exchange of carbon dioxide and water vapour between biosphere and atmosphere are both controlled by stomata (Beer et al. 2009). Water use efficiency (WUE), the ratio of gross primary productivity (GPP) to evapotranspiration (ET), can be used to quantify this coupling relationship (Yu et al. 2004;Tang et al. 2014;Wang et al. 2015;Zhu et al. 2015). Therefore, to use the rare and precious water resources efficiently and reasonably, it is essential to understanding the characteristics of GPP, ET, WUE, and their relationship with related driving factors. Meanwhile, understanding the GPP, ET, WUE characteristics, and their relationship with related factors can greatly enhance our knowledge of their control processes as well as the adaptation mechanism of carbon and water cycle to climate change .
In recent years, eddy-covariance (EC) flux tower networks have provided an opportunity to quantify WUE at the ecosystem scale (Reichstein et al. 2007). However, to estimate WUE at the regional or global scale, there is a huge challenge when using flux observations only. To date, two methods exist for evaluating carbon and water fluxes at the large scale. First, by using a process-based ecosystem model, they can be simulated (Zhu et al. 2011;Ito and Inatomi 2012). Second, remote sensing models can be used for regional-and global-scale carbon and water fluxes estimation (Xiao et al. 2004a;Wang et al. 2015). Remote sensing models are of small calculation and a strong theoretical basis, which have become an important means of simulating terrestrial ecosystem carbon and water fluxes at regional scale. GPP is quantified as a function of light use efficiency and climatic factors (Xiao et al. 2004a), while ET is calculated based on the Penman-Monteith equation (Cleugh et al. 2007).
Taking one of China's most developed, dynamic, densely populated, and concentrated industrial areas, the Yangtze River Delta (YRD) region as a study area, the primary objectives of the work described in this paper were (1) to explore the spatial-temporal characteristics of GPP, ET, and WUE in the terrestrial ecosystems of the YRD region; and (2) to quantify the relationship of GPP, ET in the terrestrial ecosystems with their biotic variable, namely leaf area index.

Description of the study Area
The center of the Yangtze River Delta (YRD) region is located at 30.5°N and 120.5°E. The study area has a typical subtropical monsoon climate with a warm and humid summer and a cold and dry winter. The annual mean air temperature (Ta) and mean precipitation (PPT) are 15.7°C and 1158 mm, respectively (Wu et al. 2014). Specifically, from late June to early July, it is persistently overcast with rain in this region. The major forest type is subtropical evergreen broadleaf forest (Fig. 1).
The spatial-temporal dynamics of GP ET and WUE in

Description of the VPM Model
Based on the assumption that nonphotosynthetically active vegetation within a canopy makes no contribution to GPP, the VPM model revises the light use efficiency model as: where FPAR PAV is the fraction of photosynthetically active radiation (PAR) absorbed by leaf chlorophyll in the canopy, PAR is the photosynthetically active radiation (μmol photosynthetic photon flux density, PPFD), and ε is the light use efficiency (μmol CO 2 /μmol PPFD) which is affected by temperature, water, and leaf phenology: where ε max is the maximum light use efficiency and T scalar , W scalar , and P scalar are used to quantify the effects of temperature, water and leaf phenology on light use efficiency of vegetation, respectively. For details of the VPM model, one can refer to Xiao et al. (2004a).

Estimation of the Maximum light use Efficiency
The maximum light use efficiency can be obtained from an analysis of the net ecosystem exchange (NEE)-PPFD relationship at flux tower sites or from a survey of the literature (Ruimy et al. 1995). The relationship between NEE and PPFD can be fitted with a linear or Michaelis-Menten function. In this study, a literature survey was conducted to collect the maximum light use efficiency determined by the Michaelis-Menten function. For each vegetation type, the average of the maximum light use efficiency was used to estimate the GPP (Table 1).

Estimation of t min , t opt , and t max
In calculation of Tscalar, the minimum, maximum and optimal temperature for photosynthetic activities (T min , T opt , and T max ) vary among different vegetation types. A literature survey was conducted to gather the values of different vegetation types (Table 2).

Description of the RRs-PM Model
Based on the model proposed by Mu et al. (2007), Yuan et al. (2010) developed the RRS-PM model. Compared with the former, there are 2 changes in the RRS-PM model. One is the equations dealing with the temperature constraint for stomatal conductance, the other is the equations' computing energy allocation between vegetation canopy and soil surface. In RRS-PM model, ET is composed of vegetation transpiration (Tr) and soil evaporation (E). The full description of the RRS-PM model is provided as follows.
Penman-monteith (PM) formula was used to calculate vegetation transpiration: where Δ is the slope of the curve relating the saturated water vapour pressure (e s ) to the temperature, A c is solar radiation received by the canopy, ρ is the air density, C p is the specific heat capacity of the air, e is the actual water vapour pressure, γ is the psychrometric constant, r a is the aerodynamic resistance, γ is the psychrometric constant, and r s is the surface resistance.

Ac Rn As Rn Rn e LAI
where Rn is net radiation, As is the solar radiation received by the soil surface, k is the extinction coefficient, the value is 0.5, and LAI is the leaf area index. ] ] where C l is the maximum stomatal conductance of the unit leaf area, T is air temperature, T opt = 25°C, VPD is the saturation vapor pressure deficit, VPD open and VPD close is the value of VPD corresponding to the stoma is unrestricted and value of VPD corresponding to the stoma is completely restricted.
The soil evaporation was obtained by combining the potential soil evaporation with the soil moisture restriction . r T where E p is the potential soil evaporation, E is the soil evaporation, r sa is the aerodynamic resistance (s m -1 ), r r is resis-tance to radiative heat transfer (s m -1 ), r tot is the total aerodynamic resistance to vapor transport, v is the boltzmann constant, RH is the relative humidity.

Parameterization of the RRs-PM Model
In this study, the values of the vapor pressure deficit (VPD) corresponding to full stomatal closure (VPD close ), the mean potential stomatal conductance per unit leaf area (C l ), and the total aerodynamic conductance to vapor transport (C tot ) were derived from the results of Xu et al. (2013). In their study, the values of parameters were determined by nonlinear curve fitting. First the initial parameter values and their bounds were determined. When the value of a parameter could not be fitted, a parameter value obtained from previous studies was assigned (Yuan et al. 2007;Mu et al. 2011), and then the nonlinear regression was run again. Finally, the values of VPD close , C l , and C tot were 5209 Pa,  No reference! E a r l y R e l e a s e 0.00149 m s -1 , 0.0039 m s -1 , respectively.

Model Inputs
The model inputs mainly included land surface water index (LSWI), enhanced vegetation index (EVI), land-use/ cover data, leaf area index, air temperature, and photosynetically active radiation (PAR). As MODIS does not provide LSWI products, 500 m surface reflectance was used to calculate the LSWI. The 8-day MODIS products were composited into monthly data using a maximum value composite method (Holben 1986). Combined with the monthly EVI and yearly land cover data, all MODIS products were downloaded at the same website (http://ladsweb.nascom. nasa.gov/data/search.html). The Global LAnd Surface Satellite (GLASS) LAI, produced by the general regression neural networks (GRNNs) method, has demonstrated that the quality of GLASS-LAI data has been greatly improved (Xiao et al. 2013;Xiang et al. 2014). GLASS-LAI data with a temporal resolution of 8 days were obtained from the Center for Global Change Data Processing and Analysis of Beijing Normal University (http://glass-product.bnu.edu. cn/en/). Monthly LAI data was integrated using a maximum value composite method (Holben 1986). In our study, we downloaded NASA's Modern ERA-Retrospective Analysis for Research and Applications of Global Modeling and Assimilation (MERRA) monthly products as meterological inputs (http://gmao.gsfc.nasa.gov/research/merra/).

Field Measurements and Data Processing
Water flux was measured with EC systems consisting of open-path infrared gas analyzers (Li-7500; Licor Inc., Lincoln, NB, USA) and a 3-D sonic anemometer (CSAT3; Campbell Scientific Inc., Logan, UT, USA). The EC systems were mounted 38 m above ground at AJ. A data logger (CR1000; Campbell Scientific Inc., Logan, UT, USA) recorded the EC signals at 10 Hz for archiving and on-line turbulence statistics computation.
2-D coordination rotation and Webb Pearman-Leuning correction were applied to obtain half-hourly mean water vapor flux. All abnormal data, caused by instrument malfunctions and weather effects, such as rain and dew, were deleted. Missing data of less than 2 hrs were filled using linear interpolation. Large gaps (more than 2 hrs) were filled using the mean diurnal variation method. The monthly value of both fluxes was integrated from the half-hourly data.

Assessment on the VPM Model
We compared the results simulated by the VPM model with those reported in the literature (Table 3). Table 3 indicates that our results were in the range of those reported by other models except that GPP of grassland in this study was much higher. The difference between GPP of grassland predicted by different models was mainly due to the different climate conditions across the different study areas (Table 4). The YRD area has a typical subtropical monsoon climate, and therefore the grassland in this area has a relatively higher productivity. When we compared the results to flux measurements with similar climate conditions, the VPM results were in the range of the flux observations (Table 3).

Assessment on the RRs-PM Model and the WuE
Water vapor flux data during 2012 -2014 at the Anji site was used for the validation of the RRS-PM model. Two statistical indicators (the determination coefficient and root mean square error) were chosen for the assessment of the RRS-PM model. The results showed that the RRS-PM model could simulate the value and seasonal variation of ET at the flux site (Fig. 2). To be specific, the RRS-PM model could explain 86% of the variability in ET, the root mean square error between the simulation value and the measurement was 8.543 mm month -1 (Fig. 3).
Subsequently, we compared the ET and WUE results in this study with those reported in the literature (Tables 5 and 6). According to the literature, the ET of the evergreen needle leaf forest, evergreen broadleaf forest, deciduous broadleaf forest, grassland and cropland ranged from 510 -822, 630 -905, 581 -957, 451 -681, and 486 -688 mm a -1 , respectively. According to the literature, the WUE of the evergreen needle leaf forest, evergreen broadleaf forest, deciduous needle leaf forest, deciduous broadleaf forest, grassland and cropland ranged from 0.52 -3.66, 1.88 -3.51, 1.29 -1.68, 1.36 -2.57, 0.41 -1.90, and 1.57 -4.02g C kg -1 H 2 O, respectively. Our modeled ET and WUE were in the range of the results reported in the literature.

temporal and spatial Variations of GPP
From 2001 -2014, GPP in all vegetation types in the YRD region showed a decline trend, but at different rates (Fig. 4). GPP in deciduous broadleaf forest and grassland exhibited greater decreasing trends than other vegetation types. In contrast, GPP in cropland only decreased slightly due to human disturbance minimizing the impact of climate change. In general, the regional GPP also showed decreasing trends. This result was consistent with that of other studies. Zhao and Running (2010) explored the temperal and spatial variations of NPP in global terrestrial ecosystems during 2001 -2009 using MODIS products. They pointed out that GPP in some parts of Southern China exhibited a decreasing tendency. Liu (2013) analyzed the temporal and spatial variations of NPP in terrestrial ecosystems in China during 2000 -2010 using the BEPS model. Their results showed that NPP in terrestrial ecosystems in the YRD region also E a r l y R e l e a s e showed a decreasing trend. This was mainly caused by global climate change, the frequency of summer droughts has increased in this region, and severe summer droughts have led to a decline in vegetation productivity (Zhao and Running 2010;Wang et al. 2016). Along with freezing rain and ice storms, GPP in the YRD region has gradually shown a decreasing trend (Xu et al. 2013). Figure 4 illustrates the spatial distribution of GPP during the study period. The value of GPP was higher in the south and northwestern YRD than in any other part of the YRD region. In general, GPP was higher in the south (Fig. 5). This is likely to be due to the different land-cover types. In the south, there is more forest, while in the north the land-cover type is almost entirely cropland . Forest ecosystems can maintain a high photosynthetic capacity for a long time in a year compared to cropland ecosystems, and as a result, the forest ecosystems fix more carbon dioxide.

temporal and spatial Variations of Et
During the study period, ET for all vegetation types in the YRD region showed different trends (Fig. 6). ET in forest ecosystems all showed an increasing tendency, with evergreen broadleaf forest increasing the most significantly. In contrast, ET in cropland and grassland both showed a decreasing trend, with the latter having a higher decreasing rate. In general, the regional ET showed an increasing trend.
By far, the results of the variation trend of ET in the YRD region were inconsistent. In the study of Liu (2013) Wang et al. (2015) suggested that ET in terrestrial ecosystems in the YRD region from 2001 -2012 also showed a dereasing trend. The discrepancy in the results of the variation trend of ET was likely caused by an inconsistency in the research method, study period, study area and the driving forces. This indicates that more studies need to be conducted by modifying the model, adding the ground-based measurement and improving the accuracy of the driving data. For the spatial distribution of ET, the maximum value of ET occurred in the south and northeastern part of the YRD region (Fig. 7).     E a r l y R e l e a s e

temporal and spatial Variations of WuE
In this study, WUE in all forest ecosystems showed a decreasing tendency. In contrast, WUE in cropland and grassland both showed an increasing trend. Among the biomes, evergreen broadleaf forest and grassland achieved a significant level (p < 0.05). In general, the regional WUE showed a decreasing trend (Fig. 8). As WUE was controlled by both GPP and ET, when GPP decreased and ET increased, the WUE inevitably decreased. This was the situation for evergreen broadleaf forest, deciduous needle leaf forest, and deciduous broadleaf forest. The increase in WUE in grassland was due to the decreased ET, for which the rate of decrease was greater than for GPP.
For the spatial distribution of WUE, the maximum value occurred in the south and northwestern part of the YRD region, with WUE greater than 2.0 g C kg -1 H 2 O. The minimum value was distributed in the middle and northwest of the YRD region, and was smaller than 1.6 g C kg -1 H 2 O (Fig. 9).

the Annual Average WuE over Vegetation types
The annual average WUE over vegetation types ranged from 1.39 -2.17 g C kg -1 H 2 O (Fig. 10). The order was deciduous broadleaf forest > evergreen broadleaf forest > cropland > grassland > evergreen needle leaf forest > deciduous needle leaf forest. Yu et al. (2008) found that WUEs of needle leaf forest were greater than that of broadleaf forest. This difference was mainly caused by the different study scale. In a study by Yu et al. (2008), they focused on the variations of WUE across a transect where the climate conditions among the vegetation types were quite different. However, in our study, we tried to explore the characteristics of WUE at the regional scale where the climate conditions among the vegetation types were comparatively similar. In general, WUE of forest was greater than that of other types and this result was consistent with that of another study (Ito and Inatomi 2012).

the Effect of lAI on GPP and Et
LAI is an important biotic factor for carbon and water cycles, and understanding the response of GPP and ET to LAI can greatly enhance our knowledge of their control processes as well as the ability to predict how climate change may affect carbon and water budgets (Reichstein et al. 2007;Hu et al. 2008). In our study, GPP increased with increasing LAI across all of the plant function types (Fig. 11), and this pattern has also been proved by other researchers (Flanagan et al. 2002;Xu and Baldocchi 2004;Li et al. 2006;Hashimoto et al. 2012).
Different from GPP, ET only increased with LAI when LAI was less than 2 m 2 m -2 across all of the plant function types, once LAI was larger than 2 m 2 m -2 , ET did not continue to increase with LAI (Fig. 12). To a great degree, this response pattern was determined by the effect of LAI on transpiration. As the most important component of ET, the way in which transpiration varies with LAI, plays a big role in the effect of LAI on ET. Benyon et al. (2006) found that the increase of changing rate of transpiration slowed down with LAI, and there was a turning point in their relationship. When LAI reached that point, transpiration did not continue to increase with LAI, which is similar to the relationship between ET and LAI found in our study.

FutuRE WoRk AnD PRosPECt
In this study, we examined the temporal and spatial variations of GPP, ET, and WUE across plant function types using remote sensing models. It should be noted that a number of models have been developed to simulate the GPP, including the process-based models and the remote sensing models. Which one is superior has been the subject of much debate in recent decades. Recently, Zhang et al. (2016) demonstrated that the VPM model was more reliable than the process-based models based on the spatial distribution and seasonal dynamics of GPP simulated by the VPM model, and sun-induced chlorophyll fluorescence showed better consistency. This enhanced our confidence in using the VPM model. Potential data being used as a benchmark to compare with ET models could also benefit the study of the simulation of ET. The change tendency of GPP, ET, and WUE across plant function types were analyzed in the present study, but further studies are needed to establish the underlying reasons and quantify their respective rates of contribution and finally to predict how climate change may affect carbon and water budgets.

ConClusIon
In this study, the vegetation photosynthesis model (VPM) and the revised remote sensing-Penman Monteith (RRS-PM) model were used to characterize the spatialtemporal dynamics of gross primary productivity (GPP) and evapotranspiration (ET) in the Yangtze River Delta (YRD) region. Our predicted GPP and ET results were in the range of those reported by other models, which enhanced confidence in using the remote sensing models. The change tendency of GPP, ET, and WUE across plant function types was different in the present study, so further studies will be needed to establish whether this is determined by climatic factors or other adjustment mechanisms. In the current study, the variation of GPP, ET, and WUE had only been monitored in the last decade or so. It is important for the change tendency to be analyzed in a sustained way in order to gain a better understanding of the impact of climate change. This study demonstrates that LAI is an important biotic factor for carbon and water cycles. Studies using LAI as an input variable to estimate carbon and water fluxes should pay more attention to the quality and accuracy of LAI.