Analysis of surface temperature bias over the Tibetan plateau in the CAS FGOALS-f3-L model

Yo Wu , , Yimin Liu , Jindong Li , , Qing Bo , Bin He , Lei Wng , , Xiocong Wng ,Jinxio Li

a State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences,Beijing, China

b University of Chinese Academy of Sciences, Beijing, China

Keywords:Tibetan Plateau Surface temperature FGOALS-f3-L Surface energy budget equation Cloud radiation

ABSTRACT The Chinese Academy of Sciences Flexible Global Ocean-Atmosphere-Land System atmospheric component model (FGOALS-f3-L) participated in Phase 6 of the Coupled Model Intercomparison Project, but its reproducibility of surface temperature ( T s ) over the Tibetan Plateau (TP) as a key climatically sensitive region remains unclear.This study evaluates the capability of FGOALS-f3-L in reproducing the climatological T s over the TP relative to the Climate Forecast System Reanalysis. The results show that FGOALS-f3-L can reasonably capture the spatial pattern of T s but underestimates the annual mean T s for the whole TP. The simulated T s for the whole TP shows a cold bias in winter and spring and a warm bias in summer and autumn. Further quantitative analysis based on the surface energy budget equation shows that the surface albedo feedback (SAF) term strongly contributes to the annual, winter, and spring mean cold bias in the western TP and to the warm bias in the eastern TP. Compared with the SAF term, the surface sensible and latent heat flux terms make nearly opposite contributions to the T s bias and considerably offset the bias due to the SAF term. The cloud radiative forcing term strongly contributes to the annual and seasonal mean weak cold bias in the eastern TP. The longwave radiation term associated with the overestimated water vapor content accounts for a large portion of the warm bias over the whole TP in summer and autumn. Improving land surface and cloud processes in FGOALS-f3-L is critical to reduce the T s bias over the TP.

1. Introduction

The Tibetan Plateau (TP) is the highest plateau in the world and has a substantial influence on regional and global weather and climate change ( Ye and Gao, 1979 ; Wu et al., 2007 ). In addition, the natural environment (e.g., vegetation, glaciers, and ice lakes) of the TP is sensitive and vulnerable to climate change ( Yang et al., 2014 ; You et al.,2020 ). The TP is a heat source for the atmosphere in spring and summer ( Yeh et al., 1957 ; Flohn, 1957 ), and its climatic effects are strongly related to its surface temperature (

T

), as a key thermal variable of the atmosphere, and its variation. The reproducibility of the spatial and temporal changes in

T

is generally regarded as an essential indicator of a climate model’s performance ( Flato et al., 2013 ; Yang et al., 2015 ). To understand the TP climatic roles and model biases, it is critical to evaluate the simulated

T

in present climate models.Most current models suffer from a considerable temperature bias at the surface and in the atmosphere across the globe.Park et al. (2014) found that the globally averaged annual mean

T

bias is approximately − 1.22 K in the Community Earth System Model.Yang et al. (2015) showed that the global mean

T

bias is 0.38 K (1.70 K)in January (July) in the Flexible Global Ocean-Atmosphere-Land System (FGOALS) spectral model. Liu et al. (2015) found that the global mean

T

bias is approximately − 1.39 K in the FGOALS grid model. These analyses of the temperature bias in model simulations were mainly conducted on a global scale. The Fifth Assessment Report of the Intergovernmental Panel on Climate Change reported that in many areas, the annual mean surface air temperature (at 2 m) bias in the Coupled Model Intercomparison Project Phase 5 (CMIP5) multimodel mean is approximately 2°C and that the bias is much larger in several locations, particularly in high elevation regions, including the TP ( Flato et al., 2013 ).Hu et al. (2014) pointed out that the CMIP5 models underestimate the annual mean temperature of the TP, with an average of − 2.3 °C, and the features of the cold bias are similar to the early climate model simulation results ( Jiang et al., 2005 ; Xu et al., 2007 ). Chen et al. (2017) suggested that improvements in snow cover parameterization, boundary layer, and hence surface turbulent fluxes may help reduce the cold bias over the TP simulated by CMIP5 atmospheric general circulation models(AGCMs). Hu et al. (2014) and Chen et al. (2017) focused on detailed local information over the TP, but they mainly attributed discrepancies to the whole TP. In fact, there are apparent differences between land surface parameters (sensible and latent heat, albedo, and vegetation) and meteorological characteristics (precipitation, cloud types, and cloud cover) between the western and eastern TP ( Ye and Gao, 1979 ;Yang et al., 2014 ; Fu et al., 2020 ). These features can obviously influence

T

over the TP. In addition,

T

has substantial seasonal variation,and the influencing factors are also different between the western and eastern TP. Hence, it is essential to consider the seasonality of the simulated

T

and the relevant differences between the western and eastern TP to identify the

T

bias in current models.The Chinese Academy of Sciences (CAS) Flexible Global Ocean-Atmosphere-Land System atmospheric component model (FGOALSf3-L) participated in CMIP6 experiments. Compared with previous versions, FGOALS-f3-L has a higher resolution and improved performance( Zhou et al., 2015 ; He et al., 2019 ). To investigate the performance of FGOALS-f3-L and understand the deviation of its physical parameterization, it is necessary to evaluate the simulation of

T

over the TP in this model. In this study, we aim to evaluate the capability of FGOALS-f3-L to simulate

T

over the TP from 1981 to 2010 relative to the Climate Forecast System Reanalysis (CFSR). We also conduct an attribution analysis of the annual and seasonal mean

T

bias through the surface energy budget equation. This study can help to further understand the general features of the

T

bias and the relative contributions to its bias over the TP, especially in the western and eastern TP .

2. Methodology

2.1. Model and Data

The CAS FGOALS-f3-L used in this study was developed at the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG) at the Institute of Atmospheric Physics (IAP) ( He et al., 2019 ; Bao et al., 2019 ) and completed the Atmospheric Model Intercomparison Project (AMIP) simulations in late 2018 ( He et al., 2019 ). In FGOALS-f3-L, longitude is divided into 384 grid cells, and latitude is divided into 192 grid cells, which has a horizontal resolution of approximately 1°×1°. In the vertical direction, the model uses hybrid coordinates over 32 layers, and the top layer is at 2.16 hPa. The unique advantage of FGOALS-f3-L is a scale-awareness scheme to resolve convective precipitation, in which convective and stratiform precipitation are calculated explicitly ( Bao and Li, 2020 ). For a more detailed description of the model, see He et al. (2019) .

Fig. 1. (a) The terrain height over the TP (units: m). Annual mean T s in (b)FGOALS-f3-L and (c) CFSR from 1981 to 2010 over the TP, in units of °C.

The reference

T

dataset is from the CFSR. The CFSR is a global highresolution dataset published by the US National Centers for Environmental Prediction (NCEP) in 2010 ( Saha et al., 2010 ), which has a horizontal resolution of 0.5°×0.5°. Wang and Zeng (2012) compared many reanalysis datasets with 63 meteorological stations across the TP. Therefore,the CFSR reanalysis datasets are applied, and the

T

bias is defined as the difference in

T

between FGOALS-f3-L and CRSR. The net cloud radiative forcing effect datasets are from the Clouds and Earth’s Radiant Energy Systems (CERES) Energy Balanced and Filled (EBAF) satellite data ( Loeb et al., 2018 ).

The reanalysis datasets and FGOALS-f3-L can provide all the required variables, which include the surface temperature, specific humidity, sensible heat, latent heat, shortwave radiation received by the surface, shortwave radiation reflected by the surface, longwave radiation received by the surface, longwave radiation emitted by the surface,clear-sky shortwave radiation, and clear-sky downward longwave radiation. CFSR reanalysis data were interpolated into the FGOALS-f3-L resolution (1.0°×1.25°) with the bilinear interpolation method. The analysis period for the FGOALS-f3-L and CFSR datasets is chosen as 1981-2010.The analysis period for the CERES EBAF satellite data is chosen as 2001-2010.

2.2. Method

Following Lu and Cai (2009) , the surface energy balance equation used in this study is written as:

Fig. 2. (a) Annual (ANN), (b) winter (December, January, and February (DJF)), (c) spring (March, April, and May (MAM)), (d) summer (June, July, and August(JJA)), and (e) autumn (September, October, and November (SON)) mean Ts difference between FGOALS-f3-L and the CFSR from 1981 to 2010, in units of °C.

where

Q

is the heat storage term;

S

and

S

are the surface downward and upward shortwave radiation, respectively;

α

is the surface albedo;

F

and

F

are the surface downward and upward longwave radiation,respectively; and

H

and LE are the surface sensible and latent heat flux,respectively. Next, we consider the difference ( Δ) between FGOALS-f3-L and CFSR over the TP. The rearranging Eq. (1) is written as:

If the surface emissivity is assumed to be one at all wavelengths:

where

σ

= 5.67 ×10W mKis the Stefan-Boltzmann constant and

T

is the surface temperature.

Cloud radiative forcing (CRF), defined as the difference between the net total sky and clear sky radiations at the surface, is used to represent the effects of clouds on climate ( Cess and Pottet, 1988 ;Ramanathan et al., 1989 ). To exclude surface albedo feedback, Lu and Cai (2009) modified the definition of surface cloud radiative forcing(CRFs):

where ( ⋅)= ( ⋅) − ( ⋅), ( ⋅) represents total sky conditions, ( ⋅)represents clear-sky conditions, and

̄α

represents the climatic mean surface albedo in CFSR in this study.Because the mathematical expression for the radiation emitted from the surface is not directly related to clouds, in Eq. (4) ,

F

,

= 0 .Eq. (4) is substituted into Eq. (2) . Furthermore, combined with Eq. (3) .

Finally:

Fig. 3. (a) Annual mean T s difference between FGOALS-f3-L and the CFSR from 1981 to 2010 due to (b) SAF, (c) CRFs, (d) non-SAF associated with clear-sky SW radiation, (e) net clear-sky LW radiation fluxes, (f) surface sensible and latent flux ( H + LE), and (g) heat storage ( Q ), and (h) the sum of (b) to (g), in units of °C.

3. Distribution of annual and seasonal mean T s bias

Fig. 1 shows the topography and the annual mean

T

in FGOALS-f3-L and CFSR over the TP, respectively. As shown in Fig. 1 (b-c), the spatial patterns of

T

over the TP in FGOALS-f3-L and CFSR are similar. A low

T

center occurs in the northwestern TP, where the altitude is high. A high

T

center occurs on the southeastern slope of the Himalayas. The annual mean

T

pattern correlation (～25°-40°N, ～70°-105°E) between FGOALS-f3-L and the CFSR is 0.98, and the seasonal mean

T

pattern correlations are also over 0.95. Thus, the FGOALS-f3-L model can reasonably capture the spatial climatology pattern of

T

over the TP. Although the major low-temperature zone and the value of the low-temperature zone can be reproduced by FGOALS-f3-L ( Fig. 1 (b)), some considerable

T

biases still exist over the TP. As shown in Fig. 2 (a), the annual mean

T

simulation results show a cold bias over the TP, especially in the western TP. The cold bias mainly occurs in winter and spring and transforms into a weak warm bias in summer and autumn ( Fig. 2 (b-e)).

4. T s bias contributions and their seasonal variation

4.1. Distribution of annual mean T s bias contributions

To detect each process’s relative importance, we decompose the surface temperature

T

bias into six terms by utilizing Eq. (5) . Fig. 3 shows the annual mean

T

bias, the six terms on the right-hand side of Eq. (5) ,and the sum of the six terms. The spatial patterns of the annual mean

T

bias ( Fig. 3 (a)) and the sum of the six terms ( Fig. 3 (h)) are relatively similar, indicating the accuracy of this method. The main error sources of the method are the interpolation and the linearization of surface upward longwave radiation adopted on the left-hand side of Eq. (5) .

Fig. 4. (a) Annual (ANN) and (b-e) seasonal (winter, spring, summer, and autumn) mean surface albedo difference between FGOALS-f3-L and the CFSR from 1981 to 2010.

As shown in Fig. 3 (a), the simulated annual mean

T

shows a cold bias over the TP, and the bias is more robust in the western TP. The SAF term is negative in the western TP and positive in the eastern TP ( Fig. 3 (b)). Combined with the first term on the right-hand side of Eq. (5) , when the SAF term is negative, the simulated surface albedo in FGOAL-f3-L is higher than that in CFSR. FGOALS-f3-L overestimates(underestimates) the annual mean surface albedo in the western (eastern) TP ( Fig. 4 (a)). The surface albedo can be significantly impacted by snow cover over the TP. Xiao et al. (2019) showed that the characteristics of snow cover simulations in climate models are opposite to those from most reanalysis datasets in the western TP. Therefore, the SAF term related to the albedo parameterization schemes is very likely the main mechanism of annual mean cold bias in the western TP. The CRFs term makes a negative contribution to the

T

bias in the eastern TP ( Fig. 3 (c)). The CRFs term is related to the reflection of solar radiation by clouds, which is more important than the greenhouse effect of clouds on longwave radiation over the TP ( Wang, 1996 ). Fig. S1 shows that the annual mean net cloud radiation effect is more influential in the eastern TP, especially in summer. Comparatively, the CRFs term has a greater contribution in the eastern TP ( Fig. 3 (c)) and is the main contributor to the annual mean cold bias in the eastern TP. The SW term also contributes to the whole TP’s cold bias, but its intensity is weaker than that of the CRFs term. Its contribution is less than that of the CRFs term ( Fig. 3 (d)). The LW term makes a warm bias contribution in the southern and northern TP, while it makes a cold bias contribution in the middle and eastern TP ( Fig. 3 (e)). When the simulated surface albedo is higher than that in the CFSR, more solar radiation is reflected in space and leads to net surface cooling. Fig. 3 (f) shows that the spatial pattern of the

H

+ LE term is opposite to that of the SAF term and that the

H

+ LE term accounts for the positive (negative)

T

bias over the western (eastern) TP. The

Q

term makes a small but positive contribution to the warm bias over the whole TP ( Fig. 3 (g)). In general, Fig. 3 shows that the annual mean cold bias contributions in the western and eastern TP are different. The main contribution to the annual mean cold bias in the western TP is the SAF term, and the main contribution to the annual mean weak cold bias in the eastern TP is the CRFs term.

4.2. Analysis of seasonal mean T s bias contributions

To examine the relative importance of different physical processes to the seasonality of the

T

bias and their differences between the western and eastern TP, we choose the area ～25°-40°N, ～70°-105°E with an altitude above 2500 m and divide this area into two parts at 90°E based on the spatial surface albedo bias ( Fig. 4 ) and annual mean

T

bias pattern ( Fig. 3 (a)). Fig. 5 shows the area-weighted decomposition results of the annual and seasonal mean

T

biases. Owing to the strong warm bias around the 2500 m altitude ( Fig. 2 (a)), the area-weighted annual mean

T

bias is approximately 0.18 °C for the whole TP ( Fig. 5 (a)).The cold (warm) bias for the whole TP is maximum in winter (autumn),approximately − 1.86 °C (1.89 °C) ( Fig. 5 (b-e)).

Fig. 5. The regional average T s bias from 1981 to 2010, the partial temperature changes due to SAF, CRFs, SW, LW, H + LE, Q , and the sum of these six terms.The selected areas are ～25°-40°N, ～70°-105°E (red), ～25°-40°N, ～70°-85°E (blue), and ～25°-40°N, ～85°-105°E (light blue), with an altitude above 2500 m. (a)Annual (ANN) mean decomposition results and (b-e) seasonal (winter, spring, summer, and autumn) mean decomposition results, in units of °C.

The SAF term contributes approximately − 0.91 °C to the annual mean

T

bias for the whole TP ( Fig. 5 (a)). However, the contribution of the SAF term in the western TP is approximately − 1.81 °C, which is more significant than the CRFterm for the whole TP, which is approximately − 1.69 °C ( Fig. 5 (a)), indicating the importance of the SAF term in the western TP. Associated with the overestimated surface albedo in the western TP in winter and spring, the strong negative SAF term in the western TP makes it the most considerable contribution to the cold bias for the whole TP in winter and spring ( Fig. 5 (b-c)). The CRFterm contributes approximately − 1.69 °C to the annual mean

T

bias for the whole TP ( Fig. 5 (a)). In the eastern TP, it contributes − 1.04 °C. The CRFterm is understandably negligible in winter. This is because the reflection of solar radiation by clouds is negligible during winter when the solar radiation is at its minimum. The SW term related to the water vapor content is shown as a negative contribution to the annual mean

T

bias for the whole TP, and the contribution in the eastern TP is slightly larger than that in the western TP ( Fig. 5 (a)). The SW term is negligible in winter for the whole TP and reaches its maximum in summer, approximately − 1.7°C. In summer, FGOALS-f3-L overestimates the water vapor content over the TP (Fig. S2). More water vapor will absorb more shortwave radiation, resulting in a lower surface temperature. Contrary to the SW term, the LW term’s contribution is a positive contribution to the annual mean

T

bias for the whole TP ( Fig. 5 (a)). Notably, the LW term is the main contribution to the warm bias for the whole TP in summer and autumn. More water vapor in summer and autumn over the TP (Fig. S2) will emit more longwave radiation, causing a higher surface temperature. The

H

+ LE term contributes positively to the annual mean

T

bias for the whole TP ( Fig. 5 (a)). It reaches its maximum in spring,approximately 3.78 °C, which may be related to snow melting. In spring,the strong western

H

+ LE term offsets the strong western negative SAF term, and the strength of the cold bias becomes weak. In summer, the

T

bias turns into a warm bias under the influence of the LW term. The

Q

term is always a positive contribution to the seasonal mean

T

bias for the whole TP, and the difference between the western and eastern TP is small ( Fig. 5 (a)).

5. Summary and discussion

This study evaluates the performance of FGOALS-f3-L in simulating

T

over the TP and analyzes the annual and seasonal mean

T

bias based on the surface energy budget equation. The main conclusions are as follows. (1) Compared with the CFSR, the model can reasonably simulate the annual mean

T

pattern over the TP. (2) This model underestimates the annual mean

T

over the TP, and the cold bias in the western TP is more robust than that in the eastern TP. In winter and spring, the simulated

T

for the whole TP shows a cold bias. In summer and autumn, it shows a warm bias. (3) The decomposition results of the annual mean

T

bias show that the main contributions in the western and eastern TP are different. The strong negative SAF term is the main contribution to the annual mean cold bias in the western TP. However, it can be offset by the surface latent and sensible heat flux term. Due to the reflection of solar radiation by clouds, the CRFterm is the main contribution to the annual mean weak cold bias in the eastern TP. (4) In winter and spring, FGOALS-f3-L overestimates the surface albedo in the western TP, making the SAF term the largest contribution to the cold bias for the whole TP. In summer and autumn, a large portion of the warm bias over the whole TP is associated with the increased clear-sky downward longwave radiation. This increased longwave radiation may be related to the greater water vapor content simulation in FGOALS-f3-L.In this study, we mainly discuss the main contributions to the

T

bias over the TP, which is defined as the difference between FGOALS-f3-L and the CRSR. Although the CFSR has corrected many known errors in the observational data input and execution of previous reanalysis, some errors may still persist ( Saha et al., 2010 ). Wang et al. (2011) found that the CFSR has a few deficiencies in the long-term variations, and these deficiencies may have an impact on the application of the CFSR for climate diagnoses and predictions. Therefore, this study only focuses on the relative contribution of each process to the

T

bias. To accurately define the source of the

T

simulation in FGOALS-f3-L over the TP, more reanalysis datasets and special numerical experiments are needed in future studies.

Funding

This work was supported by the National Key Research and Development Program of China [grant number 2018YFC1505706],the National Natural Science Foundation of China [grant numbers 91937302 , 91737306 , and 41975109 ], and the Strategic Priority Research Program of the Chinese Academy of Sciences [grant number XDA17010105].

Acknowledgments

The CERES products used in this study are produced by the NASA CERES Team, available at http://ceres.larc.nasa.gov .

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi: 10.1016/j.aosl.2020.100012 .