Validation and Spatiotemporal Analysis of Surface Net Radiation from CRA/Land and ERA5-Land over the Tibetan Plateau

Abstract

High spatial–temporal resolution surface net radiation (RN) data are of great significance to the study of climate, ecology, hydrology and cryosphere changes on the Tibetan Plateau (TP), but the verification of the surface net radiation products on the plateau is not sufficient. In this study, the China Meteorological Administration Global Land Surface Reanalysis Products (CRA/Land) and ECMWF Land Surface Reanalysis version 5 (ERA5-Land) RN data were validated using ground measurements at daily and monthly time scales, and the spatiotemporal patterns were also analyzed. The results indicate the following: (1) CRA/Land overestimated while ERA5-Land underestimated RN, but CRA/Land RN outperformed ERA5-Land in observations at the daily and monthly scale. (2) The CRA/Land RN data had a larger error in the central part and a smaller error in the northeast of the TP, while ERA5-Land showed the opposite. (3) The spatial patterns of RN revealed by CRA/Land and ERA5-Land data showed differences in most regions. The CRA/Land data showed that the RN of the TP had a downward trend during 2000 and 2020 with a slope of −0.112 W·m⁻²/a, while the ERA5-Land data indicated an upward trend with a change rate of 0.016 W·m⁻²/a. (4) Downwelling shortwave radiation (DSR), upwelling shortwave radiation (USR), downwelling longwave radiation (DLR) and upwelling longwave radiation (ULR) are the four components of RN, and the evaluation results indicate that the DSR, DLR and ULR recorded via CRA/Land and ERA5-Land are consistent with the observed data, but the consistency between the USR recorded via CRA/Land and ERA5-Land and the observed data is poor. (5) The inconsistency of the USR data is the main reason for the large differences in the spatiotemporal distribution of CRA/Land and ERA5-Land RN data across the TP.

1. Introduction

Surface net radiation (RN) is the balance between downwelling shortwave radiation (DSR), upwelling shortwave radiation (USR), downwelling longwave radiation (DLR) and upwelling longwave radiation (ULR), and is an essential parameter in the Earth’s surface energy budget [1,2,3]. Meanwhile, RN is the driving force of soil and atmospheric heat, evapotranspiration and plant photosynthetic processes, and it is also an indispensable parameter in land surface, hydrological and ecosystem models [4,5,6,7,8]. Therefore, accurate RN data are of great significance in the study of climate change, water cycle processes and cryosphere degradation.

The methods for measuring RN include ground observations, model simulations, remote sensing retrieval and reanalysis datasets [9,10,11,12,13,14]. Ground observations have the highest accuracy, but there are few RN observation stations, which makes it difficult to satisfy many research demands on the regional or global scale [2,10]. Some simple models can obtain RN through routine meteorological observations [15,16], but the RN obtained via this method is still at site-scale. Sophisticated global climate models can produce RN products on a global scale, but these products usually have low spatial resolution [12]. Remote sensing retrieval can provide high-spatial-resolution RN data. Previous studies have proposed many algorithms for retrieving RN based on Moderate Resolution Imaging Spectroradiometer (MODIS), Spinning Enhanced Visible and Infrared Imager (SEVIRI) and Landsat [17,18,19,20,21]. However, due to the influence of satellite transit time, cloud cover and other factors, the time resolution of remote sensing data cannot easily meet the research needs of some meteorological and hydrological fields. Among these methods, reanalysis data are derived by merging available observations with atmospheric models [22], which can provide high spatial–temporal resolution RN data on a global scale. However, the reanalysis data also need to be evaluated. Several RN products from reanalysis have been evaluated using ground observations on the regional and global scales. Decker et al. [22] evaluated the performance of the Climate Forecast System Reanalysis (CFSR), the 40-year European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40), ECMWF Interim Re-Analysis (ERA-Interim), MERRA and Global Land Data Assimilation System (GLDAS) using flux tower observations of RN, and proposed that all of the datasets had a positive bias in RN. Jia et al. [10] validated the ERA-Interim, Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA2) and the Japanese 55-year Reanalysis (JRA-55) on a global scale and found that these products have similar accuracies.

The Tibetan Plateau (TP) is the highest plateau in the world, with an average elevation of over 4000 m. Strong radiation causes the TP to be a heat source in summer, directly affecting the East Asian monsoon, which then affects the surrounding climate [23,24]. Because several major rivers in East and South Asia, including the Yellow River, Yangtze River and Lancang River, all originate from the TP, it is also known as the “Water Tower of Asia” [25,26]. In addition, the TP is also an important distribution area for mid-latitude permafrost and glaciers [27,28]. Therefore, accurate net radiation data are of great significance to the study of energy and moisture changes and the degradation of the cryosphere in the TP. Although many studies have verified several reanalysis datasets on the TP, they mainly focus on routine meteorological parameters, such as temperature, wind speed and precipitation [23,29], or only on a single component [30], and rarely on the accuracy of RN. RN, as an important parameter in the surface energy budget, has been widely recognized for its importance in the degradation of the cryosphere and the stability of the roadbed in cold regions on the TP [31,32,33]. Therefore, it is necessary to strengthen the evaluation of different RN products to provide more reliable data for this type of research.

With the background mentioned above, RN data from China Meteorological Administration Global Land Surface Reanalysis Products (CRA/Land) and ECMWF Land Surface Re-Analysis version 5 (ERA5-Land) were evaluated at daily and monthly time scales using available in situ observations over the TP. The main purpose of this study is to obtain more accurate net radiation data for the TP. The remaining parts of this paper are organized as follows. Section 2 introduces the RN datasets, as well as the evaluation methods used in this study. Section 3 evaluates the accuracy of the RN products from ERA5-Land and CRA/land at the daily and monthly time scales. Finally, the discussion and conclusions are provided in Section 4 and Section 5, respectively.

2. Materials and Methods

2.1. In Situ Observation Data

A total of 33 RN observation stations were selected in this study, namely Wayanshan (WYS), Wudaoliang (WDL), Xidatan (XDT), Tanggula (TGL), Liangdaohe (LDH), Zhuonaihu (ZNH), Ayakehu (AYK), Tianshuihai (TSH), BJ site of Nagqu Station of Plateau Climate and Environment (BJ), Qomolangma Atmospheric and Environmental Observation and Research Station (QOMS), Southeast Tibet Observation and Research Station for the Alpine Environment (SETORS), Ngari Desert Observation and Research Station (NADORS), Muztagh Ata Westerly Observation and Research Station (MAWORS), Nam Co Monitoring and Research Station for Multisphere Interactions (NAMORS), Qinghai Lake subalpine shrub station (QHH1), Qinghai Lake warm grassland station (QHH2), Qinghai Lake alpine meadow grassland hybrid superstation (QHH3), Qinghai Lake Station (QHH4), A’rou freeze–thaw station (AR1), A’rou north-facing station (AR2), A’rou south-facing station (AR3), Dashalong (DSL), E’bao (EB), Huangcaogou (HCG), Huangzangsi (HZS), Jingyangling (JYL), Yakou snow station (YK), Hulugou (HLG), Grassland Observation Site in the Eling Lake Basin (ELH1), Eling Lake Observation Site (ELH2), Eling Lake Observation Site (ELH3), Sidalong (SDL) and Peicuokuhu (PCKH). The geographic locations of these sites are shown in Figure 1. The observed RN data at XDT, TGL, LDH, ZNH, AYK, TSH, BJ, QOMS, SETORS, NADORS, MAWORS, NAMORS, QHH1, QHH2, QHH3, QHH4, AR1, AR2, AR3, DSL, EB, HLG, SDL and PCKH are available for free download from the National Tibetan Plateau Data Center (http://www.tpdc.ac.cn (accessed on 26 April 2023)). The observed data at ELH1, ELH2 and ELH3 are available for free download from the National Cryosphere Desert Data Center (http://www.ncdc.ac.cn (accessed on 20 April 2023)). The data selected in this study are summarized in

Table 1. Information from the RN observation stations over the TP.

Station	Longitude (°E)	Latitude (°N)	Elevation (m)	Land Cover	Time Interval	Periods
Wayanshan	100.09	37.74	3700	Alpine wet meadow	30 min	2017–2019
Wudaoliang	93.08	35.22	4783	Alpine desert	30 min	2009–2014
Xidatan	94.13	35.72	4538	Alpine meadow	daily	2004–2018
Tanggula	91.94	33.07	5100	Alpine meadow	daily	2004–2018
Liangdaohe	91.74	31.82	4808	Alpine wet meadow	daily	2014–2018
Zhuonaihu	91.96	35.49	4784	Alpine desert	daily	2013–2018
Ayakehu	88.8	37.54	4300	Alpine desert	daily	2013–2018
Tianshuihai	79.55	35.36	4844	Alpine desert	daily	2013–2018
BJ site of Nagqu Station of Plateau Climate and Environment	91.9	31.37	4509	Alpine meadow	10 min	2006–2016
Qomolangma Atmospheric and Environmental Observation and Research Station	86.95	28.36	4298	Alpine desert	1 h	2006–2016
Southeast Tibet Observation and Research Station for the Alpine Environment	94.73	29.77	3327	Alpine meadow	1 h	2008–2016
Ngari Desert Observation and Research Station	79.7	33.39	4270	Alpine desert	1 h	2009–2016
Muztagh Ata Westerly Observation and Research Station	75.05	38.41	3668	Alpine desert	1 h	2010–2016
Nam Co Monitoring and Research Station for Multisphere Interactions	90.98	30.77	4730	Alpine steppe	1 h	2005–2016
Qinghai Lake subalpine shrub station	100.1	37.52	3495	Shrub	10 min	2019–2020
Qinghai Lake warm grassland station	100.24	37.25	3210	Alpine steppe	10 min	2019–2020
Qinghai Lake alpine meadow grassland hybrid super station	98.6	37.7	3718	Alpine meadow and steppe	10 min	2018–2020
Qinghai Lake Station	100.5	36.58	3209	Lake	10 min	2018–2019
A’rou freeze-thaw station	100.46	38.05	3033	Alpine steppe	30 min	2007–2017
A’rou north-facing station	100.41	37.98	3536	Alpine steppe	30 min	2013–2014
A’rou south-facing station	100.52	38.09	3529	Alpine steppe	30 min	2013–2014
Dashalong	98.94	38.84	3739	Alpine meadow	30 min	2013–2017
E’bao	100.92	37.95	3294	Alpine steppe	30 min	2013–2016
Huangcaogou	100.73	38	3137	Alpine steppe	30 min	2013–2015
Huangzangsi	100.19	38.23	2612	Farmland	30 min	2013–2015
Jingyangling	101.12	37.84	3750	Alpine meadow	30 min	2013–2017
Yakou snow station	100.24	38.01	4148	Tundra	30 min	2014–2017
Hulugou	99.87	38.25	3232	Shrub	daily	2012–2013
Grassland Observation Site in the Eling Lake Basin	97.55	34.91	4280	Alpine steppe	30 min	2012–2019
Eling Lake Observation Site	97.65	35.02	4274	Lake	30 min	2012–2013
Eling Lake Observation Site	97.57	34.91	4275	Lake	30 min	2013–2019
Sidalong	99.93	38.43	3146	Forest	10 min	2018
Peicuokuhu	85.59	28.89	4590	Lake	daily	2015–2018

, and the details of these data can be found in the listed references [34,35,36,37,38,39,40].

2.2. Reanalysis Data

ERA5-Land is the land component of the fifth generation of the European Reanalysis (ERA5) from the European Centre for Medium-Range Weather Forecasts [15]. The ERA5-Land data span from 1950 to the present, and are available through the C3S CDS (https://cds.climate.copernicus.eu (accessed on 30 March 2023)). The spatial resolution of ERA5-Land is 0.1° and the temporal resolution is 1 h. CRA/Land is a global reanalysis dataset developed by the China Meteorological Administration, which adopts the latest international assimilation scheme [29]. The CRA/Land data cover the period of 1979 to the present, and can be downloaded from the National Meteorological Science Data Center (http://data.cma.cn (accessed on 30 March 2023)). The spatial resolution of CRA/Land is 34 km and the temporal resolution is 3 h. The ground radiation observations in ERA5-land and CRA/Land are net downward shortwave radiation (SRN), net downward longwave radiation (LRN), DSR and DLR. RN can be obtained by adding SRN and LRN.

2.3. Methods for Validation

The observation data are recorded according to Beijing time, while the reanalysis data are in world time. Before the evaluation, we first converted the observation data from Beijing time to world time. Since XDT, TGL, LDH, ZNH, AYK, TSH, HLG and PKCH can only obtain daily data and cannot convert Beijing time into world time, these eight sites were not used in evaluating the accuracy of the reanalysis data at the daily scale. The nearest grid in the reanalysis data based on the longitude and latitude of the observation stations is found, and the same time series is extracted as the observation data. Daily RN is obtained by averaging hourly data, and monthly RN is obtained by averaging daily data. The accuracy of RN data from ERA5-land and CRA/Land was evaluated using the correlation coefficient (CC), relative bias (RB) and root mean squared error (RMSE) [41]:

$$\mathrm{C}\mathrm{C}=\frac{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{S}}_{\mathrm{i}}-\overline{\mathrm{S}})({\mathrm{G}}_{\mathrm{i}}-\overline{\mathrm{G}})}}{\sqrt{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{S}}_{\mathrm{i}}-\overline{\mathrm{S}})}^2}}\sqrt{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{G}}_{\mathrm{i}}-\overline{\mathrm{G}})}^2}}}$$

(1)

$$\mathrm{R}\mathrm{B}=\frac{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{S}}_{\mathrm{i}}-{\mathrm{G}}_{\mathrm{i}})}}{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{G}}_{\mathrm{i}}}}\times 100\%$$

(2)

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{\frac{1}{\mathrm{n}}{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{S}}_{\mathrm{i}}-{\mathrm{G}}_{\mathrm{i}})}^2}}$$

(3)

The range of CC is from −1 to 1, the ranges of RB and RMSE are from −∞ to +∞ and from 0 to +∞, respectively. A positive RB value means that the reanalysis data overestimates RN, and a negative RB means that it underestimates RN. The closer the CC is to 1, the closer the RB is to 0% and the closer the RMSE is to 0, the more accurate the RN obtained by the reanalysis data will be, compared to that from the in situ observations.

3. Results

3.1. Validation of the CRA/Land and ERA5-Land RN Data

3.1.1. Validation Results at the Daily Scale

The daily RN samples from CRA/Land and ERA5-Land were compared with in situ observations, and the resulting density scatter plots is shown in Figure 2. Since CRA/Land has no RN data when the underlying surface is a lake, there are slightly fewer CRA/Land samples than ERA5-Land samples. Figure 2 shows that the CC, RB and RMSE of the CRA/Land samples were 0.78, 14.70% and 41.84 W·m⁻², respectively, while the CC, RB and RMSE of the ERA5-Land samples were 0.75, −19.78% and 45.98 W·m⁻², respectively. Previously, Jia et al. [2] evaluated the Clouds and the Earth’s Radiant Energy System (CERES) RN data at a global scale, and indicated that the R² (square of CC) of CERES RN data at a daily scale was 0.79 and the RMSE was 33.56 W·m⁻². Considering the high altitude and complex terrain of the TP, it is believed that the CRA/Land and ERA5-Land RN data have high credibility in the TP region. The RB values indicate that the CRA/Land RN data are larger while the ERA5-Land RN data are smaller than the in situ observations. The CC, RB and RMSE values also indicate that the CRA/Land RN data outperformed ERA5-Land in a comparison of the in situ observations at the daily scale.

The validation results of the CRA/Land and ERA5-Land RN at each site are shown in Figure 3. The results show that the CRA/Land and ERA5-Land samples have higher consistency with the in situ observations in the northeast of the TP, and the CC values of most sites are above 0.7 (Figure 3a,b). The RB value of each site indicates that CRA/Land overestimated while ERA5-Land underestimated the RN at most sites (Figure 3c,d). The RMSE value of each station reveals that the CRA/Land RN data have a larger error in the central part and a smaller error in the northeast of the TP, while ERA5-Land showed the opposite (Figure 3e,f). From the average of CC, RB and RMSE at each site, the CRA/Land RN data at the daily scale are generally better than ERA5-Land in the TP.

DSR, USR, DLR and ULR are four components of net radiation, and their accuracy directly affects the applicability of the net radiation data from the Tibetan Plateau. Therefore, the accuracy of these four components was also evaluated. From Figure 4 and Figure 5, it can be seen that the DSR, DLR and ULR recorded via CRA/Land and ERA5-Land are consistent with the observed data, with CC values above 0.79. The consistency between the USR recorded via CRA/Land and ERA5-Land and the observed data is relatively poor, with CC values of 0.33 and 0.40, respectively. The RB values indicate that both CRA/Land and ERA5-Land overestimate DSR and USR, and underestimate DLR and ULR. The RMSE values indicate that the accuracy of the DSR recorded via ERA5-Land is higher than that of CRA/Land, but the accuracy of the USR, DLR and ULR is lower than that of CRA/Land.

3.1.2. Validation Results at the Monthly Scale

The monthly CRA/Land and ERA5-Land RN data were also compared to the in situ observations. The evaluation method of the monthly scale data is consistent with that of the daily scale data. However, more observation sites were available at the monthly scale relative to the daily scale. The resulting density scatter plots are shown in Figure 6. From the evaluation results of the overall data, the CRA/Land and ERA5-Land samples have better consistency and smaller deviation from the observed data on the monthly scale than on the daily scale. Although the CC value of the ERA5-Land data is slightly higher than that of CRA/Land, both the RB and RMSE values of CRA/Land are better than those of ERA5-Land. Therefore, the CRA/Land RN data are generally better than the ERA5-Land’s at the monthly scale, which is consistent with the results obtained at the daily scale. Figure 7 shows the evaluation results for each site, and it can be seen that the CC values at most sites are above 0.9, and the mean values of CC based on ERA5-Land at the monthly scale are close to those of CRA/Land. CRA/Land overestimated while ERA5-Land underestimated RN at the monthly scale, but the absolute values of RB obtained based on CRA/Land and ERA5-Land are very close. The average RMSE obtained based on CRA/Land is slightly lower than that of ERA5-Land, indicating that the accuracy of CRA/Land net radiation data at the monthly scale is slightly higher than that of ERA5-Land.

3.2. Spatiotemporal Analysis

The annual RN was obtained by averaging the CRA/Land and ERA5-Land daily values, and then the spatiotemporal patterns of the mean annual RN over the TP were obtained based on data from 2000 to 2020 (Figure 8 and Figure 9). Spatially, both CRA/Land and ERA5-Land reveal that the annual average RN of the TP is mainly concentrated between 40 W·m⁻² and 120 W·m⁻², and the RN is low in the north and high in the south. However, there are significant differences in the RN patterns in the TP that were revealed by the two datasets. For example, the CRA/Land data indicate that the RN value in the central TP is relatively high (100~120 W·m⁻²), while the ERA5 Land data indicate that the RN value in this area is relatively low (40~60 W·m⁻²). In terms of time, the annual average RN of the TP from 2000 to 2020 based on CRA/Land and ERA5-Land showed opposite trends. The CRA/Land data showed that the RN of the TP showed a downward trend with a slope of −0.112 W·m⁻²/a, while the ERA5-Land data showed an upward trend with a change rate of 0.016 W·m⁻²/a.

4. Discussion

Spatially, there are large differences in RN between the CRA/Land and ERA5-Land data in the TP. In order to analyze the main reasons for this difference, we compared the spatiotemporal distribution of the four components of RN (Figure 10 and Figure 11). It can be seen from Figure 10 and Figure 11 that the spatial distribution and trend of DSR, DLR and ULR in the TP based on CRA/Land and ERA5-Land have good consistency. However, there are large differences in the spatial distribution and trend of USR between CRA/Land and ERA5-Land. The CRA/Land data indicate that the USR values are high in the western and northwestern parts of the TP, but the ERA5-Land data indicate that the USR values are high in the central and western parts of the TP. Figure 11 also shows that the CRA/Land USR in the TP showed an increasing trend from 2001 to 2020, while ERA5 Land indicated a decreasing trend. This result indicates that the inconsistency in USR data is the main reason for the large differences in the spatiotemporal distribution of CRA/Land and ERA5-Land RN in the TP.

The evaluation results based on in situ observations also indicate that the accuracies of the CRA/Land and ERA5-Land USR data are the worst compared to the other RN components. The spatial resolutions of CRA/Land and ERA5-Land are 0.1° and 34 km, respectively. The USR is mainly affected by surface conditions, and the surface conditions represented by this coarse resolution grid may have large differences between the observation stations. This may be an important reason for the inconsistency between the CRA/Land and ERA5-Land USR and the observed data. In addition, the USR data of CRA/Land were obtained through the Noah land surface model (LSM), while the USR data of ERA5-Land were obtained through the ECMWF LSM. From the physical processes of the two models, albedo is a key parameter for calculating USR, and albedo is obtained based on a static monthly climatology [15,42]. However, cold season snow cover and warm season soil moisture are the main factors affecting albedo [43,44,45], and Noah and ECMWF’s schemes make it difficult to accurately describe the changes in albedo caused by snow cover and soil moisture. In particular, in terms of snow albedo, there is a large error in the cold season precipitation data [46]. Coupled with the complex snow blowing process by the wind, it is currently difficult for existing LSMs to accurately simulate the snow cover process in the TP. This is also another important reason why the consistency between the CRA/Land and ERA5-Land USR and observation data is the worst.

5. Conclusions

In this study, we compared the accuracy of CRA/Land and ERA5-Land RN datasets over the TP by comparing them with in situ observations, and analyzed the spatiotemporal pattern of RN. The conclusions can be summarized as follows:

(1) The daily validations for CRA/Land had a CC of 0.78, RB of 14.70% and RMSE of 41.48 W·m⁻², and the monthly validations had a CC of 0.86, RB of 6.83% and RMSE of 28.72 W·m⁻²; the daily validations for ERA5-Land had a CC of 0.75, RB of −19.78% and RMSE of 45.98 W·m⁻², and the monthly validations had a CC of 0.87, RB of −21.03% and RMSE of 31.31 W·m⁻². This means that CRA/Land RN outperformed ERA5-Land when comparing observations at the daily and monthly scale. Spatially, the CRA/Land RN data have a larger error in the central part and a smaller error in the northeast of the TP, while ERA5-Land showed the opposite.

(2) DSR, USR, DLR and ULR are the four components of RN, and the evaluation results indicate that the DSR, DLR and ULR recorded by CRA/Land and ERA5-Land are consistent with the observed data with CC values above 0.79; however, the consistency between the USR values recorded by CRA/Land and ERA5-Land and the observed data is relatively poor. CRA/Land and ERA5-Land overestimate DSR and USR, and underestimate DLR and ULR. The accuracy of the DSR values recorded by ERA5-Land is higher than that of CRA/Land, but the accuracies of its USR, DLR and ULR values are lower than those of CRA/Land.

(3) During 2000 and 2020, both CRA/Land and ERA5-Land reveal that the annual average RN is low in the north and high in the south of the TP, but the spatial pattern of RN revealed by the two datasets shows significant differences in most regions. Trend analysis indicates that the CRA/Land RN of the TP showed a downward trend with a slope of −0.112 W·m⁻²/a, while ERA5-Land data showed an upward trend with a change rate of 0.016 W·m⁻²/a. The uncertainty of the USR data is the main reason for the significant differences in the spatiotemporal distribution of CRA/Land and ERA5-Land RN data in the TP.