Evaluation of Zenith Tropospheric Delay Derived from ERA5 Data over China Using GNSS Observations

: The latest reanalysis of the European Center for Medium-Range Weather Forecasts (ECMWF), ERA5, can provide atmospheric data for calculating Zenith Tropospheric Delay (ZTD) with hourly temporal resolution, which is a key factor in Global Navigation Satellite System (GNSS) high-precision application. This paper is aimed at evaluating the performance of ZTD derived from ERA5 reanalysis data over China using 219 GNSS stations of the Crustal Movement Observation Network of China (CMONOC) covering the period from 2015 to 2016. The site-speciﬁc hourly ZTD at these stations is obtained by integration method and Saastamoinen model method on ERA5 pressure-level and surface-level reanalysis data with the temporal resolution of 1 h and the spatial resolution of 0.25 ◦ × 0.25 ◦ . Firstly, the atmospheric temperature and pressure that derived from ERA5 are compared with temperature and pressure obtained from meteorological sensors available at 193 GNSS stations. The biases are 2.31 ◦ C and 1.26 mbar implying the accuracy and feasibility of ERA5 pressure and temperature for calculating ZTD over China. Secondly, the performance of ERA5 ZTD is systematically evaluated using ZTD from 219 GNSS sites. The average bias and Root Mean Square (RMS) of ERA5 pressure-level ZTD at all test stations in integration method are approximately 2.97 mm and 11.49 mm respectively, and those of ERA5 surface-level ZTD in model method are 7.97 mm, 39.25 mm, which indicates that ERA5 pressure-level ZTD has a higher accuracy over China. Further analysis indicates that the accuracies of ZTD derived from ERA5 pressure-level and surface-level data are approximately 13.8% and 10.9% higher than those from of ERA-Interim pressure-level and surface-level data. Moreover, ERA5 is able to accurately capture the short-term (hourly) variation of ZTD, which further indicates the better performance of ERA5. Thirdly, the temporal and spatial variation characteristics of ERA5 ZTD accuracy are further analyzed over China. The results show that the ZTD in the southern region has the lower accuracy compared with that in the northern region over China due to the inﬂuence of latitude and altitude. Furthermore, it is found that the ERA5 ZTD over China has obvious seasonality, with higher accuracy in winter and lower accuracy in summer.


Introduction
Tropospheric effect is a major error source in precise global navigation satellite system (GNSS) positioning [1]. Due to its impact on GNSS signal, it should be considered carefully. However, the tropospheric delay is also an important data source for Numerical Weather Model (NMW) since it contains accurate atmospheric information, such as real-time tropospheric delay, atmospheric water vapor and occultation events information and so on [2][3][4]. Thus, high-precision tropospheric delay product is critical for GNSS to improve the accuracy of positioning and water vapor retrieving [5,6].

The Method of Calculating Atmospheric Temperature and Pressure from ERA5
Atmospheric pressure and temperature are the basis for calculating ZTD, which should be firstly evaluated. The reference value is the GNSS meteorological observation data. The vertical and horizontal interpolation should be conducted to calculate the pressure and temperature using ERA5 in order to compare with the meteorological data of GNSS stations. The formula for calculating the atmospheric pressure of grid points at the station height using pressure-level data is as follows. The detailed introduction is in Jade's paper [36]. The linear vertical interpolation method is used to calculate the temperature values at four grid points. The temperature and pressure values at the height of GNSS stations are then obtained using bilinear interpolation method.

The Method of Calculating Atmospheric Temperature and Pressure from ERA5
Atmospheric pressure and temperature are the basis for calculating ZTD, which should be firstly evaluated. The reference value is the GNSS meteorological observation data. The vertical and horizontal interpolation should be conducted to calculate the pressure and temperature using ERA5 in order to compare with the meteorological data of GNSS stations. The formula for calculating the atmospheric pressure of grid points at the station height using pressure-level data is as follows. The detailed introduction is in Jade's paper [36]. The linear vertical interpolation method is used to calculate the temperature values at four grid points. The temperature and pressure values at the height of GNSS stations are then obtained using bilinear interpolation method.
(2) H of grid point j using P i−1 j and P i+1 j at layer i-1 and i+1 respectively. The method of calculating the temperature and pressure of GNSS stations using surface-level data is different from that of using pressure-level data. Temperature and pressure values of four grid points near the GNSS station can be provided at the reference height by surface-level data. Then the atmospheric temperature and pressure values at the height of GNSS stations are calculated using the standard atmosphere model [37]. Similarly, the temperature and pressure values at GNSS stations are obtained according to the following interpolation method.
where T denotes temperature, P denotes atmospheric pressure, P 0 is standard pressure at the reference height H 0 , T 0 is standard temperature at the same height, and H is the height of GNSS station. Before calculating the temperature, pressure and ZTD values, the data need to be checked first, such as the normality of the instrument, the correctness of the data format and the integrity of the collected data. After the time series have been obtained using the selected data, the data with error larger than three times standard deviation is rejected. Additionally, only 5 layers (here refer to the formula for calculating the pressure value based on the pressure-level data at the station location) of data in Network Common Data Form (NetCDF) file are extracted according to the height of the station when atmospheric pressure and temperature at grid points are calculated using pressure-level data, in order to shorten the calculation time.

Calculating ZTD Using ERA5 Meteorological Data at GNSS Stations
Since the location of grid point from ERA5 is different from the location of GNSS sites, the ZTD of the nearest four grid point needs to be calculated firstly, and then the grid ZTD are interpolated to the location of GNSS sites. Generally, the grid point ZTD is calculated based on the ERA5 pressure-level data using the integral method with the formula as follows [38]: where k 1 = 77.604 K/Pa, k 2 = 64.79 K/Pa, k 3 = 377,600.0 K 2 /Pa, P represents the atmospheric pressure, h the specific humidity, N the total refraction, and e the vapor pressure. Then the ZTD is calculated by the following formula: The integral content between different levels is calculated using the trapezoidal formula as: ZTD derived from the surface-level data mainly uses the traditional Saastamoinen method as the following [7].
Remote Sens. 2020, 12, 663 6 of 20 f (ϕ, H) = 1 − 0.00266 cos 2ϕ − 0.00028H (13) where T represents the temperature at the GNSS station, P represents the atmospheric pressure, e represents the partial pressure of water vapor, R h represents the relative humidity, ϕ represents the latitude, and H represents the height of the site. It is worth noting that the reference planes of the geopotential height and GNSS altitude are different, but should be unified [39]. The ZTD of the grid point calculated in this paper is at the height of GNSS stations. Since the surface-level data do not provide relative humidity, it can be calculated using 2 m temperature and 2 m dewpoint temperature. The relative humidity is calculated using the formula given by Haase [40] as where T d is the dewpoint temperature, T is 2 m temperature, esat approx is saturation vapor pressure. In the case that the temperature is higher than 0 • C, R 2 = 6.112 hpa, R 3 = 17.502 K, R 4 = 32.19 K, T 0 = 273.16 K. In the case that the temperature is lower than −23 • C, R 3 = 22.587 K, R 4 = −0.7 K. In the case that the temperature is higher than −23 • C and lower than 0 • C, the calculating formula is as follows where T i = 250.16 K, esat water represents the saturation vapor pressure value in the first case, esat ice represents the saturation vapor pressure value in the second case, and esat represents the saturation vapor pressure value in the third case. The deriving of ZTD from pressure-level data needs two steps: calculating the zenith delay by integration method from Level 1 to Level 37, and calculating the tropospheric delay above the top level using the Saastamoinen model to make comparable results [28]. Then the ZTD of GNSS stations was obtained by interpolation on the basis of the ZTD of grid points.

Deriving ZTD from GNSS
The GNSS data from 2015 to 2016 were processed by BERNESE5.2 software in the post-processing mode to get ZTD values, which were estimated every 1 h. GNSS ZTD can be solved with the strategies described in Table 1 [41,42]. The ZTD time series from ERA5 was compared with that from GNSS so that the residuals could be obtained. Then the deviation (bias), Mean Absolute Error (MAE) and Root Mean Square (RMS) Remote Sens. 2020, 12, 663 7 of 20 error were calculated to evaluate the ZTD accuracy for each station. The temporal and spatial features were analyzed in detail afterwards. The formulas for calculating bias, MAE and RMS are expressed as follows: where ZTD ERA5 i is the ERA5 ZTD at time i, ZTD GNSS i is the GNSS ZTD at time i. N represents the amount of epoch. Bias represents the systematical difference between ERA5 ZTD and GNSS ZTD, MAE and RMS measure the precision of ERA5ZTD against GNSS ZTD.

Temperature and Pressure Evaluation
Meteorological observations from 193 GNSS sites were selected as reference for evaluating the accuracies of the ERA5 pressure and temperature. The atmospheric pressure and temperature derived from ERA5 surface-level data and pressure-level data were calculated at each station throughout 2015 as shown in Figures 2 and 3. Further statistics of bias and RMS of temperature and pressure are shown in Table 2.    It can be seen from Figure 2, Figure 3 and Table 2 that for most stations the atmospheric pressure derived from ERA5 pressure-level data has a higher accuracy than that from ERA5 surface-level data. The accuracy difference of ERA5 temperature and pressure between stations selected for evaluation is relatively small. The temperature derived from ERA5 surface-level data is more accurate than that of pressure-level data in most stations. It can be found that the annual average RMS values of the atmospheric pressure derived from the pressure-level data and the surface-level data are 1.26 mbar and 1.91 mbar, respectively. The annual average RMS values of the temperature from pressure-level data and surface-level data are 3.32 °C and 2.31 °C , respectively. The results imply that ERA5 atmospheric temperature and pressure agree well with the measured temperature and pressure. According to China's climate characteristics, for the commonly used Saastamoinen model, if we use T = 12 °C , P = 1000 mbar, H = 5000 m, = 20°, a pressure error of 1 mbar leads to a ZHD error of 2.2 mm and a temperature error of 1 °C leads to a ZWD error of 6.2 mm. Therefore, the pressure and temperature derived from ERA5 with such accuracy can be used in the calculation of ZTD in China, which is important for GNSS tropospheric delay correction at stations not equipped with meteorological sensors. On the other hand, when just using the temperature or pressure from ERA5 over China in case of availability for both the surface-level and  It can be seen from Figure 2, Figure 3 and Table 2 that for most stations the atmospheric pressure derived from ERA5 pressure-level data has a higher accuracy than that from ERA5 surface-level data. The accuracy difference of ERA5 temperature and pressure between stations selected for evaluation is relatively small. The temperature derived from ERA5 surface-level data is more accurate than that of pressure-level data in most stations. It can be found that the annual average RMS values of the atmospheric pressure derived from the pressure-level data and the surface-level data are 1.26 mbar and 1.91 mbar, respectively. The annual average RMS values of the temperature from pressure-level data and surface-level data are 3.32 • C and 2.31 • C, respectively. The results imply that ERA5 atmospheric temperature and pressure agree well with the measured temperature and pressure. According to China's climate characteristics, for the commonly used Saastamoinen model, if we use T = 12 • C, P = 1000 mbar, H = 5000 m, ϕ = 20 • , a pressure error of 1 mbar leads to a ZHD error of 2.2 mm and a temperature error of 1 • C leads to a ZWD error of 6.2 mm. Therefore, the pressure and temperature derived from ERA5 with such accuracy can be used in the calculation of ZTD in China, which is important for GNSS tropospheric delay correction at stations not equipped with meteorological sensors. On the other hand, when just using the temperature or pressure from ERA5 over China in case of availability for both the surface-level and the pressure-level data, it is recommended that the pressure should be interpolated using the pressure-level data of ERA5, and the temperature be calculated using the surface-level data of ERA5.

Evaluation of ERA5ZTD
The evaluation of ERA5 ZTD was conducted with respect to GNSS ZTD from 219 stations which are uniformly distributed over China. The hourly ZTD time series covers the period from 2015 to 2016. The accuracies of the ERA5 ZTD from pressure-level data and surface-level data are shown in Figure 4, Figure 5 and Table 3 in terms of bias and RMS.

Evaluation of ERA5ZTD
The evaluation of ERA5 ZTD was conducted with respect to GNSS ZTD from 219 stations which are uniformly distributed over China. The hourly ZTD time series covers the period from 2015 to 2016. The accuracies of the ERA5 ZTD from pressure-level data and surface-level data are shown in Figure 4, Figure 5 and Table 3 in terms of bias and RMS.

Evaluation of ERA5ZTD
The evaluation of ERA5 ZTD was conducted with respect to GNSS ZTD from 219 stations which are uniformly distributed over China. The hourly ZTD time series covers the period from 2015 to 2016. The accuracies of the ERA5 ZTD from pressure-level data and surface-level data are shown in Figure 4, Figure 5 and Table 3 in terms of bias and RMS.

ERA5
Pressure-ZTD   The absolute values of the bias calculated by ERA5 pressure-level ZTD compared with GNSS ZTD are less than 6 mm for 90% of 219 stations. It also can be found that the RMS values of difference between ERA5 pressure-level ZTD and GNSS ZTD are less than 15 mm for 92% of 219 stations. The reason the error of few sites seems higher than others in Figures 4 and 5 may be that they are all located in mountainous areas of Qinghai province and Sichuan province, where the terrainis complex and the climate is changeable with the plateau characteristics. The average bias, MAE and RMS of ERA5 pressure-level ZTD of all GNSS stations in statistics are about 2.97 mm, 8.73 mm and 11.49 mm, respectively. Compared with ERA5 surface-level with bias of 7.97 mm, MAE of 31.19 mm and RMS of 39.25 mm, ERA5 pressure-level ZTD agrees much better than surface-level ZTD with GNSS ZTD and is more suitable for tropospheric delay correction and ZTD modeling over China. Thus, the following comparison and analysis are mainly based on ERA5 pressure-level ZTD.

Accuracy Comparison of ERA5 ZTD and ERA-Interim ZTD
To further analyze the performance of ERA5 ZTD over China, this paper compares the ERA5 ZTD with the ZTD derived from ERA-Interim, which has been widely used to develop GNSS tropospheric model. RMS values of ERA5 ZTD and ERA-Interim ZTD compared with GNSS ZTD are shown in Figure 6, Figure 7 and Table 4 The absolute values of the bias calculated by ERA5 pressure-level ZTD compared with GNSS ZTD are less than 6 mm for 90% of 219 stations. It also can be found that the RMS values of difference between ERA5 pressure-level ZTD and GNSS ZTD are less than 15 mm for 92% of 219 stations. The reason the error of few sites seems higher than others in Figure 4 and Figure 5 may be that they are all located in mountainous areas of Qinghai province and Sichuan province, where the terrainis complex and the climate is changeable with the plateau characteristics. The average bias, MAE and RMS of ERA5 pressure-level ZTD of all GNSS stations in statistics are about 2.97 mm, 8.73 mm and 11.49 mm, respectively. Compared with ERA5 surface-level with bias of 7.97 mm, MAE of 31.19 mm and RMS of 39.25 mm, ERA5 pressure-level ZTD agrees much better than surface-level ZTD with GNSS ZTD and is more suitable for tropospheric delay correction and ZTD modeling over China. Thus, the following comparison and analysis are mainly based on ERA5 pressure-level ZTD.

Accuracy Comparison of ERA5 ZTD and ERA-Interim ZTD
To further analyze the performance of ERA5 ZTD over China, this paper compares the ERA5 ZTD with the ZTD derived from ERA-Interim, which has been widely used to develop GNSS tropospheric model. RMS values of ERA5 ZTD and ERA-Interim ZTD compared with GNSS ZTD are shown in Figure 6, Figure 7 and Table 4.       It can be seen from Figures 6 and 7 and Table 4 that ERA5 performs better than ERA-Interim with less average RMS values. The average RMS values of ZTD calculated by ERA-Interim pressure-level data and surface-level data are 13.33 mm and 44.07 mm, respectively, and the average RMS of ZTD calculated by ERA5 pressure-level data and by surface-level data are 11.49 mm and 39.25 mm, respectively. The accuracies of the ZTD calculated by ERA5 pressure-level data and surface-level data are 13.8% and 10.93% higher than those of ERA-Interim respectively.

ERA5, ERA-Interim, GNSS, and GPT2w ZTD Comparison
To further analyze the performance of ERA5 ZTD calculated by integration method, it was compared with GNSS, ERA-Interim, and GPT2w ZTD time series during the two years period from 2015 to 2016. The ZTD time series of above-mentioned four schemes at four stations selected with different latitude and longitude are presented in Figure 8. Figure 9 is the ZTD time series of the first station in Figure 8 during a month in order to further reflect the temporal resolution of four schemes for calculating ZTD.    From Figure 8 and Figure 9, it can be seen the ERA5 ZTD agrees well with the ZTD derived from GNSS. The accuracy and temporal resolution of ERA5 ZTD are higher than those of ERA-Interim. The sub-daily variations of the tropospheric delay can be accurately reflected by the ERA5 which further indicates the better performance of the ERA5 than the GPT2w model and ERA-Interim. Figures 10, 11and 12 show the distribution of annual average RMS and bias of ERA5 ZTD over China. From Figures 8 and 9, it can be seen the ERA5 ZTD agrees well with the ZTD derived from GNSS. The accuracy and temporal resolution of ERA5 ZTD are higher than those of ERA-Interim. The sub-daily variations of the tropospheric delay can be accurately reflected by the ERA5 which further indicates the better performance of the ERA5 than the GPT2w model and ERA-Interim. ERA5 which further indicates the better performance of the ERA5 than the GPT2w model and ERA-Interim. Figures 10, 11and 12 show the distribution of annual average RMS and bias of ERA5 ZTD over China.     It can be seen that the agreements between the ERA5 ZTD and GPS ZTD vary from region to region. The trend of ERA5 ZTD bias in latitude is not so obvious, whereas RMS appears a clear trend. The values of RMS is larger at low latitude region and smaller at high latitude region. It has the same conclusion as Mops's paper that the accuracy of ZTD varies in space, which is mainly related to latitude [43]. Thus, ERA5 ZTD agrees better with GNSS ZTD in the northern region with high latitude than in the southern region with low latitude. In the southern region of China with low latitude, the water vapor change is relatively active, which makes the error of ERA5 ZTD larger, resulting in larger bias and RMS. There are also differences between the east and west of the southern region. This feature is more likely due to the different terrain of the two areas (mountains and basin), which causes large difference of the altitude between grid points and GNSS stations. It can be seen that the agreements between the ERA5 ZTD and GPS ZTD vary from region to region. The trend of ERA5 ZTD bias in latitude is not so obvious, whereas RMS appears a clear trend. The values of RMS is larger at low latitude region and smaller at high latitude region. It has the same conclusion as Mops's paper that the accuracy of ZTD varies in space, which is mainly related to latitude [43]. Thus, ERA5 ZTD agrees better with GNSS ZTD in the northern region with high latitude than in the southern region with low latitude. In the southern region of China with low latitude, the water vapor change is relatively active, which makes the error of ERA5 ZTD larger, resulting in larger bias and RMS. There are also differences between the east and west of the southern region. This feature is more likely due to the different terrain of the two areas (mountains and basin), which causes large difference of the altitude between grid points and GNSS stations. Another important reason is the complicated and changeable water vapor in mountainous region. The trend of ERA5 ZTD MAE is similar as the trend of RMS.

The Spatial and Temporal Distribution Characteristics of ERA5 ZTD Precision
Further, the RMS of ERA5 ZTD compared with GNSS ZTD for 219 selected stations over China is shown in Figure 13. It can be seen from the figure that the lower the latitude of the station is, the larger the RMS of ERA5 ZTD deviation is, which shows the same trend as the former analysis. The reason for this phenomenon may be the tropospheric activity and water vapor distribution. On the one hand, affected by the continental monsoon climate over China, the lower the latitude, the higher the water vapor content is. On the other hand, the tropospheric activity is directly affected by surface heat. In the lower latitude region, the heat is higher and the tropospheric activity is stronger, as a result the uncertainty of tropospheric calculation-related parameters derived from ERA5 increase, and the accuracy of the ZTD decreases [44,45].
reason for this phenomenon may be the tropospheric activity and water vapor distribution. On the one hand, affected by the continental monsoon climate over China, the lower the latitude, the higher the water vapor content is. On the other hand, the tropospheric activity is directly affected by surface heat. In the lower latitude region, the heat is higher and the tropospheric activity is stronger, as a result the uncertainty of tropospheric calculation-related parameters derived from ERA5 increase, and the accuracy of the ZTD decreases [44,45]. one hand, affected by the continental monsoon climate over China, the lower the latitude, the higher the water vapor content is. On the other hand, the tropospheric activity is directly affected by surface heat. In the lower latitude region, the heat is higher and the tropospheric activity is stronger, as a result the uncertainty of tropospheric calculation-related parameters derived from ERA5 increase, and the accuracy of the ZTD decreases [44,45]. As shown in Figure 14, most of the biases values are within ±30 mm which shows the accuracy of these ZTD time series. On the other hand, from the first subfigure in Figure 14, it can be found that the ERA5 ZTD agrees better with GNSS ZTD during winter, with the worst agreement being during summer at all 10 test GNSS stations.
To further analyze the temporal variation features of ERA5 ZTD over China, the month average bias, MAE and RMS of the ERA5 ZTD at all 219 GNSS stations from 2015 to 2016 were calculated, and the results are shown in Figures 15-17. As shown in Figure 14, most of the biases values are within ±30 mm which shows the accuracy of these ZTD time series. On the other hand, from the first subfigure in Figure 14, it can be found that the ERA5 ZTD agrees better with GNSS ZTD during winter, with the worst agreement being during summer at all 10 test GNSS stations.
To further analyze the temporal variation features of ERA5 ZTD over China, the month average bias, MAE and RMS of the ERA5 ZTD at all 219 GNSS stations from 2015 to 2016 were calculated, and the results are shown in Figures 15-17. As shown in Figure 14, most of the biases values are within ±30 mm which shows the accuracy of these ZTD time series. On the other hand, from the first subfigure in Figure 14, it can be found that the ERA5 ZTD agrees better with GNSS ZTD during winter, with the worst agreement being during summer at all 10 test GNSS stations.
To further analyze the temporal variation features of ERA5 ZTD over China, the month average bias, MAE and RMS of the ERA5 ZTD at all 219 GNSS stations from 2015 to 2016 were calculated, and the results are shown in Figures 15-17.      It can be seen from Figure 15 that all monthly average biases are negative and there appears to be a seasonal trend, indicating that the ZTD derived from ERA5 is lower than the GNSS ZTD over China. The monthly average biases is greater in summer, smaller in winter and medium in autumn and spring. Figures 16 and 17 show that the temporal variation trend of MAE and RMS are more obvious. Since MAE and RMS both reflect the precision characteristics of ERA5 ZTD, the monthly average MAE and RMS values have similar trends. It can be found that from January to July, the MAE and RMS have an obvious increasing trend, while from July to December they have a slow decreasing trend in the two years, which indicates their obvious seasonal variation features. Due to the influence of land-sea thermal differences, the climate is complicated and changeable in summer, with heavy rainfall and large water vapor content which leads to the increase of the uncertainty of ERA5 ZTD. As a result, RMS, MAE and bias values in summer are larger than those in winter.

Conclusions
High-precision ZTD plays a vital role in GNSS positioning and meteorological study. Compared with GNSS ZTD, this paper evaluates the accuracy of ERA5 ZTD and analyzes the spatial distribution of ZTD RMS and bias, as well as the seasonal variation feature. The accuracies of It can be seen from Figure 15 that all monthly average biases are negative and there appears to be a seasonal trend, indicating that the ZTD derived from ERA5 is lower than the GNSS ZTD over China. The monthly average biases is greater in summer, smaller in winter and medium in autumn and spring. Figures 16 and 17 show that the temporal variation trend of MAE and RMS are more obvious. Since MAE and RMS both reflect the precision characteristics of ERA5 ZTD, the monthly average MAE and RMS values have similar trends. It can be found that from January to July, the MAE and RMS have an obvious increasing trend, while from July to December they have a slow decreasing trend in the two years, which indicates their obvious seasonal variation features. Due to the influence of land-sea thermal differences, the climate is complicated and changeable in summer, with heavy rainfall and large water vapor content which leads to the increase of the uncertainty of ERA5 ZTD. As a result, RMS, MAE and bias values in summer are larger than those in winter.

Conclusions
High-precision ZTD plays a vital role in GNSS positioning and meteorological study. Compared with GNSS ZTD, this paper evaluates the accuracy of ERA5 ZTD and analyzes the spatial distribution of ZTD RMS and bias, as well as the seasonal variation feature. The accuracies of temperature and pressure are assessed by comparing with measured GNSS meteorological observation data firstly. The bias values between interpolated atmospheric temperature and pressure derived from ERA5 and the measured temperature and pressure data from 193 GNSS stations is about 2.31 • C and 1.26 mbar respectively. It implies that ERA5 pressure and temperature can be used to calculate the ZTD over China with enough accuracy, which is important for GNSS tropospheric delay correction at stations that are not equipped with meteorological sensors.
The performance of ZTD derived from ERA5 is systematically evaluated using GNSS measurements from 2015 to 2016 at 219 GNSS stations of CMONOC. The average bias and RMS of ERA5 pressure-level ZTD at 219 selected GNSS stations are approximately 2.97 mm and 11.49 mm respectively, and those of ERA5 surface-level ZTD are 7.97 mm, 39.25 mm, which indicates that ERA5 pressure-level ZTD has a higher accuracy over China. To further analyze the accuracy improvement of ERA5 ZTD, the calculation of ZTD from ERA-Interim is conducted for comparison. It is shown that the accuracies of ZTD derived from ERA5 pressure-level and surface-level data are approximately 13.8% and 10.8% higher, respectively, than those from of ERA-Interim pressure-level and surface-level data. Moreover, the comparison of ZTD time series of ERA5, GNSS, ERA-Interim, GPT2w shows that ERA5 ZTD agrees well with the GNSS ZTD and ERA5 can well capture the sub-daily variation feature of the ZTD, which indicates the superiority of the ERA5 to the GPT2w model and ERA-Interim.
Additionally, the temporal and spatial variation features of ZTD bias and RMS are analyzed over China. The results show that, due to the effects of latitude and altitude, the accuracy of ZTD in southern China is lower than that in northern China. Furthermore, it is found that the ERA5 ZTD over China has obvious seasonality, with higher accuracy in winter and lower accuracy in summer.
In summary, the latest generation of global reanalysis data, ERA5, has higher precision than ERA-Interim with improved temporal and spatial resolution, which indicates ERA5 is quite promising in tropospheric modeling and application over China. Due to China's vast territory, diverse climate types, and wide range of altitudes, the conclusions drawn in this paper may still be useful for research in other regions. Further research will focus on regional or global modeling of ZTD and long-term analysis on the water vapor variability and trend using ERA5 data.