Future changes of global potential evapotranspiration simulated from CMIP5 to CMIP6 models

ABSTRACT

This research evaluated the ability of different coupled climate models to simulate the historical variability of potential evapotranspiration (PET) for the time period 1979–2017 in phases 5 and 6 of the Coupled Model Intercomparison Project (CMIP5 and CMIP6, respectively). Their projected future changes of PET under two emission scenarios for the 21st century were also compared. Results show that PET has an increasing trend of 0.2–0.6 mm d⁻¹/50 yr over most land surfaces and that there are clear regional differences. The future value of PET is higher in the CMIP6 multi-model simulations than in the CMIP5 ones under the same emissions scenario, possibly because CMIP6 models simulate stronger warming for a given forcing or scenario. The contributions of each individual climate driver to future changes in PET were examined and revealed that the surface vapor pressure deficit makes a major contribution to changes in PET. Shortwave radiation increases PET in most terrestrial regions, except for northern Africa, East Asia, South Asia, and Australia; the effect of longwave radiation is the opposite to that of shortwave radiation. The contribution of surface wind speed to PET is small, but results in a slight reduction.

摘要

为了评估国际耦合模式比较计划第6阶段(CMIP6)的模拟能力, 本研究选用CMIP5和CMIP6两代气候模式对潜在蒸散发进行对比分析, 并比较了21世纪潜在蒸散发的未来变化及各驱动因素的相对贡献。结果表明: CMIP5与CMIP6模式均可以较好地模拟出潜在蒸散发的增加趋势。基于相同排放情景下, CMIP6多模式结果模拟的未来潜在蒸散发量值将高于CMIP5, 这可能是CMIP6模式体现出了比CMIP5模式更强的升温效应。表层水汽压差仍是导致未来潜在蒸散发变化的主要驱动因子。

KEYWORDS:

• Potential evapotranspiration
• simulation evaluation
• contribution analysis
• CMIP5
• CMIP6

关键词:

• 潜在蒸散发
• 模拟评估
• 贡献分析
• CMIP5
• CMIP6

1. Introduction

Potential evapotranspiration (PET), a basic land climate variable (Hartmann 1994), is a measure of the atmospheric demand for evaporation and is independent of the supply of water. PET has been widely used to characterize environmental, hydrological, and global changes (Burke, Brown, and Christidis 2006; Dai 2011a; Feng and Fu 2013; Cook et al. 2014; Scheff and Frierson 2014; Dai, Zhao, and Chen 2018). It is defined as the amount of water that would potentially be removed from a vegetated surface through the processes of evaporation or transpiration with no forcing other than atmospheric demand (Allen et al. 1998; Yoder, Odhiambo, and Wright 2005). As such, a higher PET value represents more arid, evaporative conditions. PET can be used to calculate a variety of aridity, drought, and soil moisture indices (Burke, Brown, and Christidis 2006; Cook et al. 2014; Dai 2011b, 2013; Zhao and Dai 2015), as well as to prepare input data for hydrological models (Aouissi et al. 2016; Li, Zheng, and Liu 2012). Unbiased estimates of changes in PET are especially important in these frameworks to understand the hydroclimatic changes of the land surface and the impacts of climate variability on terrestrial systems (Kim and Hogue 2008).

Various studies have assessed the performance of general circulation models in the prediction of hydroclimatic variables in terms of individual dimensional quantities such as precipitation (Scheff and Frierson 2012; Dai and Zhao 2017), evapotranspiration (Wang and Dickinson 2012; Dong and Dai 2017), runoff (Taylor et al. 2013; Koirala et al. 2014), and soil moisture (Wang 2005; Sheffield and Wood 2008), or in terms of complex metrics of local drought relative to some reference period, such as the Palmer drought severity index (Zhao and Dai 2017). By contrast, few studies have compared the performance of the PET produced in general circulation models where the PET is derived from ground-based observations. Some studies have used the outputs of phase 5 of the Coupled Model Intercomparison Project (CMIP5) to project future changes in PET at the regional or global scale. Based on CMIP5 multi-model projections, Wang, Chen, and Zhou (2014) showed that an increase in PET could outweigh an increase in precipitation, resulting in intensified droughts in southwestern China. Scheff and Frierson (2014) used outputs from 13 CMIP5 models and reported that the percentage change in the local annual mean PET during the 21st century is almost always positive and usually increases with latitude, but is divergent between models.

Improved climate model simulations under the sixth phase of CMIP (CMIP6) have now been released (Eyring et al. 2016; Stouffer et al. 2016; Checa-Garcia et al. 2018). The future climate projections in CMIP6 use improved emissions, land use scenarios, model parameterizations, and physical processes driven by scenarios based on shared socioeconomic pathways (O’Neill et al. 2015; Eyring et al. 2016; Riahi et al. 2017). A comprehensive evaluation of the historical and future changes in PET has not yet been performed and is the primary motivation for this study.

The aim of this study was to assess the performance of two generations of climate models (CMIP5 and CMIP6) and to compare their ability to simulate historical changes in PET. Future projections under two scenarios (RCP4.5 vs. SSP2.45 and RCP8.5 vs. SSP5.85) were also compared. Section 2 describes the data and methods, and section 3 presents the global historical and future changes in PET. A summary and conclusions are given in section 4.

2. Data and methods

2.1 Data

The observational dataset used in this study is the CRU-TS4.02 monthly PET dataset from the University of East Anglia (UK), which has a spatial resolution of 0.5° × 0.5°; more details about these data are given in Harris et al. (2014). Also used were monthly reanalysis data for near-surface air temperature, near-surface dewpoint temperature, net solar radiation at the surface, net thermal radiation at the surface, surface air pressure, and wind speed at 10 m, from the global European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis dataset, to calculate the reanalyzed PET with the Penman–Monteith equation. These reanalysis data span the time period 1979–2017 with a spatial resolution of 1° × 1°.

The model data were compiled to calculate the PET separately from each of 16 CMIP5 models and 7 CMIP6 models (see Table S1 for the CMIP5 models and Table S2 for the CMIP6 models). The all-forcing historical simulations from 1900 to the early 21st century (2005 and 2014 for the CMIP5 and CMIP6 models, respectively) and the future simulations under low-to-moderate emission scenarios (RCP4.5 and SSP2.45) and high emission scenarios (RCP8.5 and SSP5.85) for the CMIP5 and CMIP6 models were used separately in this study; more details are given in Taylor, Stouffer, and Meehl (2012) and Eyring et al. (2016), respectively. The PET values for individual models were averaged over all the models to create a multi-model ensemble mean for the CMIP5 and CMIP6 models.

All the PET data were re-mapped onto a common 1° × 1° grid over global land areas from 60°S to 75°N. The ERA5 reanalysis dataset does not include data before 1979; thus, to better discuss the long-term changes in PET, this paper uses the estimated results under the low-to-moderate emission scenarios (RCP4.5 and SSP2.45) to supplement the historical simulation data to 2017, and sets the historical study period to be the same as the observation data and reanalysis data from 1979 to 2017.

The Penman–Monteith equation is used to quantitatively analyze the changes in PET caused by driving factors and to allow the contributions of these individual factors driving the PET to be further analyzed. The method involves changing one driving factor of PET with time whilst the values of the remaining driving factors remain fixed in 1979. The resulting PET is therefore only affected by a single driving factor.

2.2 PET formula

In this study, the equation for calculating PET used by the Penman–Monteith formula (Shuttleworth 1993) is as follows:

(1) $\mathrm{P}\mathrm{E}{\mathrm{T}}_{\mathrm{p}\mathrm{m}}=\frac{\mathrm{\Delta }}{\mathrm{\Delta }+\gamma }\left(Rn+A_{\mathrm{h}}\right)+\frac{\gamma }{\mathrm{\Delta }+\gamma }\frac{6.43\left(1+0.536U_2\right)D}{\lambda }$(1)

where Rn is the surface net radiation in mm d⁻¹, A_h is the surface horizontal energy convergence, which is ignored here, U₂ is the surface wind speed at 2 m height in m s⁻¹, D is the surface vapor pressure deficit in kPa, ∆ is the slope of the saturated water vapor pressure–temperature curve in kPa/°C, γ is the coefficient in kPa/°C and γ is the latent heat of water vapor in MJ kg⁻¹. The Penman–Monteith PET has been shown to perform better than other formulas for calculating PET in some studies (Donohue, Mcvicar, and Roderick 2010; Dai 2011b).

3. Results

3.1 Historical changes in PET

Figure 1 shows the spatial distribution of the linear trends of the annual PET during 1979–2017 using the observational dataset, the ERA5 reanalysis dataset, and the CMIP5 and CMIP6 multi-model all-forcing simulations. There is a significant increasing trend of 0.2–0.6 mm d⁻¹/50 yr over many terrestrial areas in the observational dataset and the ERA5 reanalysis dataset, including southern North America, southern Europe, Australia, the Sahel and southern Africa, and parts of Asia (Figure 1(a,b)). By contrast, there is a weak downward trend of PET in South Asia and Central Africa.

Figure 1. Spatial distribution of the linear trend of annual mean PET during 1979–2017 (units: mm d⁻¹/50 yr). The pattern correlation (r) is the linear trend of (a) the CRU observational data with (b) the ERA5 dataset, (c) the CMIP5 simulation, and (d) the CMIP6 simulation. The shading indicates that the area passed the 0.05 significance test

In general, the ERA5 reanalysis dataset and the two multi-model ensemble means capture the spatial distribution of the linear trend of PET. However, there is an overall poor ability of the models and reanalysis to reproduce the wetting trend in North America, particularly at higher latitudes (Figure 1(b–d)). The problem here is that the trend analyzed is subject to uncertainties in the observations, the complications of natural variability in the real world and models, and uncertainties in feedbacks and how they may change in the future (Räisänen 2007; Knutti 2010).

Despite the large regional differences, the leading modes of the PET from the CRU observational dataset, the ERA5 reanalysis dataset, and the CMIP5 and CMIP6 model simulations, all show an increasing trend over most of the land area (Figure 2). The increasing trend of PET in southern North America, southern Europe, the Sahel and southern Africa, and the decreasing trend in southern Asia, are detected well, but the magnitude of the change in PET in East Asia and Australia is not consistent among the different datasets. The drying trend over southern Asia in the CMIP6 models is much closer to the observational data and ERA5 data than CMIP5. However, the CMIP6 models show a significant wetting trend in southern Asia that is absent in the observational dataset and the CMIP5 models (Figures 1(a,c,d) and 2(a,c,d)). Overall, there is an increasing trend of PET worldwide and the performance of each dataset is consistent (Figure 2(e)).

Figure 2. Leading empirical orthogonal function (EOF) of the monthly global terrestrial PET anomalies during 1979–2017. The EPV is the explained percentage of the total variance in (a-d), and the value of EOF multiplied by 100. The pattern correlation (R) is the leading EOF of (a) the CRU observational data with (b) the ERA5 dataset, (c) the CMIP5 simulation, and (d) the CMIP6 simulation. (e) Nine-point moving average principal component (PC) and correlation coefficient (R1–R3) between the CRU observational dataset and other datasets

3.2 Simulated future changes in PET

Figure 3 shows the time series of long-term changes in terrestrial PET. The changes in PET in the Northern Hemisphere (Figure 3(c,d)) are similar to the global range (Figure 3(a,b)), and the changes in the Southern Hemisphere (Figure 3(e,f)) are larger than others as a result of the smaller land area in the Southern Hemisphere. In the historical stage, the observational CRU dataset, the ERA5 reanalysis dataset, and the two model simulations, all show a relatively consistent increasing trend, with a global increase of about 0.2 mm d⁻¹ from 1979 to 2017; and among these datasets, ERA5 shows a greater increase in the average PET. In the future, the increase in PET estimated by the CMIP5 and CMIP6 models under the RCP4.5 and SSP2.45 low-to-moderate emission scenarios (Figure 3(a,c,e) will stabilize after the 2060s, but will continue to increase under the RCP8.5 and SSP5.85 high emission scenarios (Figure 3(b,d,f)). The CMIP6 models produce larger changes than the CMIP5 models after about the 2040s for the same emissions scenario. Since PET is most sensitive to perturbations in temperature (Guo, Westra, and Maier 2017), it may be associated with the CMIP6 models simulating stronger warming for a given forcing or scenario (Forster et al. 2020).

Figure 3. Time series of the interannual global terrestrial PET anomalies for the CRU observational dataset (black) from 1979 to 2017, the ERA5 reanalysis dataset (green) from 1979 to 2017, and the CMIP5 (blue) and CMIP6 (red) ensemble means from 1979 to 2099 (units: mm d⁻¹) under (a, c, e) low-to-moderate emission scenarios and (b, d, f) high emission scenarios, for (a, b) global land (60°S–75°N), (c, d) Northern Hemisphere land (0°–75°N), and (e, f) Southern Hemisphere land (60°S–0°). The shaded areas indicate the ranges of the changes in the results of different models of CMIP5 and CMIP6

Figure S1 shows the change in PET for both the CMIP5 and CMIP6 multi-model simulations under different future emission scenarios. The PET increases by 0.2–0.6 mm d⁻¹ in the low-to-moderate emission scenarios and increases by more than 0.8 mm d⁻¹ in the high emission scenarios over southern North America, southern Europe, the Sahel, western Australia, and other regions. Overall, this shows the same increasing trend as in Figure 3(a,b).

3.3 Contribution analysis of PET

Figure 4 shows the spatial distribution of the contribution of each driving factor to PET under different emission scenarios. The change in PET due to surface vapor pressure deficit (VPD) (Figure 4(a–d)) shows a uniform increasing trend of greater than 0.8 mm d⁻¹ in southern North America, central and eastern South America, the Sahel, southern Africa, southern Europe, and western Australia. The change in PET due to net shortwave radiation (Rn_s) (Figure 4(e–h)) shows a downward trend in the Sahel, the Indian peninsula, the Middle East, and western Australia; by contrast, there is an increasing trend in other land regions. The contribution of net longwave radiation (Rn_l) to PET (Figure 4(i–l)) is the opposite to that of ∆Rn_s. The surface wind speed (sfcWind) (Figure 4(m–p)) causes a slight increase in PET in South America, Africa, and southern Asia, but a slight increase in other terrestrial regions.

Figure 4. Spatial distribution of the annual change in PET caused by the (a–d) VPD, (e–h) Rn_s, (i–l) Rn_l, and (m–p) sfcWind, over land areas from 1979–99 to 2079–99 (units: mm d⁻¹) in the CMIP5 models under the RCP4.5 scenario (left-hand column) and the RCP8.5 scenario (left-middle column), the CMIP6 models under the SSP2.45 scenario (right-middle column), and the SSP5.85 scenario (right-hand column). Shading indicates that at least 80% of the models agree on the sign of the change

These changes in PET caused by various driving factors show a greater change in the high emission scenarios than in the low-to-moderate ones. Also, the future changes in PET caused by ∆VPD and ∆Rn_s in the CMIP6 multi-model ensemble mean are larger than those for CMIP5 under the same emissions scenario.

Figure S2 shows the time series of the global area mean PET change, as caused by different driving factors, for the CMIP5 and CMIP6 multi-model ensemble means under low-to-moderate and high emission scenarios. The actual change in PET is similar to that in Figure 3(a,b); the increase in PET caused by the change of ∆VPD can approach or exceed the actual change in PET, which means that the VPD is the most important driving factor for PET. The contribution of ∆VPD to PET in CMIP6 multi-model simulations is increased relative to the contribution in CMIP5. In the low-to-moderate emission scenarios, the impact of ∆Rn_s in PET in the late 21st century is larger than that of ∆Rn_l. By contrast, the influence of ∆Rn_l is larger than that of ∆Rn_s under the high emission scenarios. The contribution of ∆sfcWind to PET is small and slightly reduced. The surface air pressure (Ps) has almost no effect and can be ignored.

4. Summary

We compared historical change in PET from 1979 to 2017 simulated by CMIP5 and CMIP6 models with CRU observational data and ERA5 reanalysis data. Then, we examined the projected changes in PET from the CMIP5 and CMIP6 model simulations for the 21st century under low-to-moderate and high emission scenarios and performed a preliminary quantitative contribution analysis. The main findings can be summarized as follows.

The linear trend of PET from 1979 to 2017 increases by 0.2–0.6 mm d⁻¹/50 yr in many land areas, including southern North America, northern central South America, southern Europe, the Sahel, southern Africa, and Australia. There is a clear weakening trend in South Asia and Central Africa. This result is also shown in the leading mode. The change in PET in the historical period is fairly consistent in many global terrestrial regions.

The future increases in PET estimated by the CMIP5 and CMIP6 models under the low-to-moderate emission scenarios are projected to stabilize after the 2060s, but PET will continue to increase under the high emission scenarios. The CMIP6 models produce larger changes than the CMIP5 models after about the 2040s for the same emissions scenario, possibly associated with the CMIP6 models simulating stronger warming for a given forcing or scenario.

In the analysis of the contributions of different variables to PET, the surface vapor pressure deficit made a major contribution to the change in PET, with the result that the increase in PET approached or exceeded the actual change in PET. The main contribution of shortwave radiation is to increase PET in most terrestrial regions of the world, except northern Africa, East Asia, South Asia, and Australia. The effect of net longwave radiation is the opposite to that of net shortwave radiation. The contribution of surface wind speed to PET is small, but results in a slight reduction.

Supplementary material

Supplemental data for this article can be accessed here.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Key Research and Development Program of China [grant number 2018YFC1507704] and the National Natural Science Foundation of China [grant numbers 41675094 and 41975115].