Performance of surface radiation products of Greenland Ice Sheet using in-situ measurements

CHE Jiahang,HUAI Baojuan*,SUN Weijun,DING Minghu,WANG Lei,ZHANG Qinglin,WU Jiake,KANG Limin,TENG Xinru,YANG Xiaohong,YAN Jinpei & ZHAO Shuhui

1 College of Geography and Environment,Shandong Normal University,Jinan 250358,China;

2 State Key Laboratory of Severe Weather,China Academy of Meteorological Sciences,Beijing 100081,China;

3 Key Laboratory of Global Change and Marine-Atmospheric Chemistry,Third Institute of Oceanography,Ministry of Natural Resources,Xiamen 361005,China;

4 School of Tourism,Taishan University,Taian 271021,China

Abstract Radiation is the direct energy source of the surface natural environment and the main driving force of climate change.It has increasingly become an important meteorological factor affecting the surface heat exchange and glacier mass balance,especially in the glacier changes of the Greenland Ice Sheet (GrIS).Due to the harsh climatic conditions of GrIS and sparse observed data,it has become an important way to obtain radiation data from reanalysis datasets.However,the applicability of these radiation data on GrIS is uncertain and worth exploring.In this work,we evaluate five reanalysis datasets (the fifth generation of European Centre for Medium-Range Weather Forecasts (ERA5),European Centre for Medium-Range Weather Forecasts Interim Reanalysis (ERA-Interim),Japanese 55-year Reanalysis (JRA55),National Centers for Environmental Prediction Reanalysis II (NCEP2) and Modern-Era Retrospective analysis for Research and Applications,Version 2 (MERRA-2)) during 1997-2022 using observations from 26 Program for Monitoring the Greenland Ice Sheet (PROMICE) automatic weather stations (AWSs) and 3 K-transect AWSs on GrIS.The conclusions are as follows: ERA5 has the best performances in downward shortwave radiation (SWD) as well as downward and upward longwave radiation (LWD and LWU),but the performance is not the best in upward shortwave radiation (SWU).Based on the radiation budget analysis with ERA5 during 1979-2022,the fluctuation of longwave radiation is greater than that of shortwave radiation.The seasonal variation of shortwave radiation is obvious,while that of longwave radiation is small.The increasing trend of longwave radiation may result from global warming,in which ice sheets absorb more solar radiation and the surface heats up significantly,emitting more LWU.

Keywords Greenland Ice Sheet,downward shortwave radiation,upward shortwave radiation,downward longwave radiation,upward longwave radiation,reanalysis datasets,automatic weather stations

1 Introduction

The Greenland Ice Sheet (GrIS) covers an area of about 1.7 × 106km2,with a length of about 2200 km from south to north.It is the second-largest ice sheet in the world except for Antarctica (Steffen and Box,2001).Solar radiation is one of the most important sources of energy on the Earth’s surface (Wang et al.,2021).The radiation budget affects the energy balance at the Earth’s surface,and also affects the melting and disintegration of the ice sheet.Therefore,it is particularly important to study the radiation flux and its change for the prediction of future melting changes of the GrIS (Bamber and Riva,2010).

The sum of longwave and shortwave radiation determines the energy budget of the earth-atmosphere system and greatly control the local climate (Power and Mills,2005).There are automatic weather stations (AWSs) observing radiation flux on the GrIS,but they are few in number and mostly located in coastal areas due to the harsh climate in the region,which makes equipment maintenance difficult.As a result,there are many missing values in the observed data and we cannot obtain the long-term radiation datasets of the entire Greenland surface (Fausto et al.,2012).Remote sensing and reanalysis datasets make up for this defect (Trenberth and Guillemot,1998).Remote sensing mainly obtains observation data through satellites and simulates surface radiation by using the physical model of atmospheric radiation transmission (Schroeder et al.,2009;Zhang et al.,2015),which has the advantage of high spatial resolution,but its time coverage is limited (Liang et al.,2006).Moreover,due to the complex characteristics of remote sensing data,the accuracy of radiation products may be affected (Peng,2019).In the 1990s,reanalysis datasets emerged to avoid this time series of shortcomings by optimally combining observations from different types and sources with short-term weather forecasts through a data assimilation system,which has the advantages of continuous time series,easy access and coverage of the entire Greenland surface (Trenberth and Olson,1988).Thus it has played an increasing role in estimating radiation changes in various regions of the global (Bengtsson et al.,2007;Zhao et al.,2010).Due to the reanalysis datasets are assimilated by different data and means,the global performance capacity of each reanalysis dataset is also different,which will also make the analysis related to radiation quantity show great uncertainty and error (Wang and Dickinson,2013).Therefore,it is necessary to evaluate the performance of each reanalysis dataset,so that the reanalysis datasets with better performance in different regions can be selected for further study.Radiation is the most important energy input for ice sheet surface melt,and the change of radiation will directly affect the amount and area of melt.The study of radiation is of great significance to the surface melt of the GrIS.

At present,previous studies have evaluated the radiation data of reanalysis (Griggs and Bamber,2008;Cox et al.,2014;Lenaerts et al.,2017;Seo et al.,2020).Although some reanalysis radiation datasets have been evaluated on GrIS,they use less measured data and lack comparison between multiple datasets.For example,Cox et al.(2014) analyzed the downward longwave radiation (LWD) over GrIS using surface-based observations from Summit Station on GrIS and European Centre for Medium-Range Weather Forecasts Interim Reanalysis (ERA-Interim) LWD fields.The results showed that ERA-Interim performed reasonably well at Summit Station in simulating of LWD.However,other radiation components were not studied and comparison with other reanalysis data was not made.In evaluating cloud cover characteristics using Greenland satellite datasets and three reanalysis products ((NCEP1),the second NCEP-Department of Energy (DOE) Atmospheric Model Intercomparison Project (NCEP2),and ERA-40),Griggs and Bamber (2008) found that the reanalysis datasets were inconsistent in modeling radiation.This result suggests that different reanalyses may be inaccurate in estimating radiation balance over the GrIS.More research has focused on the Arctic rather than GrIS.For example,Seo et al.(2020) assessed net surface radiation over the Arctic using data from the fifth generation of European Centre for Medium-Range Weather Forecasts (ERA5),National Centre for Environmental Prediction (NCEP),Cloud and Earth Radiant Energy System (CERES) Energy Balanced and Filled (EBAF),and Global Energy and Water Exchanges (GEWEX).They found that ERA5 had the highest accuracy.Lenaerts et al.(2017) compared the downward radiation of the multi-reanalysis combined with CMIP5 and found significant but inconsistent deviations in the downward radiation component.Huang et al.(2017) used five reanalysis datasets in Arctic region (70°N-90°N) to compare surface longwave and shortwave radiation fluxes derived from NASA CERES-MODIS (CM).All reanalysis showed positive deviations from the northern and eastern coasts of Greenland.Both these studies all contribute to understand the accuracy of radiation products at high latitudes and the differences between products,but they do not focus on reanalyses over the GrIS.Serreze et al.(1998) found the NCEP-NCAR products capture 50%-60% of the observed spatial variance in global radiation during most months.At present,there are few assessments of radiative products on the GrIS,and there is a lack of systematic and complete assessment of radiative products from reanalysis data.Therefore,this work uses the data of 29 AWSs in the Programme for Monitoring of the Greenland Ice Sheet (PROMICE) and K-transect on GrIS from 1997 to 2022 and five reanalysis datasets (ERA5,ERA-Interim,JRA55,NCEP2,MERRA-2) are evaluated and analyzed for their applicability.

The organization of this paper is as follows: The data used in this study (observational data and five reanalysis datasets) are presented in section 2;The sections 3.1 and 3.2 introduce the assessment results and the radiation budget analysis based on ERA5 (1979-2022),and section 3.3 is the discussion part.Finally,a summary is given.

2 Data and methods

2.1 In-situ observations

We used daily downward shortwave radiation (SWD),upward shortwave radiation (SWU),downward longwave radiation (LWD) and upward longwave radiation (LWU) data from 26 AWSs in the PROMICE network (Fausto et al.,2021).These data are operated by the Geological Survey of Denmark and Greenland (GEUS) in collaboration with the National Space Research Institute of the Technical University of Denmark (Ahlstr?m and Team,2008).We used data from three AWSs along the K-transect in southwest GrIS,i.e.,three Institute for Marine and Atmospheric research Utrecht (IMAU) AWSs (S5,S6,S9).The K-transect was established by the Greenland Ice Margin Experiment (GIMEX) (Oerlemans and Vugts,1993;Kuipers Munneke et al.,2018).

The distribution of AWSs in Greenland is shown in Figure 1a and the topography is shown in Figure 1b.The locations,elevations and observation periods of these stations are shown in Table S1.

Figure 1 a,Location map of the research area and AWSs used in this study: The blue circle represents the AWSs of the K-transect,and the red triangle represents the PROMICE AWSs;b,Topographic map of Greenland.

2.2 Reanalysis products

2.2.1 ERA5

The ERA5 reanalysis is the fifth generation of global atmospheric reanalysis from the European Centre for Medium-Range Weather Forecast (ECMWF),covering the period from 1940 to the present.It provides hourly data on SWD,LWD,SWN (net shortwave radiation) and LWN (net longwave radiation) from which we can calculate SWU and LWU.The ERA5 has a temporal resolution of 1 h and a spatial resolution of 0.25°×0.25°.Assimilation schemes for ERA5 reanalysis are 10 mixed 4-Dimensional Variational Analysis (4D-Var) methods,Ensemble of Data Assimilation technology,the latest version of the high-resolution Integrated Forecast System (IFS),Earth System Models and observation data assimilation system CY41r2 Global Spectral Model.There are 137 mixed sigma-pressure (mode) levels in the vertical direction,the resolution of which is T639 (～31 km) in the horizontal direction and T319 (～63 km) in the vertical direction.The vertical direction is the highest level from class 137 to 0.01 hPa (80 km) (Hoffmann et al.,2019;Hersbach et al.,2020).

2.2.2 ERA-Interim

The ERA-Interim reanalysis is sponsored by the European Union and run by the ECMWF.The dataset adopts 4D-Var method,IFS and Cy31r2 Global Spectral Model,combined with improved humidity analysis,satellite data error correction and other technologies.Moreover,the horizontal resolution is improved to T255 (～79 km) (Berrisford et al.,2011;Dee et al.,2011).In this paper,12-hour grid data with a spatial resolution of 0.125°×0.125° is obtained from the ECMWF.The data includes SWD,LWD,SWN and LWN,from which we calculated SWU and LWU.And the time range was January 1997 to August 2019.

2.2.3 JRA55

JRA55 was jointly launched by the Japan Meteorological Agency (JMA) and the Central Research Institute of Electric Power Industry (CRIEPI).JRA55 uses a more advanced 4D-Var method to provide data from January 1958 to the present (Ebita et al.,2011;Kobayashi et al.,2015).Many deficiencies of JRA-25 have been well corrected in the JRA55 version,such as upgrading the spatial resolution grid from T106L40 to TL319L60.In short,compared with JRA25,JRA55 reanalysis data is more complete (Kobayashi et al.,2015;Peng,2019).In this work,we use SWD,SWU,LWD and LWU data with 3 h,0.56°×0.56° resolution.

2.2.4 MERRA-2

MERRA-2 is a high-resolution global reanalysis product developed by Global Modeling and Assimilation Office (GMAO) in the National Aeronautics and Space Administration (NASA).It aims to produce a regularly gridded,uniform global atmospheric record,covering the era of satellite observations from 1980 to the present.To achieve this goal,MERRA-2 updates the analysis protocols and models of the Goddard Earth Observation System (GEOS),using version 5.12.4 of the GMAO/GEO-5 data assimilation system.The spatial resolution of MERRA-2 is 0.5°×0.625° (Gelaro et al.,2017).We use the data of SWD,LWD,LWU,and SWN of MERRA-2,and calculate SWU through SWN and SWD.

2.2.5 NCEP2 (NCEP-DOE)

NCEP2 (NCEP-DOE) reanalysis is a global reanalysis from 1979 to the present,with a time resolution of the day (Kistler et al.,2001).The NCEP2 reanalysis project was established to provide an improved version of the original NCEP-NCAR reanalysis (Kalnay et al.,1996;Kanamitsu et al.,2002).The NCEP2 model has the same resolution as the T62 Gaussian grid (2.5°×2.5°) with 28 vertical sigma levels in the NEP-NCAR reanalysis project.NCEP2 is a “second generation” reanalysis,which uses an improved assimilation procedure based on 4D-Var with greater emphasis on accuracy,resolution,and long-term trends.Chou and Lee (1996) replaced the parameterization method of Lacis and Hansen (1974) and alleviated the overestimation of surface solar radiation in NCEP-NCAR (Saha et al.,2010).Here we use SWD,SWU,LWD and LWU data.The information for all the reanalyses is shown in Table 1.

Table 1 Summary of information of the five reanalyses used in this study

2.3 Methods

For PROMICE observational data,monthly shortwave and longwave radiation were calculated only when valid daily data were more than or equal to 90% available.Seasonal and annual data were not calculated when monthly data were missing (Zhang et al.,2022).For the K-transect observed hourly data,daily data can be calculated only when the effective hourly data is greater than or equal to 90% available,and the other data is the same as PROMICE.

We used bilinear interpolation method to extract SWD,SWU,LWD and LWU data of corresponding AWSs,and carried out comparative analysis.Then three classical statistical indicators were used to evaluate the performance of the five reanalyses,that is correlation coefficient (R),root mean square error (RMSE),and bias (BIAS).

By calculating the values of three evaluation indicatorsR,RMSE and BIAS between the reanalysis data and the measured data,rank scoring (Rs) was performed as the objective function,and a score of 0-10 was assigned according to the differences in the performance of the five reanalyses.The formula is as follows:

xiis the relative error between the statistical eigenvalues of the reanalyzed data and the measured data,andxmaxandxminare the maximum and minimum values of the error,respectively.First,theRsof different evaluation indicators is calculated for each set of reanalysis data,and the final score is obtained by the average of the corresponding scores of all evaluation indicators.The smaller the score,the better the applicability of reanalysis data.However,theRsvalue cannot represent the actual simulation accuracy of specific reanalysis data,and is currently mostly used in the applicability evaluation of atmospheric circulation models in various regions (Li et al.,2011;Liu et al.,2011,2013).

3 Results and discussions

3.1 Performance of the reanalysis datasets

3.1.1 Daily scale performance

Figure 2 Comparison of daily observed SWD,SWU,LWD,LWU with 5 reanalysis datasets.a-d,Observed and ERA5 SWD,SWU,LWD and LWU,respectively ;e-h,Same as a-d but for ERA-Interim;i-l,Same as a-d but for JRA55;m-p,Same as a-d but for MERRA-2;q-t,Same as a-d but for NCEP2.

Figure 2 shows the performance of daily radiation components from five reanalyses based on observations.All reanalyses could present SWD change (R＞0.94,p＜0.05),suggesting they could reflect long-term SWD change well.However,MERRA-2 and NCEP2 have higher BIAS relative to AWSs.MERRA-2 systematically underestimates SWD,but NECP2 overestimates SWD below 350 W·m-2due to more SWD outputs in most seasons except June,July and August (JJA).Compared to ERA-Interim,ERA5 has not improved the performance of SWU and even become more terrible.JRA55 overestimates SWU in all seasons and the high RMSE (62.61 W·m-2) shows it hardly presents SWU conditions over the GrIS.Compared with other reanalyses,ERA-Interim still is a good SWU product in the GrIS.Besides,all reanalyses underestimate LWD over the GrIS,with systematically negative BIAS occurring in all reanalyses.However,their correlation with observations still performs well (R＞0.82,p＜0.05),suggesting that reanalyses could be used for simulating LWD anomalies change.In fact,the underestimation of LWD results from reanalyses underestimating LWU over the GrIS except JRA55.Given the BIAS difference between LWD and LWU in JRA55 and NCEP2,there may be potential problems for the cloud radiation parameterization scheme in these two reanalyses (Yeo et al.,2022).In addition,MERRA-2 performs worst for LWU,it underestimates greatly summer LWU conditions.Overall,reanalyses could simulate SWD well over the GrIS and all reanalyses could present daily radiation change well with high correlation values,especially ERA5.Rssequencing results of four radiation components of five reanalyses are shown in Figure 3a.

Figure 3 Rs score results of five reanalyses radiation four components at daily (a) and monthly (b).

A comprehensive analysis of the four radiation components shows that ERA5 has a good performance in radiation simulation.To further look into the performance of ERA5 at each AWS,we analyze theR,RMSE and BIAS of SWD in detail (Figure 4).SWD of ERA5 at different stations and regions shows great differences and has a highR-value in each station (R＞0.86,p＜0.05).RMSE ranges from 20 to 55 W·m-2.Among the magnified 27 AWSs,16 AWSs are underestimated and 11 stations are overestimated.Among them,3 AWSs in THU-transect (Figure 4d) and 2 AWSs in UPE-transect (Figure 4g) all underestimate SWD.Both AWSs in the SCO-transect (Figure 4v) overestimate SWD.The RMSE and BIAS of THU_U2 are relatively large,which may be caused by the effect of elevation,there is a big difference between the station location of THU_U2 and the elevation of adjacent cells (Minola et al.,2020;Molina et al.,2021).It is also possible that the actual received SWD is different from the SWD flux simulated by ERA5 due to the topography (Gao and Hao,2014;Longo-Minnolo et al.,2022;Vanella et al.,2022).

The three indicators (R,RMSE,BIAS) of SWU,LWD and LWU of ERA5 are respectively shown in Supplementary Figure S1,Figure S2,Figure S3.The three indicators (R,RMSE,BIAS) of SWD,SWU,LWD and LWU of ERA-Interim,JRA55,MERRA-2 and NCEP2 are shown in Supplementary Table S2,Table S3,Table S4 and Table S5,respectively.The simulation of ERA5 is better for LWD,but all the stations show negative deviation.

3.1.2 Monthly scale performance

Figure 5 shows the performance of monthly radiation components from five reanalyses based on observations,respectively.All reanalyses could present SWD change (R＞0.99,p＜0.05),and have extremely highR-values,suggesting they reflect long-term SWD change well.However,ERA-Interim,MERRA-2 and NCEP2 have higher BIAS relative to AWSs.MERRA-2 systematically underestimates SWD,but NECP2 overestimates SWD.This is the same as the daily data as the monthly data is averaged from the daily data.Compared to ERA-Interim,ERA5 has not significantly improved the performance of SWU,and the difference of performance between them is small.JRA55 overestimates SWU in all seasons and the high RMSE (50.52 W·m-2) indicating that it has little ability to simulate SWU on GrIS.This may be due to the fact that the parameterization scheme of JRA55’s snow ice albedo is fixed,and the change of snow ice albedo is instantaneous (Dorman and Sellers,1989;Kobayashi et al.,2015).Compared with other reanalyses,ERA5 and ERA-Interim still are good SWU products in the GrIS.Compared to daily data,ERA5 has improved the simulation performance of SWU in monthly data.Besides,all reanalyses underestimate LWD over the GrIS,with systematically negative BIAS occurring in all reanalyses.However,their correlation with observations still performs well (R＞0.89,p＜0.05),suggesting that reanalyses could be used for simulating LWD anomalies change.In fact,with the exception of JRA55,the negative deviation of LWD is due to the underestimation of LWU on GrIS.Given the BIAS difference between LWD and LWU in JRA55,there may be potential problems for the cloud radiation parameterization scheme in the reanalyses (Yeo et al.,2022).In addition,MERRA-2 performs worst for LWU,it underestimates summer LWU conditions by about 40 W·m-2.Overall,reanalyses simulate SWD and LWD well over the GrIS and all reanalyses could present monthly radiation change well with high correlation values,especially ERA5.Rssequencing results of four radiation components of five reanalyses are shown in Figure 3b.

Figure 4 Spatial distribution of R (yellow),RMSE (purple) and BIAS (blue or red) of SWD at daily scale ERA5.All AWSs are located on the GrIS (a).THU-transect (b-d);UPE-transect (e-g);KAN-transect (h-j);NUK-transect (k-m);QAS-transect (n-p);TAS-transect (q-s);SCO-transect (t-v);KPC-transect (w-y).

Figure 5 Comparison of monthly observed SWD,SWU,LWD,LWU with 5 reanalysis datasets.a,Observed and ERA5 SWD;b,Observed and ERA5 SWU;c,Observed and ERA5 LWD;d,Observed and ERA5 LWU;e-h,Same as a-d but for ERA-Interim;i-l,Same as a-d but for JRA55;m-p,Same as a-d but for MERRA-2;q-t,Same as a-d but for NCEP2.

A comprehensive analysis of the four radiation components shows that ERA5 has a good performance in radiation simulation.To further look into the performance of ERA5 at each AWS,we analyze theR,RMSE and BIAS of LWD in detail (Figure 6).The spatial distribution ofR,R MSE and BIAS of LWD of ERA5 is significantly different.RMSE ranges from 8.50 to 43.07 W·m-2.TheRof KAN_M station is the highest (R=0.99),and that of TAS_A is the worst.All AWSs show negative BIAS,indicating that ERA5 significantly underestimates LWD.Among them,SCO_L (Figure 6v) severely underestimates LWD,and in the daily data,the same problem exists for the two AWSs in the SCO-transect (Figure S2v).

Figure 6 Spatial distribution of R (yellow),RMSE (purple) and BIAS (blue) of LWD at monthly scale ERA5.All AWSs are located on the GrIS (a);THU-transect (b-d);UPE-transect (e-g);KAN-transect (h-j);NUK-transect (k-m);QAS-transect (n-p);TAS-transect (q-s);SCO-transect (t-v);KPC-transect (w-y).

TheR,RMSE and BIAS of SWD,SWU and LWU of ERA5 are shown in Figure S4,Figure S5,Figure S6.TheR,RMSE and BIAS of SWD,SWU,LWD and LWU at the other four reanalysis datasets are shown in Table S6,Table S7,Table S8,Table S9.The simulation of ERA5 on the AWSs is also that SWD is better than SWU,and there is underestimation in most AWSs of LWU.

Figure 7 Monthly changes of R,RMSE and BIAS of SWD of five reanalysis datasets (ERA5,ERA-Interim,JRA55,MERRA-2,NCEP2).a,R of observed and five reanalysis datasets for SWD;b,RMSE of observed and five reanalysis datasets for SWD;c,BIAS of observed and five reanalysis datasets for SWD;d-f,Same as a-c but for SWU;g-I,Same as a-c but for LWD;j-l,Same as a-c but for LWU.

By analyzing the seasonal cycle of three evaluation indexes (R,RMSE and BIAS) with these reanalyses in different radiation component in Figure 7,the following conclusions can be drawn.For SWD,ERA5 has a higherRin each month than the other four reanalysis datasets (ERA-Interim,JRA55,MERRA-2,NCEP2).JRA55’sRdecreases significantly in March and ERA-Interim’sRdecreases significantly in November.The simulation performance of five reanalysis datasets are higher in summer than in other seasons.When comparing RMSE,NCEP2 and MERRA-2 are higher than the other three reanalysis datasets (ERA5,ERA-Interim,JRA55) with a large deviation.In all reanalysis datasets,RMSE is largest in summer and smallest in winter,which is because SWD is basically absent in winter and SWD is large in summer.Compared with SWD,the simulation performance of SWU decreases significantly.The values of RMSE and BIAS are large and there are obvious seasonal fluctuations,with the worst performance in summer and the best performance in winter.ERA-Interim and JRA55 perform worse than the other three reanalysis datasets.For LWD,there is a significant decrease inRof MERRA-2 in summer,and the simulation performance is poor.Four reanalysis datasets have no seasonal fluctuations in RMSE and BIAS,and the change is little.ERA5 performs better than the other four reanalysis datasets.RMSE and BIAS do not have seasonal fluctuations,which is the same for LWU and LWD.However,the RMSE and BIAS of MERRA-2 is significantly higher than those of the other four reanalysis datasets,and theRis significantly lower,indicating its poor ability to simulate LWU on GrIS.The poor performance of longwave radiation may be caused by the problem of parametric scheme (Yeo et al.,2022).TheRvalues of the five reanalysis datasets have significant seasonal fluctuations in simulating LWU,with poor performance in summer and better performance in winter.

3.2 Radiation budget of Greenland during 1979-2022

Through the evaluation of the four components of radiation (SWD,SWU,LWD,LWU) of five reanalysis datasets,ERA5 is relatively a good reanalysis product.Therefore,this work analyzes the four components of seasonal and annual radiation in Greenland from 1979 to 2022 by using the data of ERA5 (Rossow and Due?as,2004;Pinker et al.,2005).

The intensity of shortwave radiation is affected by solar elevation angle,cloud cover,aerosol and water vapor (Gu et al.,2001;Pfister et al.,2003),and the main source of the SWD is solar radiation.The time series of summer anomaly shows that the summer shortwave radiation flux has a downward trend since 2000,and the annual variation is mainly affected by the summer radiation flux,which also has a downward trend (Figure 8).There is no shortwave radiation in winter,so the anomaly time series fluctuation is minimal.However,longwave radiation shows an upward trend at seasonal and annual scales.The interannual fluctuation of longwave radiation may be greatly affected by summer,and less affected by other seasons.In winter,the time series fluctuates the most and the annual longwave radiation flux fluctuates the most smoothly.The anomaly time series fluctuation of longwave radiation is larger than that of shortwave radiation.The increasing trend of longwave radiation is explained in detail in the spatial distribution of the trend below.

Figure 8 The four radiation components relative to the multi-year average time series of radiation anomalies from 1979-2022.a,SWD;b,SWU;c,LWD;d,LWU.MAM indicates March,April,May;JJA indicates June,July,August;SON indicates September,October,Novermber;DJF indicates December,January,February.

Figure 9 Multi-year average radiation from 1979 to 2022.a,SWD;b,SWU;c,LWD;d,LWU.

The shortwave radiation flux is lower than longwave radiation flux.The main source of shortwave radiation is solar radiation,while the longwave radiation can come from the radiation emitted by the surface,atmospheric radiation,etc.The main reason for the low shortwave radiation flux is that there is no shortwave radiation in winter,so the average annual radiation value is low.The average annual radiation flux of shortwave radiation is concentrated in 80-160 W·m-2(Figure 9),and the radiation flux in the south is larger than that in the north,showing the latitude zonality.SWU is mainly affected by SWD,so the multi-year average of SWU is lower than that of SWD.The average annual radiation flux of longwave radiation is greater than 140 W·m-2.LWD is mainly affected by LWU,so LWD flux is lower than LWU flux.Shortwave radiation and longwave radiation present different situations at the edge of the GrIS and the coastal areas of Greenland.Shortwave radiation has a low radiation flux in the coastal areas,while longwave radiation has a high radiation flux in the coastal areas.This is mainly influenced by the underlying surface,such as the underlying surface of the ice sheet,tundra,ocean and sea ice,and therefore the albedo is different.The low albedo in the southern coastal areas leads to the obvious low SWU flux.The longwave radiation absorbs more heat and has a larger longwave radiation flux.

Because Greenland has polar nights in winter,shortwave radiation is almost zero in winter.In spring and autumn,due to the influence of solar elevation angle,the shortwave radiation flux is small,and there is significant latitude zonality (Figure 10).The ice sheet receives more solar radiation in summer than in other seasons.SWD fluxes are greater on the ice sheet than in coastal areas.We speculate that the main reason is cloud cover (Abe et al.,2016;Kim et al.,2020;Li and Xu,2020).On GrIS,there is less evaporation and less cloud cover.Therefore,clouds have little shading effect on solar radiation,and most solar radiation directly reaches the surface,forming a large radiation flux.On the other hand,the coastal area of Greenland has high evaporation,high cloud cover and strong shading,and the solar radiation reaching the surface will be significantly reduced.Therefore,the SWD flux in the seas around Greenland is smaller than that of GrIS.The obvious difference of SWD in the east and west of Greenland Ice Sheet in summer may be due to the influence of Greenland blocking index (GBI).When GBI intensifies,there is less cloud cover and more SWD.When the GBI intensity of western Greenland and eastern Greenland is different,the SWD of eastern and western Greenland is different (Lewis et al.,2021;Preece et al.,2022).Longwave radiation is not affected by the solar elevation angle,there is no the latitude zonality.Overall,the radiation flux within the ice sheet is lower than in the coastal area,which may be mainly influenced by the (underlying surface) albedo.Summer radiation flux is greater than other seasons.For the GrIS,the trend of shortwave radiation is not significant,varying in the range of -0.4-0.4 W·m-2·(10 a)-1,while the trend of longwave radiation is obviously rising in the range of 0.8-2.0 W·m-2·(10 a)-1(Figure 11).The variation trend in the interior of the ice sheet is not obvious,but the radiation in the coastal areas of Greenland have a large variation.SWU has a decreasing trend in the eastern and northwestern coastal areas of Greenland,probably due to global warming,melting of sea ice and lower albedo.The increasing trend in longwave radiation may be result from global warming,in which ice sheets absorb more solar radiation and the surface heats up significantly,emitting more LWU.There is an anomaly in northwest Greenland that is significantly different from other regions.The reason may be that the sea ice concentration areas have a higher albedo than other regions,which leads to an increasing trend of SWU and a decreasing trend of longwave radiation (Renfrew et al.,2021).

Figure 10 Multi-year average seasonal radiation from 1979 to 2022.a,SWD;b,SWU;c,LWD;d,LWU.

Figure 11 Radiation change trend from 1979 to 2022.a,SWD;b,SWU;c,LWD;d,LWU.

3.3 Discussions

It is found that ERA5,ERA-Interim,JRA55,MERRA-2,and NCEP2 are all able to roughly fit SWD,SWU,LWD,and LWU.However,the simulation performance of SWU on both daily and monthly scales is relatively poor,which may be caused by inaccurate estimation of underlying surface types in reanalysis products (Wei and Li,2003;Liang et al.,2015;Zhang et al.,2021,2023).On the monthly scale,there is a significant underestimation in the LWD simulation,and a certain deviation in the LWU radiation simulation (Wang,2022).

On both daily and monthly time scales,none of the reanalysis is optimal when simulating the four components of radiation (Wang,2021).This may be caused by different boundary layer processes and observations using different models when producing reanalysis.ERA5 has better performance in radiation simulation because it uses the 4D-Var data assimilation and model prediction generation of the IFS CY41r2 (Hoffmann et al.,2019;Hersbach et al.,2020).The JRA55 adopts a 4D-Var data assimilation system and variational deviation correction of satellite radiation,and adds a new data source (Kobayashi et al.,2015).However,these optimizations do not significantly affect the SWU simulation performance,and the radiation performance of JRA55 needs to be improved.The MERRA-2 reanalysis adopts the GEOS 5.12.4 model and updates the Clintrial Integration Solution scheme to reduce some errors in the observation system (Wu et al.,2002;Rienecker et al.,2011;Gelaro et al.,2017),the simulation of longwave radiation is poor,and the simulation quality is obviously different in different months,so the simulation is relatively poor.In order to fundamentally solve the accuracy of the radiation reanalysis,further research should be carried out from the aspects of improving satellite sensors,reducing the influence of cloud cover and quantifying the influence of complex terrain on reanalysis products (Xu et al.,2020;Wang,2021).

In addition to being affected by the assimilation system of the reanalysis,the accuracy of satellite sensors and topographic conditions,meteorological factors such as aerosol,cloud cover,water vapor,solar elevation angle also affect the solar radiation and thus affect the accuracy of the reanalysis (Wang,2022;Wang et al.,2022).At the same time,the surface albedo is also an important factor affecting the reflectivity of solar radiation,directly affecting the SWU and LWU fluxes (Ruckstuhl et al.,2008;Zhou et al.,2013;Wang,2021).

Cloud cover is an important factor affecting the four components of radiation,among which cloud cover will not only weaken solar radiation,but also exert various influences on solar radiation by its various physical properties,cloud cover conditions and optical properties (Gu et al.,2001;Pfister et al.,2003).Previous studies have shown that underestimation of cloud cover can lead to overestimation of solar radiation in reanalysis (Zhu,1982;Peng,2019).Above,the spatial patterns of the SWD in summer presented in this work further confirmed that SWD was affected by cloud cover (Abe et al.,2016;Kim et al.,2020;Li and Xu,2020).However,Zhang (2019) suggested that there was a small correlation between cloud cover and SWD,which might be caused by different study regions.The influence of cloud on solar radiation is very complicated.Different geometric properties such as cloud height,cloud shape and cloud thickness,as well as different physical,chemical and optical properties of cloud will have different effects on solar radiation (Shen et al.,2008).Thus,although changes in total cloud cover currently observed are difficult to fully explain changes in ground-based solar radiation,the important effect of changes in cloud on solar radiation cannot be denied (Shen et al.,2008).In addition,clouds cannot directly affect LWU,but they can affect surface temperature and thus LWU by influencing SWD.Aerosols are another important and complex influence factors on solar radiation,either directly reflecting,scattering or absorbing solar radiation,or indirectly by altering the microphysical properties of clouds (Shen et al.,2008).At the same time,the influence of different aerosols on solar radiation also varies greatly (Shi et al.,2008).But in general,an increase (decrease) in aerosols also generally reduces (increases) the amount of solar radiation reaching the surface.In the last 50 years,there have been two major eruptions with global impacts: The eruptions of Mount El Chichón (Mexico 1982) and Mount Pinatubo (Philippines 1991) significantly increased the natural aerosol concentration and optical thickness on a global scale (Mishchenko et al.,2007),which significantly weakened the direct radiation reaching the ground and enhanced the scattered radiation (Robock,2000).The apparent low value of shortwave radiation in 1983 may be related to the 1982 eruption (Figures 8a,8b),but the 1991 eruption did not show apparent effect.Anthropogenic aerosols are mainly the aerosols emitted by burning fossil fuels in recent decades due to global industrial and economic development,such as sulfate and black carbon,which contribute about 1/3 of the global average aerosol optical thickness (Streets et al.,2006).Aerosols have a negative forcing effect on the surface shortwave radiation,which may cause the reduction of shortwave radiation.

4 Conclusions

Based on five reanalysis datasets from 29 AWSs on the GrIS,we evaluate SWD,SWU,LWD,and LWU of ERA5,ERA-Interim,JRA55,MERRA-2 and NCEP2.On the daily scale,ERA5 is superior for all radiation components except SWU.And the comprehensive analysis suggests that ERA5 should be the better choice to fill the blank regions of Greenland.The five reanalyses all show relatively poor performance of SWU.In simulating of LWD and LWU,ERA5 has the best simulation performance.

By analyzing the radiation budget during 1979-2022 with ERA5,the time series of shortwave radiation is smaller than longwave radiation.The multiyear seasonal average of shortwave radiation is mainly affected by the solar elevation angle in spring,autumn and winter,while the multiyear seasonal average of longwave radiation is mainly affected by surface albedo.The inland GrIS has high albedo and low longwave radiation flux,whereas ice sheet margins and coastal areas are the opposite.Longwave radiation has an obvious increasing trend,while shortwave radiation trend is not obvious,but there is an increasing trend in some regions.

Author contributionsChe J H,Zhang Q L,Wu J K,Kang L M,Teng X R and Yang X H wrote the paper;Che J H,Zhang Q L,Wu J K,Teng X R and Huai B J analyzed the data;Huai B J,Sun W J,Ding M H,Wang L,Yan J P,Zhao S H edited the paper.

Conflicts of interestThe authors declare no conflict of interest.

Data availability statementThe PROMICE observations are available online (http://promice.org/download-data).ERA5 data are available online (https://cds.climate.copernicus.eu/#!/home).ERA-Interim data are available online (https://apps.ecmwf.int/datasets/data/interim-mdfa/levtype=sfc/).JRA55 data are available online (https://rda.ucar.edu/ datasets/ds628.0/).MERRA-2 data are available online (https://search.earthdata.nasa.gov/search?fp=MERRA-2).NCEP2 data are available online (https://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis2.html).

AcknowledgmentsThis work was funded by the Natural Science Foundation of China (Grant no.42171121) and the open fund of Key Laboratory of Oceanic Atmospheric Chemistry and Global Change,Ministry of Natural Resources,China (Grant no.GCMAC2206).The authors gratefully acknowledge support from data availability from PROMICE and ERA5,ERA-Interim,JRA55,MERRA-2,NCEP2.We thank Guest Editor,Dr.Tong Zhang and two anonymous reviewers for reviewing this manuscript.

Supplementary Figures and Tables

Figure S1 Spatial distribution of R (yellow),RMSE (purple) and BIAS (blue or red) of SWU at daily scale ERA5.All AWSs are located on the GrIS (a).THU-transect (b-d);UPE-transect (e-g);KAN-transect (h-j);NUK-transect (k-m);QAS-transect (n-p);TAS-transect (q-s);SCO-transect (t-v);KPC-transect (w-y).

Figure S2 Spatial distribution of R (yellow),RMSE (purple) and BIAS (blue or red) of LWD at daily scale ERA5.All AWSs are located on the GrIS (a).THU-transect (b-d);UPE-transect (e-g);KAN-transect (h-j);NUK-transect (k-m);QAS-transect (n-p);TAS-transect (q-s);SCO-transect (t-v);KPC-transect (w-y).

Figure S3 Spatial distribution of R (yellow),RMSE (purple) and BIAS (blue or red) of LWU at daily scale ERA5.All AWSs are located on the GrIS (a).THU-transect (b-d);UPE-transect (e-g);KAN-transect (h-j);NUK-transect (k-m);QAS-transect (n-p);TAS-transect (q-s);SCO-transect (t-v);KPC-transect (w-y).

Figure S4 Spatial distribution of R (yellow),RMSE (purple) and BIAS (blue or red) of SWD at monthly scale ERA5.All AWSs are located on the GrIS (a).THU-transect (b-d);UPE-transect (e-g);KAN-transect (h-j);NUK-transect (k-m);QAS-transect (n-p);TAS-transect (q-s);SCO-transect (t-v);KPC-transect (w-y).

Figure S5 Spatial distribution of R (yellow),RMSE (purple) and BIAS (blue or red) of SWU at monthly scale ERA5.All AWSs are located on the GrIS (a).THU-transect (b-d);UPE-transect (e-g);KAN-transect (h-j);NUK-transect (k-m);QAS-transect (n-p);TAS-transect (q-s);SCO-transect (t-v);KPC-transect (w-y).

Figure S6 Spatial distribution of R (yellow),RMSE (purple) and BIAS (blue or red) of LWU at monthly scale ERA5.All AWSs are located on the GrIS (a).THU-transect (b-d);UPE-transect (e-g);KAN-transect (h-j);NUK-transect (k-m);QAS-transect (n-p);TAS-transect (q-s);SCO-transect (t-v);KPC-transect (w-y).

Table S1 Basic information of AWSs used in this study

Table S2 Daily data evaluation indexes of SWD for five reanalysis datasets for all stations

Continued

Table S3 Daily data evaluation indexes of SWU for five reanalysis datasets for all stations

Continued

Table S4 Daily data evaluation indexes of LWD for five reanalysis datasets for all stations

Table S5 Daily data evaluation indexes of LWU for five reanalysis datasets for all stations

Table S6 Monthly data evaluation indexes of SWD for five reanalysis datasets for all stations

Continued

Table S7 Monthly data evaluation indexes of SWU for five reanalysis datasets for all stations

Continued

Table S8 Monthly data evaluation indexes of LWD for five reanalysis datasets for all stations

Table S9 Monthly data evaluation indexes of LWU for five reanalysis datasets for all stations