Responses of land evapotranspiration to Earth’s greening in CMIP5 Earth System Models

Satellite-observed Earth’s greening has been reproduced by the latest generation of Earth System Models (ESMs) participating in the Coupled Model Intercomparison Project Phase 5 (CMIP5). Land evapotranspiration (ET) is expected to rise with increasing leaf area index (LAI, Earth’s greening). The responses of ET play a key role in the land–climate interaction, but they have not been evaluated previously. Here, we assessed the responses of ET to Earth’s greening in these CMIP5 ESMs. We verified a significant and positive response of ET to the modeled greening in each model. However, the responses were not comparable across the ESMs because of an inherent bias in the sensitivity of ET to LAI ( ∂ E T / ∂ L A I ) in the models: ∂ E T / ∂ L A I is precisely and inversely proportional to the trend of LAI ( ∂ L A I / ∂ t ) across the ESMs. Constrained by this inversely proportional relationship with the satellite-observed ∂ L A I / ∂ t , the Earth’s ∂ E T / ∂ LAI is 0.26 (0.21–0.34) mm d−1 per m2 m−2, equaling the independent estimates from satellite-derived reconstructions of ET and LAI. Thus, the Earth’s greening-induced acceleration of ET is about 11.4 mm yr−1, accounting for more than 50% of the observed increase in land ET over the last 30 years. To better model the land–climate interaction, ∂ E T / ∂ L A I in these ESMs should be calibrated. A feasible means is to improve the representation of the magnitude of LAI in these CMIP5 ESMs.


Introduction
The greening of the Earth over the last three decades has been documented by several studies based on the National Oceanic and Atmospheric Administration-Advanced Very High Resolution Radiometer (NOAA-  Mahowald et al 2015). Furthermore, these ESMs also unequivocally and consistently project continuation of Earth's greening, at least until the end of the 21st century (Mahowald et al 2015). As the change in land surface properties has a profound impact on land-atmosphere exchanges of water and energy, and ultimately on the climate system, the modeled Earth's greening should incorporate boundary forcing in these climate model simulations.
The key flux determining the strength of the greening-induced boundary forcing is terrestrial evapotranspiration (ET). ET, as a central process in the climate system, represents the exchanges of energy and water between the land and the atmosphere. As for its driver, because transpiration through vegetation crown dominates terrestrial ET (Jasechko et al 2013), the greening of the Earth has been one of the most important drivers in the rise of global land ET over the past 30 years (Zhang et al 2015). As for its effect, terrestrial ET plays a fundamental role in shaping the climate. It cools the land surface temperature by consuming more than half of the solar radiation absorbed by the land surface (Trenberth et al 2009), and drives atmospheric dynamics by the released latent heat during condensation (Makarieva et al 2013). Therefore, the greening of the Earth would be expected to reshape the Earth's climate (e.g., rainfall, temperature, and circulation) by producing evapotranspiration (Shukla and Mintz 1982, Bounoua et al 2000, Sewall et al 2000, Buermann et al 2001.
In the ESMs, the response of ET to the modeled Earth's greening determines the strength of vegetation feedback in the land-climate interaction. Assuming that all CMIP5 ESMs simulated the sensitivity of ET to LAI ¶ ¶ ( ) ET LAI following the laws of nature, ¶ ¶ ET LAI in these ESMs should be close to the Earth's ¶ ¶ ET LAI, resulting in a tendency to be constant across the models. Terrestrial ET increased more in the ESMs with a higher greening rate ¶ ¶ ( ) t LAI ; thus a stronger evaporative cooling effect and faster moisture recycling were modeled. That is, greeninginduced boundary forcing was underestimated (or overestimated) by ESMs that modeled a weaker (or stronger) greening rate than the satellite-observed greening rate. This implies that, to better model the land-climate interaction in the ESMs, the modeling community should focus on improving the simulation of vegetation dynamics (i.e., ¶ ¶t LAI ). To our knowledge, however, no evaluation of the responses of ET to the Earth's greening in these CMIP5 ESMs has been performed. If biases exist in the modeled ¶ ¶ ET LAI, the greening-induced boundary forcing would not be comparable across the ESMs. In this case, it is important to understand why the modeled ¶ ¶ ET LAI differs from the ESMs. In addition, the modeling community should pay more attention to calibration of the representation of the sensitivity of ET to greening (i.e., ¶ ¶ ET LAI). In this study, we assessed the responses of ET to Earth's greening in the CMIP5 ESMs by integrating the historical simulations from CMIP5 ESMs with satellite-derived reconstructions of ET and LAI over the last 30 years. The objective was to address the following questions (1) is there a significant and positive response of land ET to Earth's greening in each ESM? A positive answer to this question would verify that land LAI is an important driver of the interannual variation of land ET in the ESMs.
(2) Is ¶ ¶ ET LAI constant across the ESMs? If the answer is positive, we would expect a greater increase in land ET in the ESMs with a higher greening rate. A negative answer would prompt us to ask, were chosen based on their data availability. The outputs of both LAI (in m 2 m −2 ) and ET (in mm d −1 ) in the historical simulations of these ESMs were downloaded from the CMIP5 archive (http://pcmdi9. llnl.gov/). The annual area-weighted global landaverage LAI and ET during 1982-2005 were calculated using the average of all initial condition ensemble members available in the archive.

Observational datasets
The long-term NOAA-AVHRR satellite measurements were used to generate an 8 km global LAI product from 1982 to 2011 (AVHRR GIMMS LAI3g) (Zhu et al 2013). The quality of the satellite LAI product used in this study has been extensively evaluated through the comparisons with field measurements, with an accuracy of about 0.68 m 2 m −2 (root mean square error) relative to field measured LAI  table 2). All these products have been proven of quality for scientific researches. Yet, because of the lack of direct global observations, it is hard to estimate the biases in these products and to tell which product is superior to the others. Thus, all these products were applied to calculate the annual area-weighted global land-average ET of the last three decades, with the difference among them representing an uncertainty range in the observed land ET. The Earth's sensitivity of ET to LAI ¶ ¶ ( ) ET LAI was thus estimated with these observed land ET and LAI over the last 30 years.

Results and discussion
3.1. Response of land ET to Earth's greening in each CMIP5 ESM The magnitude of land LAI found among the CMIP5 ESMs covered a large spectrum, and most models overestimated the magnitude of land LAI compared to observations from the AVHRR satellites (figure 1(a)), probably due to the systematic underestimate of NPP in these models (Shao et al 2013, Anav et al 2013b). Despite this, the greening of the Earth, defined as a significant and positive trend of land LAI observed from the AVHRR satellites for the last 30 years (e.g., Zhu et al 2013, 2016, Piao et al 2015, was reproduced by 16/27 CMIP5 ESMs ( figure 1(b)). For the other 11/ 27 ESMs, land LAI either did not vary annually (ACCESS1-0, ACCESS1-3, FIO-ESM), decreased unreasonably by a constant rate each year (MIROC5, figure S1), or did not change significantly over the last 30 years (CESM1-WACCM, CanESM2, GFDL-ESM2M, MIROC-ESM, MPI-ESM-LR, NorESM1-ME, inmcm4). Land LAI, if not fixed in models, is a key driving factor of interannual variation of land ET in all the models (figures 1(c) and S2). As the goal of this study is to investigate the responses of land ET to Earth's greening in CMIP5 ESMs, we analyzed the responses of land ET to increasing LAI in the 16 CMIP5 ESMs that successfully reproduced the Earth's greening of the last 30 years (figure 1(c)).
As shown in figure 1(c), there was a significant interannual correlation between the modeled land LAI and ET for all 16 ESMs that reproduced the Earth's greening. The strongest correlation between the modeled land LAI and ET was found in GFDL-CM3 (figure 1(c5), R=0.81, P<0.01), and the weakest correlation was found in CESM1-CAM5 (figure 1(c4), R=0.35, P<0.1). Consistent with the significant correlation, land ET increased significantly with  1(c)). That is, land ET responded positively to the modeled Earth's greening in all of the models. Land LAI indeed is a key driver of the interannual variability in land ET in each ESM (e.g., Zhang et al 2015).
In theory, LAI is one of the key parameters of land ET which is also co-determined by factors like soil moisture supply, solar radiation, and wind speed. LAI could change land ET by its role in the regulations of the surface area of vegetation in direct contact with the atmosphere and thus the efficiency by which water can be transferred from within the vegetation to the atmosphere (e.g., canopy conductance ) g , c surface radiation budget (e.g., albedo, and radiation partitioning between canopy and soil), boundary layer aerodynamic characteristics (e.g., aerodynamic conductance), and redistribution of rainfall (e.g., interception and through fall). Among them, regulating the canopy conductance ( ) g c has been suggested as a dominant one by dozens of studies where i , 1 i , 2 and j are parameters that are dependent on aerodynamic roughness and vegetation type. Figure 2 shows the curves of equations (1) and (2) with = i 1, and j=1, which clearly demonstrates that, in each ESM, land ET should increase with LAI.

Responses of land ET to Earth's greening across CMIP5 ESMs
As 1) land ET responds to the modeled greening in each ESM (figures 1(c)) and 2) the greening rate (i.e., ¶ ¶ ) t LAI differs among the models ( figure 1(b)), the increase in land ET ¶ ¶ ( ) t ET is expected to be greater in the ESMs with a higher greening rate, i.e., ¶ ¶ µ ¶ ¶ t t ET LAI . However, as shown in figure 3, there is no correlation between the modeled ¶ ¶t ET and   figure 4).
across the ESMs tends to be constant, as shown in figure 3. This explains the lack of correlation between ¶ ¶ ET LAI and ¶ ¶t LAI across the ESMs (figure 3). We then investigated why ¶ ¶ ET LAI is inversely proportional to ¶ ¶t LAI across the ESMs. In theory, as shown in figure 2, while ET increases with LAI in each ESM, ¶ ¶ ET LAI decreases with the magnitude of LAI across the ESMs, which is also shown in the partial derivatives to LAI for equations (1) and (2) Thus, the inherent bias of ¶ ¶ ET LAI is primarily due to the bias in the magnitude of LAI across the CIMP5 ESMs ( figure 1(a)). In addition, the bias in the magnitude of LAI is also responsible for the difference in ¶ ¶t LAI across the models: ¶ ¶t LAI is significantly proportional to the magnitude of land LAI across the models (P<0.01; figure 5). As a result, ¶ ¶ ET LAI is inversely proportional to ¶ ¶t LAI across the CMIP5 ESMs ( figure 4).

The Earth's sensitivity of land ET to land LAI
If all the ESMs modeled the land-climate interaction well, the modeled sensitivity of land ET to land LAI should be almost constant across the models, with the constant being the Earth's ¶ ¶ ET LAI. However, due to the bias in the magnitude of modeled LAI ( figure 1(a)), there are biases in the modeled ¶ ¶ ET LAI in the CMIP5 ESMs. To better model the land-climate interaction, ¶ ¶ ET LAI in these ESMs should be calibrated to equal the Earth's ¶ ¶ ET LAI.  Here, we further applied two approaches to provide a reference for the Earth's ¶ ¶ ET LAI making use of model simulations from CMIP5 and satellite observations of ET and LAI over the last 30 years.
First, the Earth's ¶ ¶ ET LAI can be estimated by the observational constraints on the precise inversely proportional relationship between the modeled ¶ ¶ ET LAI and ¶ ¶t LAI across the CMIP5 ESMs. As  where P and T are the observed annual precipitation and annual average temperature from the CRU dataset, respectively (Harris et al 2014), and k 2 is the sensitivity controlling precipitation and temperature. In this approach, the observed ¶ ¶ ET LAI ranges from 0.08 to 0.39 mm d −1 per m 2 m −2 depending on ET products (green bars in figure 6). The ensemble of the observed ¶ ¶ ET LAI controlling precipitation and temperature provides the optimal estimate of the Earth's ¶ ¶ ET LAI, namely 0.29 (0.25-0.33) mm d −1 per m 2 m −2 (dark green bar in figure 6).
Thus, the two independent estimates of the Earth's ¶ ¶t LAI match very well with each other. Using the sensitivity estimated by the observational constraints, the satellite-observed Earth's greening (i.e., the increasing LAI at a rate of 0.04 ± 0.01 m 2 m −2 per decade) has accelerated land ET by 11.4 mm yr −1 in the past 30 years. The total increase in land ET during the last 30 years is 16.3±6.9 mm yr −1 according to the CMIP5 ESMs, and ranges from 13.1 to 51.2 mm yr −1 according to the satellite-derived reconstructions of ET (average: 26.6 mm yr −1 ), depending on the model and/or ET dataset used as a reference. Therefore, the Earth's greening-induced acceleration of ET has contributed more than 50% of the observed increase in land ET over the last 30 years. The greater capacity of water loss associated with increasing LAI has become a dominant driver of the increasing land ET in the past 30 years.

Implications for model improvement
Considering the key role of ET in the water cycle and energy fluxes, the biophysical feedback of vegetation growth activity should play an important role in shaping the climate. However, in the CMIP5 ESMs, the biophysical feedback of vegetation growth activity has not been represented well due to the biases in the greening rate ¶ ¶ ( ) t LAI and the sensitivity of ET to greening ¶ ¶ ( ) ET LAI . The modeling community should calibrate the modeled ¶ ¶ ET LAI to make it equivalent to the Earth's ¶ ¶ ET LAI (∼0.26 mm d −1 per m 2 m −2 ). As the biases of both ¶ ¶t LAI and ¶ ¶ ET LAI are primarily caused by the modeled bias of the magnitude of land LAI, feasible and effective methods for improving the representation of vegetation-climate feedback are therefore required to improve the representation of the magnitude of LAI in these state-of-the-art ESMs.
In addition, we found that magnitude of global land LAI of four ESMs was similar to the value observed by satellite (MPI-ESM-MR, IPSL-CM5B-LR, Figure 6. The Earth's sensitivity of land ET to land LAI estimated from the observational constraint on the CMIP5 ESMs (red bar), and the linear regressions of satellite-derived reconstructions of ET and LAI over the last 30 years (blue and green bars). For the sensitivity from the observational constraint, the error bar shows the range constrained by the mean± standard error (SE) of the satelliteobserved trend in global land LAI. For the sensitivities of the regressions, *** , P<0.01; n.s., P > 0.05 for sensitivity, and error bars show the standard errors of the sensitivities. IPSL-CM5A-LR, IPSL-CM5A-MR, see figure 5; the observed magnitude is 1.5 m 2 m −2 ). However, all of these models still underestimated the trend in global land LAI compared to the satellite-observed trend (i.e., 0.04±0.01 m 2 m −2 per decade). Among these four models, the closer the modeled ¶ ¶t LAI is to the satellite-observed trend, the closer the modeled ¶ ¶ ET LAI is to the Earth's ¶ ¶ ET LAI ( figure 4(a)). Therefore, another work at a next stage for the modeling community is to investigate the mechanisms driving the temporal changes in land LAI (e.g., Piao et al 2015, Zhu et al 2016) and then improve the representation of temporal variation of LAI in the ESMs.

Conclusions
Our results demonstrated a significant and positive response of land ET to increasing LAI in all of the 16 CMIP5 ESMs that reproduced the Earth's greening for the last three decades. However, the responses of land ET to the modeled greening are not comparable across the ESMs due to an inherent bias in the modeled ¶ ¶ ET LAI: ¶ ¶ ET LAI is precisely and inversely proportional to ¶ ¶t LAI across the ESMs. Furthermore, this inherent bias in the modeled sensitivity was found to be primarily due to the bias in the magnitude of LAI. The bias in the modeled ¶ ¶ ET LAI indicates that greening-induced biophysical feedback has not been represented well in these ESMs. Thus, it is necessary to improve the representation of the magnitude of LAI, which is an easy, feasible, and effective means for an ESM to calibrate the response of land ET to greening and thus to better represent the climate effect of Earth's greening in the model.
We estimated the Earth's ¶ ¶ ET LAI with two independent approaches, including the observational constraints on the precise inversely proportional relationship between the modeled ¶ ¶ ET LAI and ¶ ¶t LAI across the CMIP5 ESMs, and linear regression of the satellite-derived reconstructions of ET and LAI during the last 30 years. The suggested sensitivity of land ET to land LAI in the Earth's climate is 0.26 mm d −1 per m 2 m −2 . With this sensitivity, the satellite-observed Earth's greening can be translated into acceleration of land ET by a rate of 3.8 mm yr −1 per decade, accounting for more than 50% of the observed acceleration rate of land ET over the last 30 years. As the response of land ET deeply affects the climate (water cycle and energy fluxes), it will be important to investigate the climate effect of Earth's greening by combining satellite-derived observations and the state-of-the-art ESMs. Furthermore, because the biophysical feedback induced by the response of land ET is dominated over the regions where vegetation has changed, the understanding of the climate effect of Earth's greening would improve future projections of regional climate change and benefit local policy decisions.