Deviations of Boundary Layer Height and Meteorological Parameters Between Ground-Based Remote Sensing and ERA5 over the Complex Terrain of the Mongolian Plateau

Abstract

This study provides a comprehensive evaluation of the vertical accuracy of ERA5 reanalysis data for boundary layer height and key meteorological variables, based on high-precision observational data from Baotou, located on the Mongolian Plateau, during the winter (January–March) and summer (July–August) months of 2021. Results indicate that ERA5 exhibits significant biases in horizontal wind speed, with deviations ranging from −5 to 8 m/s at 50 m, primarily driven by sandstorms in winter and convective weather in summer. The most pronounced errors occur below 500 m. Vertical wind speeds are consistently underestimated in both seasons, with biases reaching up to 1 m/s, particularly during active summer convection. ERA5 also struggles to reproduce low-level wind directions accurately. In winter, correlation coefficients range from 0.43 to 0.64 below 200 m and improve to above 0.7 at 500 m. In summer, correlation coefficients are lower, ranging from 0.3 to 0.5 below 200 m, with reduced accuracy at 500 m compared to winter. Temperature deviations increase above 2000 m, with a relative overestimation of 3% at 3000 m. Relative humidity is generally overestimated by 5–20% between 1000 and 2000 m in winter and by 10–30% in summer. For boundary layer heights, ERA5 overestimates daytime mixed-layer heights by up to 2000 m in summer and 500–800 m in winter. In contrast, ERA5 captures nocturnal stable boundary layer heights well during winter. This comprehensive evaluation of the vertical structure accuracy of ERA5 reanalysis data, conducted in a heavily industrialized city on the Mongolian Plateau, offers essential insights for improving meteorological studies and refining climate models in the region. The findings provide valuable reference data for enhancing weather forecasting and supporting climate change research, particularly in complex terrain areas.

1. Introduction

The planetary boundary layer height (PBLH) and essential meteorological variables—such as temperature, humidity, horizontal and vertical wind speeds, and wind direction—are vital for understanding atmospheric dynamics and the transfer of heat, moisture, and pollutants between the Earth’s surface and the atmosphere [1,2]. The accurate characterization of these variables is essential for reliable weather predictions, air quality modeling, and climate research. Among the widely utilized tools for such studies, the ERA5 reanalysis dataset offers extensive temporal and spatial coverage of these variables, making it an invaluable resource for large-scale meteorological and climate applications. ERA5 (the fifth generation of the ECMWF atmospheric reanalysis of the global climate) is a high-resolution global reanalysis dataset that integrates various observational data and advanced numerical weather prediction models, providing accurate, continuous, and consistent atmospheric and surface meteorological variables [3,4]. It has made significant research progress across multiple fields. Sleem et al. (2024) developed an improved weighted mean temperature (Tm) model, named EGWMT, for accurate Tm estimation at any site in Egypt, utilizing hourly ERA5 reanalysis data from 2008 to 2019 [5]. The EGWMT model was evaluated against two data sources, including ERA5 data from 2019 to 2022 and radiosonde profiles from 2017 to 2022, and was shown to outperform existing models (Bevis, Elhaty, ANN, and GGTm-Ts) with significant improvements in RMSE and mean absolute bias (MAB), confirming its superior performance for GNSS-based precipitable water vapor retrieval. Wilczak et al. (2024) evaluated the applicability of the ERA5 reanalysis for estimating wind and solar energy generation over the contiguous United States, highlighting significant biases in ERA5-derived wind and solar power [6]. They found that ERA5 overestimates solar power and underestimates wind power, with errors varying by season and location. Cardoso et al. (2024) evaluated the performance of the ERA5 reanalysis data in representing precipitation for the Brazilian side of the Mirim-São Gonçalo watershed (MSGW). By comparing ERA5 grid point data with observed rain gauge data from 1981 to 2020, their study found that ERA5 effectively captured precipitation patterns in the region, showing low error measures (MAPE and RMSE) and high correlation (rpearson), demonstrating the model’s potential for water resource management and climate data analyses [7]. Hanesiak et al. (2024) used ERA5 data to analyze tornadic storm environments in Canada from 1980 to 2020 [8]. They found that ERA5 convective parameters accurately represent observed data, with eastern Canada having more favorable conditions for tornado formation due to higher humidity, while central Canada had the highest CAPE. Despite its high spatial and temporal resolution, ERA5 data may lack sufficient detail in regions with complex terrain, such as mountainous plateaus and coastal areas. This limitation can result in errors and biases when representing the meteorological and climatic features of these regions [9,10,11].

A number of studies have investigated the performance and suitability of ERA5 reanalysis data [12,13]. Li et al. (2024) assessed the accuracy and impact of ERA5 reanalysis precipitation data on hydrological cycling in the Ili River Basin (IRB) and tested several bias correction methods, including linear scaling (LS), local intensity scaling (LOCI), power transformation (PT), and distribution mapping (DM) [14]. They found that ERA5 overestimates precipitation, with the DM method providing the most effective bias correction, significantly improving runoff simulations and addressing inaccuracies in peak flow and flow duration curves. Zhi et al. (2013) combined sounding data to study the applicability of upper-air temperature fields in various reanalysis datasets across China, revealing that ERA5’s upper-air temperature data accurately depicted the upper troposphere in northern China [15]. Yoshida et al. (2024) investigated large-amplitude inertia gravity waves (GWs) over Syowa Station, Antarctica, using PANSY radar data and ECMWF reanalysis v5 (ERA5) data from October 2015 to September 2016. Their study found that ERA5 significantly underestimated the absolute momentum flux (AMF) of the GWs, particularly in the stratosphere, with a factor-of-5 underestimation between 5 and 12.5 km altitudes, primarily due to the model’s larger vertical grid spacing and the shorter vertical wavelengths of the dominant GWs [16]. Graham et al. (2019) evaluated the accuracy of ERA5, ERA-Interim, JRA-55, CFSv2, and MERRA-2 in the Arctic, finding that ERA5 had higher correlation coefficients and smaller biases and root mean square errors for temperature, wind speed, and specific humidity compared to other reanalysis datasets [17]. Wei et al. (2024) assessed the vertical accuracy of various meteorological elements in ERA5 data over mountainous regions, finding that ERA5 data overestimated temperatures above 2000 m in mountainous areas [18]. However, the vertical structure accuracy of ERA5 data over the Inner Mongolia Plateau has not been fully evaluated. Therefore, evaluating and analyzing the vertical structure accuracy of ERA5 data in Baotou is of great significance for understanding the meteorological characteristics and changes in this region. The complex terrain near Baotou, bordered by the Yinshan Mountains to the north and the Yellow River to the south, induces intricate local circulations due to topographical forcing, impacting the accuracy of ERA5 reanalysis data. Previous research has indicated discrepancies between ERA5 boundary layer heights and actual heights, as ERA5 boundary layer heights are computed using the Richardson number, which relies on numerous basic meteorological variables [19,20]. Therefore, the accuracy of ERA5’s basic meteorological elements is crucial.

This study utilizes core meteorological variables from ERA5, including temperature, humidity, horizontal and vertical wind speeds, wind direction, and planetary boundary layer height (PBLH), and compares them with high-precision observational data to evaluate the vertical structure accuracy of ERA5 in the complex terrain of Baotou. The inclusion of PBLH data in this comparison is particularly valuable, as the boundary layer height influences the dispersion of pollutants and impacts various atmospheric processes, which are especially crucial in industrial regions with complex terrain like Baotou. This research provides critical reference data for meteorological studies and climate modeling in Inner Mongolian industrial cities, offering reliable scientific evidence to support weather forecasting and climate change studies. The observational period spans January, February, and March (winter) and July and August (summer) of 2021. Extreme weather events, such as heavy rainfall and sandstorms, are common on the Mongolian Plateau, with significant impacts on local agriculture, livestock, and the daily lives of residents, resulting in substantial economic losses. This study analyzes the vertical structure and boundary layer height accuracy of ERA5 reanalysis data, with the goal of improving the accuracy and timeliness of disaster warnings, thereby aiding in more effective responses to extreme weather events.

This study leverages high-precision instruments deployed by the Baotou Municipal Ecological Environment Bureau to observe the vertical profiles of key meteorological variables and uses the observational data to assess the accuracy of the ERA5 reanalysis, particularly in relation to planetary boundary layer height (PBLH). Wind lidar and microwave radiometers (MWRs) have proven to be highly accurate and widely applicable across diverse terrains [18,21], with numerous significant results achieved using these instruments [22,23].

2. Materials and Methods

2.1. Introduction of Observation Sites

Baotou’s geographic location is shown in Figure 1a, with the site located at (40.65°N, 109.84°E; 1067 m A.S.L.). The Windcube 100S Doppler wind lidar (Figure 1c, left) and the RPG-HATPRO-G5 microwave radiometer (Figure 1c, right, hereafter referred to as MWR) are placed at the observation site. The observation period (January–March and July–August 2021) was selected to capture key seasonal variations in the Mongolian Plateau region, including winter sandstorms and summer convective weather. The limited observation period was due to the operational schedule of the Baotou Municipal Ecological Environment Bureau.

2.2. Introduction of Observation Instruments

The Windcube 100S Doppler wind lidar is used to observe 3D wind fields in complex terrain. The wind lidar relies on aerosol backscattered laser signals to provide accurate and reliable results [24,25,26]. The Windcube 100S is a widely used lidar system for obtaining three-dimensional wind fields at observation sites. It typically uses the Carrier-to-Noise Ratio (CNR) as a key parameter for data quality control. Higher CNR values generally indicate that the signal strength is high relative to the noise, suggesting that the measurements are more reliable and accurate. In contrast, lower CNR values may indicate poorer data quality, which could result from factors such as interference, signal attenuation, or environmental disturbances. Data with CNR values below −30 dBZ are discarded, while data at or above −30 dBZ are considered credible [27,28].

The adopted CNR threshold of −30 dB is a widely validated metric for ensuring data reliability, as shown in previous studies [29,30]. The WindCube 100S has a horizontal wind speed accuracy of ±0.1 m/s and a wind direction accuracy of ±2° under optimal conditions [31]. Regular maintenance and calibration were performed to ensure the accuracy of the data during the observation period.

The Windcube 100S with a time resolution of 5 s collects backscattered echo signals from moving particles by emitting laser pulses into the atmosphere, using specific signal processing algorithms to determine the Doppler shift in the backscattered signal and analyzing it to calculate radial wind speed and direction along the measurement path. The lidar has a blind spot from the ground to a height of 50 m, and its theoretical maximum detection range is 3 km, limited by signal attenuation. Detailed information about the Windcube 100S and its detection principles can be found in the Supplementary Materials. Detailed information about the Windcube 100S can be acquired from www.leosphere.com (accessed on 4 December 2024). In this study, wind data were averaged to hourly resolution for comparison.

The RPG-HATPRO-G5 microwave radiometer (MWR) is a sophisticated instrument designed for continuous, real-time monitoring of the thermodynamic structure of the troposphere, extending from the surface to 10 km in altitude. This advanced system provides a comprehensive set of atmospheric data, including the temperature, absolute and relative humidity, total water vapor content, liquid water path, liquid water profiles, cloud base heights, atmospheric stability indices, and surface meteorological variables such as wind speed and direction. It also includes atmospheric pressure measurements at 2 m above the surface [11,25,32,33,34]. These parameters are critical for understanding dynamic processes within the boundary layer and the broader atmospheric system, positioning the MWR as an essential tool for both atmospheric research and operational meteorology. The MWR observations exhibit minimal differences when compared to radiosonde data, demonstrating its suitability for boundary layer studies [35]. It provides reliable data for vertical structure analyses by offering temperature and humidity profiles from the surface up to 10 km. The MWR operates with a high temporal resolution of 1 s, enabling continuous, near-real-time observations. The RPG-HATPRO-G5 MWR is designed to provide high-precision atmospheric measurements. According to the manufacturer’s technical note, the instrument achieves a temperature profile accuracy of approximately 0.25 K RMS in the lower atmosphere (0–500 m) and up to 1 K RMS in higher altitudes (up to 10 km). These accuracies are maintained through regular calibrations, including the use of a precision target for liquid nitrogen (LN2) calibration, which ensures an absolute accuracy of ±0.25 K and a calibration repeatability better than 25 mK. A key advantage of the MWR is its ability to maintain continuous operation even under challenging weather conditions, such as thick cloud cover or precipitation, where other instruments may encounter limitations or produce less reliable data. This robust capability supports a variety of applications, including weather forecasting, atmospheric research, boundary layer analyses, and studies of cloud and precipitation dynamics. The real-time nature of the MWR’s measurements allows for the continuous tracking of atmospheric changes, making it an invaluable tool for meteorologists and atmospheric scientists. The MWR contributes significantly to the study of atmospheric science, particularly in areas such as atmospheric stability, water vapor and cloud dynamics, and boundary layer turbulence. Its high sensitivity and precision in measuring key thermodynamic parameters make it an indispensable instrument for advancing our understanding of weather and climate processes. For more information on the RPG-HATPRO-G5 microwave radiometer, please visit http://www.radiometer-physics.de (accessed on 4 December 2024).

2.3. ERA5 Reanalysis Data

ERA5 (https://cds.climate.copernicus.eu/datasets, accessed on 4 December 2024) is the fifth-generation atmospheric reanalysis dataset developed by ECMWF, covering the global climate from January 1940 to the present [4]. This dataset provides hourly estimates of numerous atmospheric, land, and oceanic climate variables with a spatial resolution of 0.25° × 0.25° and includes 137 vertical levels from the surface up to 80 km, allowing comprehensive atmospheric resolution. Additionally, ERA5 includes uncertainty information for all variables, which is provided at reduced spatial and temporal resolutions [36,37,38,39]. In this study, we used the ERA5 hourly data on pressure levels from the 1940-to-present dataset, which includes the U-component of wind, V-component of wind, vertical velocity, temperature, relative humidity, and geopotential. We also used the boundary layer height data from the ERA5 hourly data on single levels from the 1940-to-present dataset. Bilinear interpolation was applied to obtain site-specific data, followed by cubic spline interpolation to align the data with observation heights.

2.4. Principle of Planetary Boundary Layer Height (PBLH) Retrieval

In this study, the potential temperature (θ) curve is used to retrieve the planetary boundary layer height (PBLH). From the temperature profile, the potential temperature θ is calculated for each height using Equation (1), where T is air temperature, P₀ is the reference pressure, p_(z) is air pressure, R is the ideal gas constant, and C_p is the specific heat capacity at constant pressure.

$$\mathsf{\theta }=\mathrm{T}·{\left({\displaystyle \frac{{\mathrm{P}}_0}{{\mathrm{p}}_{\left(\mathrm{z}\right)}}}\right)}^{{\displaystyle \frac{\mathrm{R}}{{\mathrm{C}}_{\mathrm{P}}}}}$$

(1)

The thermal planetary boundary layer height (PBLH), derived from microwave radiometer (MWR) observations, is represented as H_θ. The retrieval algorithm for H_θ is outlined as follows:

1.

Parcel Condition Check: First, assess whether there exists a height at which the potential temperature θ_(z) is lower than θ₍₀₎. This parcel condition is applicable only under convective conditions [40,41]. When the condition is met, H_θ is defined as the height where θ_(z) = θ₍₀₎, representing the Mixing Layer (ML) height.

2.

Stable Condition for PBLH: If the parcel condition is not satisfied, the PBLH is calculated under stable conditions. In this scenario, H_θ is defined as the top of the stable boundary layer (SBL), where the potential temperature gradient θ′_(z) shows a decreasing trend [20,41]. When the minimum of θ′_(z) is greater than or equal to zero, H_θ is the altitude corresponding to the minimum θ′_(z). Otherwise, H_θ is determined as the height at which θ′_(z) = 0 [11,22,25].

To calculate the pressure at different altitudes, we use the hypsometric equation (also known as the barometric formula), which relates the pressure at a given height to the surface pressure and the temperature profile. The hypsometric equation is given by

$${\mathrm{p}}_{\left(\mathrm{z}\right)}={\mathrm{p}}_0·\mathrm{e}\mathrm{x}\mathrm{p}\left(-{\displaystyle \frac{\mathrm{g}\mathrm{z}}{\mathrm{R}\mathrm{T}}}\right)$$

(2)

where p_(z) is the pressure at the height, p₀ is the surface pressure, g is the acceleration due to gravity, R is the ideal gas constant, T is the temperature at the height, and z is the altitude.

By applying this equation, we can compute the atmospheric pressure at each measurement level from the ground up. This allows us to calculate the potential temperature (θ) for each altitude, which is required for retrieving the planetary boundary layer height (PBLH) as described in the methodology.

3. Results

3.1. ERA5 Horizontal Wind Speed Assessment

Figure 2a,b show the absolute deviations of horizontal wind speed at various heights. In Figure 2a, for January, February, and March, most absolute deviations of horizontal wind speed at 50 m are between −5 m/s and 5 m/s, with some blue outliers. This may be related to several dust storms and sandstorms during the observation period, with specific times of large deviations shown in different colors in Figure S1. The median absolute deviation at 100 m is near 0 m/s. In Figure S1(a1,b1), the horizontal wind speed deviations are small for most of the time series, but significant deviations are noted from 10 January to 18 January, coinciding with a sandstorm event. At 200 m, the median absolute deviation is below 0 m/s, indicating an underestimation of horizontal wind speed by ERA5, with the maximum underestimation reaching 8 m/s. This phenomenon is evident in Figure 3c and is likely due to local circulation influenced by the Yinshan Mountains northwest of the Baotou observation site [42]. At 500 m, ERA5 overestimates horizontal wind speed in winter, as shown by the higher median absolute deviation above 0 m/s and evident blue outliers in Figure 2. At 1000 m, most deviations are above 0 m/s with some outliers reaching 10 m/s while limited observations at 1500 m result in fewer deviation samples, as shown in Figure S5(f1), where deviations within 20 m/s are minor during observed periods. In Figure 2b, for July and August, most absolute deviations at 50 m are between −5 m/s and 5 m/s, with some blue outliers exceeding 5 m/s, due to strong convective weather events such as thunderstorms and heavy rain on 17 July, 25 July, and 18 August, causing low wind speed estimates by ERA5. These phenomena are evident in box plots at all heights in Figure 2b. At 200 m, 500 m, and 1000 m, northwest wind speeds observed are higher than ERA5 estimates, influenced by both strong convective weather and Foehn winds, where northwest winds accelerate over the Yinshan Mountains. Figure 2c,d show many blue outliers below 200 m, indicating the overestimation of horizontal wind speeds by ERA5. Overall, summer’s convective weather causes lower horizontal wind speed estimates by ERA5, with leeward topography also contributing to the underestimation. Winter predictions are better, but high-level deviations still occur.

3.2. ERA5 Vertical Wind Speed Assessment

Figure 3a,b display the absolute deviations of vertical wind speed at various heights. In Figure 3a, for January, February, and March, most absolute deviations of vertical wind speed at 50 m range between −1 m/s and 1 m/s, with several blue outliers related to dust storms and sandstorms, as highlighted in different colors in Figure S2. The median absolute deviation at 100 m is near 0 m/s, while at 200 m, 500 m, and 1000 m, the median absolute deviation is below 0 m/s, indicating an underestimation of vertical wind speed by ERA5. The scatter plots in Figure S2(a2–i2) show a low correlation between ERA5 vertical wind speed and lidar observations. Previous studies also noted significant discrepancies between ERA5 vertical wind speed and actual observations [18].

Figure 3. Absolute and relative deviations of vertical wind speed at various heights. (a) Absolute deviations of vertical wind speed (ERA5 reanalysis data minus Doppler wind lidar observations) for January, February, and March; (b) absolute deviations for July and August; (c) relative deviations (ERA5 reanalysis data minus Doppler wind lidar observations, divided by Doppler wind lidar observations) for January, February, and March; (d) relative deviations for July and August. Red circles and blue dots indicate the distribution of deviation data, with deeper colors indicating denser areas. Blue dots represent outliers, and the light green box plot shows the interquartile range (75th percentile, median, and 25th percentile).

Figure 3. Absolute and relative deviations of vertical wind speed at various heights. (a) Absolute deviations of vertical wind speed (ERA5 reanalysis data minus Doppler wind lidar observations) for January, February, and March; (b) absolute deviations for July and August; (c) relative deviations (ERA5 reanalysis data minus Doppler wind lidar observations, divided by Doppler wind lidar observations) for January, February, and March; (d) relative deviations for July and August. Red circles and blue dots indicate the distribution of deviation data, with deeper colors indicating denser areas. Blue dots represent outliers, and the light green box plot shows the interquartile range (75th percentile, median, and 25th percentile).

In Figure 3b, for July and August, most absolute deviations at 50 m range between −1 m/s and 1 m/s, with some blue outliers exceeding 1 m/s, due to local heating from industrial activities in Baotou, causing air expansion and ascent. Strong convective weather also contributes to deviations, with thunderstorms and heavy rain on 17 July, 25 July, and 18 August leading to underestimations by ERA5. Figure 3c,d show large relative deviations at various heights in both winter and summer.

The relatively low spatial resolution of ERA5 data may smooth out localized strong convective or turbulent phenomena, leading to an underestimation of vertical wind speeds. In contrast, observational data are capable of capturing fine-scale localized vertical motions, particularly in areas characterized by intense convection and turbulence. As a result, such localized dynamics may not be adequately represented or resolved in the ERA5 reanalysis, potentially leading to discrepancies when compared with observational datasets. Overall, the vertical wind speed from ERA5 reanalysis data shows larger deviations during both winter and summer.

3.3. ERA5 Wind Assessment

In Figure 4a–l, discrepancies between observed mean wind speeds and directions at various heights and the corresponding ERA5 reanalysis data are clearly illustrated. The wind rose diagrams (Figure 4a–l) compare mean wind speeds and directions at different heights, while the scatter plots (Figure 5a–l) demonstrate the alignment between observed and ERA5 reanalysis wind directions. At 50 m (Figure 4a), ERA5 tends to overestimate wind speeds, except for the northwest and southwest directions. This overestimation may be related to local terrain effects (such as buildings and land use) that influence near-surface wind speeds but are not fully captured by the ERA5 model. The alignment of observed and reanalysis wind directions is particularly strong in the northwest direction, although discrepancies are observed in the southeast direction, which could be due to local turbulence or wind deflection effects. At 100 m (Figure 4b), ERA5 overestimates mean wind speeds, especially in the west and south directions, with poor alignment of wind directions (Figure 5b) across all directions. This discrepancy may be linked to the relatively low resolution of ERA5, which fails to capture the small-scale terrain features and their influence on wind flow. At 200 m (Figure 4c), ERA5 underestimates wind speeds by 1 to 2 m/s in the east-southeast, northwest, and north directions, which may be due to localized wind channeling effects around the terrain in Kundulun District, Baotou, not fully represented by ERA5. However, as the height increases, the wind speed becomes less dependent on surface roughness, and the alignment of wind directions improves at this height compared to lower levels. At 500 m (Figure 4d), ERA5 overestimates wind speeds by 2 to 3 m/s in the east-northwest and west directions, possibly because the larger-scale atmospheric circulation dominates at this height, though local topographic features still cause some variation. Wind direction alignment is good, with fewer outliers in the northwest and northeast directions, indicating the dominance of larger-scale weather systems. At 1000 m (Figure 4e) and 1500 m (Figure 4f), ERA5 consistently overestimates wind speeds, especially in the west, west-northwest, and northwest-north directions, with errors ranging from 3 to 6 m/s. This could be due to ERA5 capturing the large-scale atmospheric circulation patterns well, but it still fails to fully account for the local terrain influences on wind flow in Kundulun District, Baotou. Wind direction alignment improves at these heights, with prevailing west and northwest winds, though occasional outliers are still observed, possibly due to short-term weather systems [18]. At higher altitudes (Figure 4g–l), the comparison of wind speeds and directions shows that ERA5 consistently overestimates wind speeds, although wind direction alignment improves with altitude as the influence of local terrain decreases. Seasonal variations make the analysis more complex. In winter, significant deviations in wind directions are observed below 200 m, which may be related to cold air accumulation and local atmospheric effects. In summer, poor alignment of wind directions is observed in the southwest direction below 500 m, likely due to local thermal effects, such as valley winds. Overall, discrepancies in wind direction alignment can be attributed to various factors, including underlying surface topography, temperature, and local weather conditions, particularly in Kundulun District, Baotou, where local topographic features have a significant influence on wind flow.

3.4. ERA5 Temperature Assessment

Figure 6a,b illustrate the absolute temperature deviations between the ERA5 reanalysis data and microwave radiometer (MWR)-observed temperatures across various altitudes. In Figure 6a, temperature deviations for January, February, and March are presented, with green dashed boxes highlighting notable blue outliers. These outliers are attributed to frequent sandstorms and severe pollution during winter in Inner Mongolia. The ERA5 reanalysis data, based on model simulations, do not account for the impact of aerosols and only assimilate limited high-altitude atmospheric observation data [43]. In Figure 7a–i, black-circled outliers represent data points under conditions of sandstorms and heavy pollution.

In Figure 6a, during winter, ERA5 overestimates air temperatures at approximately 500 m and above 1000 m while underestimating them below 200 m. In Figure 6b, for summer, the median of absolute deviations remains near 0 K below 1500 m, though temperature deviations increase with altitude above 2000 m. At 3000 m, the ERA5 reanalysis data overestimate temperature by an average of 7 K compared to observations. Additionally, Figure 6b displays several outliers in absolute deviations due to intense convective weather events, such as thunderstorms and hailstorms. In Figure S3, the time series of absolute deviations below 1000 m reveals a significant overestimation of temperatures by ERA5 from 12 July to 15 July during strong convective weather, as indicated by prominent red shading in the temperature series. Persistent rainy weather from 23 August to 30 August results in the underestimation of nighttime temperatures by ERA5 in the absolute deviation time series.

Overall, above 2000 m, ERA5 reanalysis data tend to overestimate temperatures in both winter and summer, with absolute temperature deviations decreasing with increasing altitude during winter. Below 2000 m, ERA5’s overestimation of winter temperatures gradually diminishes, while absolute temperature deviations during summer mostly remain within ±5 K.

3.5. ERA5 Relative Humidity Assessment

Figure 8a,b display the absolute deviations in relative humidity at various heights. In Figure 8a, for winter, absolute humidity deviations from 500 to 2000 m are mostly above 0%, indicating that ERA5 reanalysis data overestimate local relative humidity. Green dashed boxes in Figure 8c highlight outliers between 1000 and 2000 m, indicating poor predictions of relative humidity at these heights by ERA5. Time series in Figure S4(f1–h1) show periods of the overestimation of actual relative humidity by ERA5 reanalysis data during winter.

At the ground level, the majority of winter relative humidity deviation points are above 0%, indicating a slight overestimation by ERA5. Between 50 and 200 m, ERA5 slightly underestimates relative humidity. In the time series of Figure S4, the bottom layer below 100 m (Figure S4(a1–c1)) shows a distinct underestimation during dust storms from January 7 to 17, indicated by purple coloring. Above 1000 m, ERA5 overestimates relative humidity in winter, shown in Figure S4(f1–h1), while at 3000 m, ERA5 underestimates it, as shown by purple outliers in Figure S4(j1).

In Figure 8c, ERA5 reanalysis data underestimate actual relative humidity at 3000 m in winter. Similarly, in summer, Figure 8d shows that ERA5 underestimates relative humidity at 3000 m.

In Figure 8b, for summer, absolute humidity deviations from 500 to 2000 m are mostly above 0%, indicating overestimation by ERA5. Between 50 and 200 m, ERA5 slightly underestimates relative humidity. In the time series of Figure S4(o1–q1), many red points indicate the significant overestimation of relative humidity by ERA5 reanalysis data between 1000 and 2000 m in summer. At 3000 m, ERA5 underestimates relative humidity, as shown by the purple coloring in Figure S4(r1).

Overall, ERA5 reanalysis data overestimate relative humidity between 1000 and 2000 m, while underestimating it at 3000 m.

3.6. Comparison of ERA5 with Observed Meteorological Profile

Figure 9a shows that, for January, February, and March, ERA5 horizontal wind speeds are slightly lower than observed values below 500 m, but higher above 500 m. For July and August, ERA5 overestimates horizontal wind speeds at all heights, with a 4 m/s overestimation at 2000 m.

Figure 9b indicates that, for both January, February, and March, and July and August, ERA5 vertical wind speeds show large deviations above 1000 m, while deviations below 1000 m are smaller. However, Figure 3 reveals that ERA5 vertical wind speed accuracy is low at all heights, with average vertical wind speeds near 0 m/s, not closely matching observations.

Figure 9c demonstrates that, for January, February, and March, ERA5 wind direction shows small deviations below 500 m, with good alignment between 500 and 1000 m. Above 1000 m, deviations increase with height, with average deviations between −15° and 30°. For July and August, ERA5 wind direction shows large deviations below 100 m, good alignment between 100 and 1000 m, and slightly larger deviations between 1000 and 2000 m, with average deviations between −10° and 40°.

Figure 9d shows that, during January, February, and March, ERA5 underestimates temperatures in the lower layers compared to observations, while it overestimates temperatures in the upper layers, with deviations increasing above 1500 m. For July and August, ERA5 overestimates temperatures below 500 m, underestimates them between 500 and 1200 m, and again overestimates temperatures above 1200 m, with deviations increasing with height.

Figure 9e shows that ERA5 relative humidity data exhibit large deviations for both January, February, and March, and July and August. Although ERA5 humidity data follow a similar vertical trend to observed data, significant differences exist between ERA5 and observed humidity values. One reason for ERA5’s humidity inaccuracies is the Foehn effect, where humidity decreases after wind passes over mountains east, west, and north of the observation site [44]. ERA5’s low resolution, especially in complex terrain like Baotou, limits its data accuracy.

3.7. Evaluation of ERA5 Boundary Layer Heights

During winter (Figure 10a,b), the observed boundary layer heights reveal a nocturnal stable layer typically below 500 m, with a daytime mixed boundary layer ranging between 800 and 1500 m. Throughout most of the observation period, ERA5 estimates of the maximum daytime boundary layer height are higher than the MWR-derived values. ERA5 overestimated the mixed-layer height in 43% of the cases when observational data were available. In Figure 10b, the scatter plots overlaid with surface temperature differences indicate that ERA5 surface temperatures are lower than observed surface temperatures during polluted periods in winter over Baotou. From January to March 2021, multiple pollution events occurred, including haze, blowing dust, and sandstorms. These pollution episodes caused a dome effect, where aerosols suppressed surface heating during the daytime, leading to a more stable thermal stratification [45]. The dome effect caused ERA5 to underestimate surface temperatures, while insufficient heating of the atmosphere above the observation site resulted in lower observed daytime thermal boundary layer heights compared to ERA5 estimates. Conversely, when ERA5 overestimated surface temperatures, it also overestimated surface-heating effects. This overestimated heating enhanced buoyancy-driven turbulence, increased vertical mixing intensity, and strengthened boundary layer turbulence, ultimately leading to an overestimation of boundary layer heights.

During summer (Figure 10c,d), the boundary layer height exhibits a nocturnal stable layer mainly ranging between 10 and 350 m, and a daytime mixed layer ranging between 500 and 2100 m. Similarly to winter, ERA5 generally overestimated the maximum daytime boundary layer heights; however, the discrepancies were more pronounced in summer, with maximum daytime overestimations reaching up to 2000 m. ERA5 overestimated the mixed-layer height in 59% of the cases when observational data were available. During July and August, ERA5 overestimated surface temperatures, which further led to an overestimation of the surface-heating effect, ultimately resulting in overestimated boundary layer heights. Additionally, multiple strong convective weather events occurred during this period, such as the intense convection on August 29. Vertical transport induced by mountain airflow and radiative forcing likely contributed to variations in mixed boundary layer heights [21,46,47,48].

4. Discussion

The results show that ERA5 data exhibit relatively small biases in horizontal wind speed at lower altitudes (50 m to 1000 m), but at higher altitudes (above 1000 m), significant overestimations occur, particularly during winter, where wind speed errors increase progressively above 500 m. This indicates that ERA5 tends to smooth out local strong convective or turbulent phenomena, making it difficult to accurately capture atmospheric dynamics. Additionally, during the strong convective weather conditions in summer, ERA5 significantly underestimates horizontal wind speeds, which corresponds to the frequent occurrence of strong convective events such as thunderstorms and heavy rain in this region. These weather phenomena hinder ERA5 from fully reproducing the variability in local wind fields, with the most pronounced discrepancies occurring at lower levels between 50 m and 500 m.

Regarding vertical wind speed, ERA5 reanalysis data consistently underestimate values at all altitudes, especially during the convectively active summer months, where ERA5 fails to capture the local vertical motion induced by convective activity. This underestimation is likely due to the relatively low spatial resolution of ERA5, which cannot resolve fine-scale convective or turbulent phenomena, particularly in regions with complex terrain. In contrast, the Windcube 100S lidar effectively captures the local vertical wind speed variations during strong convective weather, further emphasizing the limitations of ERA5 in reproducing local-scale processes.

In the analysis of temperature and humidity, ERA5 data in the lower layers (500 m to 1000 m) align relatively well with the observations, but above 1000 m, it shows significant temperature overestimation, with the magnitude of the overestimation increasing with altitude. This overestimation is particularly pronounced in summer, where temperature deviations above 2000 m are notably significant, possibly due to ERA5’s inability to accurately represent the modulation of local thermal processes by the terrain. In terms of humidity, ERA5 tends to overestimate relative humidity during both winter and summer, especially at higher altitudes (1500 m to 3000 m), where the overestimation is more pronounced. The Foehn effect may partly explain this phenomenon, as the interaction of local airflow with the complex terrain exacerbates the biases in ERA5 data.

In the comparison of boundary layer heights, ERA5’s performance is closely tied to seasonal and weather conditions. During the winter, ERA5 generally overestimates daytime boundary layer heights, while underestimating nocturnal stable boundary layer heights. This discrepancy is linked to the “dome effect” observed during polluted weather conditions, where aerosols suppress surface heating, limiting the full development of the thermal boundary layer [45]. In summer, the overestimation of daytime boundary layer heights in ERA5 becomes even more pronounced, with deviations reaching up to 2000 m, largely due to overestimated surface temperatures and heating effects. Furthermore, intense convective weather events, such as the thunderstorm on August 29, led to strong updrafts that transported boundary layer air into the free atmosphere, causing a substantial local increase in boundary layer height, which ERA5 failed to capture. Additional factors, such as vertical transport driven by mountain airflow and radiative forcing, further exacerbate the discrepancies between ERA5 data and observations. In Baotou, a geographically complex industrial region, variations in boundary layer height significantly impact pollutant dispersion and other atmospheric processes, making the inclusion of boundary layer height data in comparative analyses crucial. This study offers valuable reference data for meteorological research and climate modeling in Inner Mongolia’s industrial cities, providing essential scientific support for weather forecasting and climate change studies. To overcome the limitations of ERA5, recent advances in artificial intelligence, particularly machine learning, have been applied to enhance its accuracy. For example, machine learning techniques have proven to be effective in reducing errors in boundary layer height estimates in ERA5 reanalysis data [11].

Furthermore, this study reveals that ERA5 struggles to accurately reproduce local meteorological phenomena in complex terrain regions. For instance, when northwesterly winds pass over the Yinshan Mountains, the orographic forcing leads to a significant acceleration of wind speeds, which ERA5 fails to capture, resulting in underestimated wind speeds. The presence of complex terrain increases the complexity of the interactions between local circulations and meteorological variables, and ERA5’s relatively coarse spatial resolution hinders its ability to faithfully represent these localized phenomena.

5. Conclusions

In conclusion, this study provides an in-depth evaluation of ERA5 reanalysis data in the complex terrain of Baotou, highlighting both its strengths and limitations. The results indicate that ERA5 is effective in simulating horizontal wind speeds at lower altitudes, but it struggles with overestimating wind speeds above 1000 m, particularly during winter, and underestimating wind speeds during summer convective events. Vertical wind speeds are consistently underestimated, likely due to the model’s coarse resolution, which is insufficient to capture fine-scale convective processes.

ERA5 performs reasonably well in simulating temperature and humidity in lower layers, but it shows the significant overestimation of temperature and relative humidity at higher altitudes, particularly during summer, which may be related to terrain-induced thermal effects like the Foehn effect. Boundary layer heights are generally overestimated by ERA5, with seasonal and weather-related biases further complicating accurate simulation. These overestimations are particularly problematic during convective weather events and polluted conditions.

The study also demonstrates that ERA5’s performance is compromised in regions with complex terrain, where local effects such as orographic forcing and convective processes are inadequately represented. While ERA5 is a valuable tool for large-scale atmospheric studies, its limitations in capturing local-scale phenomena necessitate the integration of higher-resolution data and regional models, particularly in areas with complex terrain.

Given these findings, it is recommended that ERA5 be used in conjunction with higher-resolution observational data and advanced modeling techniques, such as machine learning, to enhance its accuracy in simulating local meteorological phenomena in complex regions like Baotou. Further research should focus on improving ERA5’s spatial resolution and incorporating new technologies to reduce simulation errors, particularly in capturing the dynamics of convective events and terrain-induced phenomena. This would significantly improve weather forecasting, air quality modeling, and climate studies in regions with similar topographies.