Evaluation and statistical bias correction of ERA5-Land meteorological variables for a humid river basin in Southwest China

Abstract

High-quality meteorological data are essential for climate monitoring and renewable energy applications. ERA5-Land, a newly released high-resolution reanalysis dataset, provides a wide range of meteorological variables, but its accuracy remains a concern. This study evaluated the performance of ERA5-Land in the Lower Jinsha River Basin, the largest clean energy base in China, focusing on precipitation, wind speed, air temperature, and solar radiation. A statistical bias correction procedure was developed, combining month-specific regression fitting with daily and hourly adjustments. Results indicated that air temperature estimates agreed best with ground observations, with a coefficient of determination (R²) exceeding 0.87 and percent bias (Pbias) below 15%, followed by solar radiation. Precipitation and wind speed, in contrast, exhibited larger uncertainties (R² < 0.31, Pbias up to 67.76%). After applying the statistical bias correction, systematic biases were largely eliminated across all examined variables. Absolute errors decreased by more than 10%, and temporal consistency also improved moderately, especially for wind speed and solar radiation, where R² increased by 29.5% and 25.8%, respectively. The corrected dataset captured basin-wide climatic variations from 1980 to 2019, including decreasing precipitation, increasing temperature and solar radiation, and the spatial heterogeneity changes in wind speed. Overall, this study contributes to better knowledge of ERA5-Land uncertainties in multiple meteorological variables and provides a practical statistical correction framework, which can serve as a reference for data-scarce regions with similar climatic and geographical conditions and clean energy development contexts.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-24942-4.

Introduction

Meteorological monitoring provides essential information for characterizing the climate distribution and is fundamental for the rational use of natural resources^1,2. The recent rapid expansion of clean energy has further highlighted the importance of meteorological conditions, such as precipitation, wind speed, air temperature, and solar radiation, for power generation³. High-quality observations of these variables are increasingly important for the scientific assessment of clean energy resources and evaluating climatic risks to clean energy systems^4–6.

Traditionally, meteorological data have been obtained primarily from ground-based gauges, which provide accurate point-scale measurements^7,8. However, due to financial and topographical constraints, ground-based observations are often sparse, particularly in remote regions where clean energy power plants are typically located^9,10. The development of reanalysis products offers new opportunities for continuous meteorological monitoring across space and time. By integrating numerical weather prediction models, remote sensing monitoring, and data assimilation, reanalysis datasets can provide extensive coverage, high temporal resolution, and multiple meteorological variables^11–13. Among various reanalysis products, ERA5-Land, developed by the European Centre for Medium-Range Weather Forecasts (ECMWF), has gained widespread attention due to its high spatiotemporal resolution, rapid updating, and comprehensive meteorological variables¹⁴.

Nevertheless, reanalysis estimates are the results of assimilating numerical forecasts with observations, and inherent uncertainties can affect their quality^15,16. Therefore, evaluating the accuracy and reliability of reanalysis products is essential before their practical applications. ERA5-Land is derived from the ECMWF’s ERA5 reanalysis, the successor to ERA-Interim^17,18. Compared with ERA5, ERA5-Land offers higher spatial resolution (~ 9 km) while retaining the exact hourly temporal resolution, and emphasizes land-surface variables with a focus on land-surface variables. This makes it particularly suitable for terrestrial applications like renewable energy power estimation. Previous studies have assessed ERA5-Land at national or regional scales, demonstrating its superior performance compared with several other widely used reanalysis products, including Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2)¹⁹, NASA Prediction of Worldwide Energy Resources (NASA POWER)²⁰, and North American Regional Reanalysis (NARR)¹¹.

Although ERA5-Land shows particular capability in capturing meteorological events and reproducing the spatiotemporal distribution of meteorological variables, biases have been reported in most studies, and performance may vary under different climatic and topographic conditions^21–23. Among the meteorological variables, precipitation has been most extensively studied^12,23–25, while air temperature has also received notable attention^13,26,27. Variables crucial for clean energy development, such as solar radiation²⁸ and wind speed²⁹, have received comparatively less focus.

During the 14th Five-Year Plan, China planned to construct several large-scale clean energy bases based on cascade hydropower stations along major rivers, including the Jinsha, Yalong, and Yellow Rivers³⁰. This initiative provides an unprecedented opportunity for reanalysis products to support the national clean energy development. While existing studies offer some references for ERA5-Land performance, its specific feasibility in clean energy base areas warrants further attention, particularly given the influence of climate and topography. Furthermore, previous studies generally focused on single or a few meteorological variables, underscoring the need for a more systematic understanding of multiple variables that affect different clean energy types and for strategies to manage estimation uncertainties.

To this end, this study focuses on the Lower Jinsha River Basin, China’s largest clean energy base, aiming to evaluate the applicability of ERA5-Land for regional clean energy planning and sustainable development. The specific objectives are: (1) to comprehensively assess the accuracy of ERA5-Land estimates for multiple meteorological variables, including precipitation, wind speed, air temperature, and solar radiation; (2) to develop a statistical bias correction procedure for the above variables to enhance estimate quality at daily and hourly scales; and (3) to investigate the spatial and temporal variations of multiple variables using the corrected ERA5-Land, providing insights for the sustainable development of clean energy bases.

Compared with previous studies that focused on single variables or employed conventional bias correction approaches, this study attempts to provide a more systematic evaluation across multiple meteorological variables and introduces a practical correction framework suitable for data-scarce regions. The findings may serve as a valuable reference for similar geographic settings and for supporting clean energy development.

Materials and methods

Study area

The lower Jinsha River Basin (LJRB), located in southwest China at the junction of the Sichuan and Yunnan Provinces, forms an important part of the upper reaches of the Yangtze River. The basin spans approximately 340,000 km² and is characterized by complex terrain with high mountains and deep valleys, where altitude differences exceed 3500 m (Fig. 1). The LJRB is rich in hydro-energy resources, with a multi-year average runoff of ~ 140 billion m³. Four hydropower stations with a total installed capacity of 46,460 MW form the largest hydropower base in China. In addition, the basin has abundant wind and solar resources, with planned installed capacity of 6767.5 and 8785 MW for wind and photovoltaic power stations, respectively³¹. Together, these developments are expected to transform the LJRB into a major hydro-wind-solar multi-energy complementary base in China.

Fig. 1.

Location of the lower Jinsha River Basin and the distribution of streams, meteorological stations, and ERA5-Land grid cells. The background topographic layer is based on the standard map (No. GS(2020)4619) provided by the Ministry of Natural Resources of China (http://bzdt.ch.mnr.gov.cn), and the boundaries have not been altered.

Despite this resource potential, many planned wind and photovoltaic power stations have not yet been entirely constructed, and long-term meteorological observations at these sites are lacking. Furthermore, ground gauges deployed by the China Meteorological Administration are sparsely distributed, with only eight gauges covering the basin and surrounding areas (detailed information on the ground gauges can be found in Table S1 in the supplementary materials). Consequently, high-resolution meteorological data with accurate magnitudes are urgently required to support the planning and stable development of clean energy in the LJRB.

Meteorological data

This study used two types of meteorological data: ground gauge observations and the ERA5-Land reanalysis product. Basic information on the examined meteorological variables is provided in Table 1.

Table 1.

Information on the examined meteorological variables from the CMA observations and ERA5-Land reanalysis.

Meteorological variable (abbreviation)	Description	Unit
CMA
Precipitation (P)	The total amount of precipitation accumulated over a day	0.1 mm
Wind speed (W)	Mean wind speed of a day at a height of 10 m above the Earth’s surface	0.1 m/s
Mean air temperature (T)	The mean air temperature of a day	0.1 ℃
Maximum air temperature (Tmax)	The maximum air temperature of a day	0.1 ℃
Minimum air temperature (Tmin)	The minimum air temperature of a day	0.1 ℃
Total surface solar radiation (R)	Cumulative component of total radiation irradiance per unit area over a given period	0.01 MJ/m2
ERA5-Land
Total precipitation (P)	Accumulated liquid and frozen water, including rain and snow, that falls to the Earth’s surface	m
10 m u-component of wind (Wu)	The horizontal speed of air moving towards East, at a height of ten meters above the Earth’s surface	m/s
10 m v-component of wind (Wv)	The horizontal speed of air moving towards North, at a height of ten meters above the Earth’s surface	m/s
2 m temperature (T)	Air temperature at 2 m above the Earth’s surface	K
Surface solar radiation downwards (R)	The amount of solar radiation reaching the Earth’s surface comprises direct and diffuse solar radiation	J/m2

Daily-scale observations of precipitation, wind speed, air temperature, and solar radiation were obtained from the China Meteorological Administration (CMA, http://data.cma.cn/) for the period 1980–2019 (1980–2016 for solar radiation). Precipitation data were recorded in two sub-periods, i.e., UTC (Coordinated Universal Time) 12-UTC 00 and UTC 00-UTC 12, while daily wind speed and air temperature were calculated from four observations (UTC 18 on the previous day, and UTC 00, 06, and 12 on the current day). Solar radiation was measured from UTC 16 on the previous day to UTC 16 on the current day. All ground gauge observations underwent standard quality control procedures, including removal of obvious outliers and handling of missing values, to ensure data integrity before being used as a reference.

ERA5-Land is a high-resolution (~ 9 km) reanalysis product developed by the ECMWF, which replays the land component of ERA5 climate reanalysis while retaining hourly temporal resolution. Estimates of the examined variables from 1980 to 2019 were obtained from the Climate Data Store (CDS) (https://cds.climate.copernicus.eu/). Precipitation and solar radiation are accumulated values from UTC 00 to the specific hour of the day, whereas wind speed and air temperature are instantaneous values valid at the specified times. Detailed information on ERA5-Land is available in previous studies and official documentation^32,33.

ERA5-Land estimates were extracted from the grid cell in which the gauge is located. Units of all variables were harmonized. Observations from the eight ground gauges (Table S1) and the paired ERA5-Land grid estimates were then used to evaluate data accuracy and derive statistical correction strategies and coefficients.

Statistical bias correction

Although ERA5-Land has been widely applied as a meteorological data source, it has not been systematically corrected for regional estimation errors and may exhibit substantial biases in specific areas. To address this issue, this study developed a statistical bias correction procedure for multiple meteorological variables in ERA5-Land, suitable for regions with sparse gauge coverage.

As illustrated in Fig. 2, the procedure consists of the following steps: (1) collect ground gauge observations and corresponding ERA5-Land grid estimates; (2) aggregate hourly ERA5-Land data into daily values, including accumulated precipitation and solar radiation, mean wind speed, and mean, maximum, and minimum air temperatures; (3) develop daily-scale bias correction strategies for each calendar month, specifying the correction mode, function form, and associated parameters (see “Daily-scale bias correction strategy” section for details); (4) compute hourly-scale correction coefficients, including scaling factor and offsetting factor, based on comparisons between daily statistics of raw and corrected ERA5-Land data (see “Hourly-scale correction under bias-corrected daily statistics” section); and (5) generate the corrected ERA5-Land dataset by applying the daily- and hourly-scale corrections uniformly across all grids covering the study area.

Fig. 2.

The correction process for the ERA5-Land dataset.

Daily-scale bias correction strategy

Two correction modes were employed to quantitatively capture the monthly patterns of estimation bias for the examined meteorological variables.

Mode 1 adjusts ERA5-Land estimates based on their own estimation errors by modeling the relationship between the ERA5-Land values and the discrepancies relative to ground gauge observations (Eq. 1). This mode is suitable for variables with relatively stable bias patterns over time.

Mode 2 performs a direct correction by establishing the relationship between ERA5-Land estimates and the corresponding ground gauge observations (Eq. 2). This mode is appropriate when the ERA5-Land biases exhibit spatial or temporal variability, enabling more targeted corrections informed by actual measurements.

1

2

where denotes the bias-corrected daily ERA5-Land estimate; and represent the daily meteorological values from ERA5-Land and ground gauges, respectively; is the estimation error, calculated as the difference between ERA5-Land and ground gauge observations for each day of month; and denote the mathematical relationship in Mode 1 (between ERA5-Land estimates and estimation errors) and Mode 2 (between ERA5-Land estimates and ground observations), respectively; i and j correspond to month i and day j, respectively.

The daily correction model for each meteorological variable in each calendar month was separately constructed. For each mode, seven candidate function forms were considered to quantify the monthly relationships, either between ERA5-Land estimates and their errors (Mode 1) or between ERA5-Land estimates and ground observations (Mode 2), as listed in Table S2 in the supplementary materials. The function with the highest coefficient of determination was then selected for the monthly bias correction.

Modes 1 and 2 were eventually jointly applied to the statistical bias correction of precipitation and solar radiation, while Mode 2 was applied to wind speed and the three air temperature-related factors (Fig. S1). The cubic function was found to hold a generally superior performance in all meteorological variables and all months, with the specific parameters provided in Table S3 in the supplementary materials.

Hourly-scale correction under bias-corrected daily statistics

The hourly-scale correction procedure was developed based on the bias-corrected daily statistics of the meteorological variables, which include daily accumulated, mean, maximum, and minimum values depending on the variable type. For precipitation and solar radiation, the daily accumulated values were used to constrain the hourly correction. Specifically, a scaling factor was calculated as the ratio of the bias-corrected daily accumulated value and the raw daily accumulated value for each day of each month, and this factor was applied to adjust the raw hourly ERA5-Land estimates (Eq. 3). The hourly correction for wind speed followed a similar approach, with daily mean values instead of accumulated values. For air temperature, where daily mean, maximum, and minimum values are available from ground gauges, both scaling and offsetting factors were applied to adjust the hourly estimates (Eq. 4).

3

4

where and are the scaling and offsetting factors, respectively; and denote the daily accumulated or mean values of the bias-corrected and raw ERA5-Land estimates, respectively; , , and are the maximum, minimum, and mean value of the bias-corrected daily ERA5-Land estimates, respectively; , , and are their corresponding raw statistics; n represents the number of ground gauge observations used to compute daily air temperature statistics (i.e., UTC 18 on the previous day, UTC 00, 06, and 12 on the current day), and t denotes the observation times excluding those at which the daily maximum and minimum occurred. P and are the raw and corrected hourly ERA5-Land estimates, respectively. Indices i, j, and k refer to the kth hour on day j in month i.

Statistical methods

The capability of ERA5-Land to capture the temporal dynamics of the meteorological variables was evaluated using several statistical metrics, including the coefficient of determination (R²), root mean square error (RMSE), relative RMSE (rRMSE), mean absolute error (MAE), relative MAE (rMAE), and percent bias (Pbias) (Table 2). R² was used to assess the consistency between ERA5-Land estimates and ground gauge observations. RMSE and MAE quantified the absolute errors, with RMSE giving greater weight to larger deviations and thus being more sensitive to outliers, while MAE reflected the average error magnitude. rRMSE and rMAE normalized errors across variables of different magnitudes, enabling direct comparison of estimation accuracy. Pbias reflected the relative bias of the estimates.

Table 2.

Statistical indices used for evaluating the performance of ERA5-Land estimates.

Statistic	Formula	Range	Optimal value
R2		[0, 1]	1
RMSE		[0, +∞)	0
rRMSE		[0, +∞)	0
MAE		[0, +∞)	0
rMAE		[0, +∞)	0
Pbias		(− ∞, +∞)	0

The coefficient of variation (CV) was employed to investigate performance differences of ERA5-Land across diverse terrain and climatic conditions. CV, calculated as the ratio of the standard deviation to the mean of a metric across paired grids, provides a normalized measure of variability independent of variable magnitude. The improvement ratio (IR), defined as the ratio of the change in each index after correction to its original value, was used to quantify and visually compare the effectiveness of the bias correction procedure.

To analyze spatiotemporal variations in the meteorological variables, daily observations were first aggregated to monthly values and subsequently to annual values. The Mann-Kendall test was applied to detect trends using the Z statistic³⁴. Linear regression was employed to quantify long-term changes, with the slope representing the annual change rate. Statistical significance was determined using p-values, where p < 0.05 indicates a significant trend³⁵.

and denote the observed precipitation from the ground gauge and the corresponding ERA5-Land grid cell, respectively; represents the mean of the gauge observations, and N is the total number of samples.

Results

Evaluation of the ERA5-land estimates

Based on the obtained ERA5-Land estimates and ground gauge observations, the ability of ERA5-Land to represent the dynamics of precipitation, wind speed, air temperature, and solar radiation was quantitatively assessed. Figure 3 presents scatter plots of daily ERA5-Land estimates against ground gauge observations together with performance assessment indices.

Fig. 3.

Daily-scale scatter density plots of (a) accumulated precipitation (P), (b) mean wind speed (W), (c) mean air temperature (T), (d) maximum air temperature (T_max), (e) minimum air temperature (T_min), and (f) accumulated solar radiation (R) from the ERA5-Land dataset versus ground gauge observations during 1980–2019. The blue-to-red color scale represents the data density from low to high.

It can be found that the accuracy of ERA5-Land varied among different variables. Precipitation and wind speed exhibited relatively weak agreement with gauge records, with R² of 0.31 and 0.22 and Pbias of 67.76% and − 23.36%, respectively (Fig. 3a and b). In contrast, air temperature-related metrics were reproduced with high fidelity: daily mean, maximum, and minimum values all show R² exceeding 0.87, absolute errors below 4 °C, and systematic underestimations smaller than 18% (Fig. 3c–e). Solar radiation also moderately agreed with ground observations (R² = 0.63, Fig. 3f).

Figure 4 compares evaluation metrics across different paired grid cells. rRMSE and rMAE were used instead of RMSE and MAE to eliminate the influence of magnitude differences among meteorological variables. To quantify spatial variability in ERA5-Land performance, the coefficient of variation (CV) of each metric across grids was calculated (see “Statistical methods” section). Overall, temporal consistency between ERA5-Land estimates and ground gauge observations was relatively stable across grids, with CV(R²) below 10% for all examined variables except wind speed (37.7%). Spatial variability in absolute errors was most pronounced for air temperature-related variables, where CV(rRMSE/rMAE) exceeded 30%, followed by wind speed (< 20%), precipitation (< 15%), and solar radiation (< 10%). For Pbias, substantial inter-grid variations were observed for wind speed and for mean and minimum air temperature, with CV(Pbias) values exceeding 80%.

Fig. 4.

Bar charts of (a) R², (b) rRMSE, (c) rMAE, and (d) Pbias for the paired grids based on daily records from 1980–2019. Note that data in only one paired grid were available for solar radiation as depicted.

Specific correlations between evaluation metrics and elevation were also observed in this study. Specifically, absolute errors in precipitation showed moderate negative correlations with elevation, with r of − 0.58 and − 0.45 for rRMSE and rMAE, respectively (see Fig. S2(a) in the supplementary materials). Temporal consistency of wind speed was enhanced significantly with higher elevation (r = 0.79 for R²). Whereas, estimation errors in air temperature-related variables were more pronounced at higher altitudes, particularly for minimum air temperature (r > 0.60). Moreover, larger elevation mismatches between ERA5-Land grid cells and ground gauges were associated with increased absolute errors in mean and maximum air temperature (r > 0.58, Fig. S2(b)) and significantly reduced Pbias (r<− 0.77).

The performance of daily meteorological estimates from the ERA5-Land dataset exhibited notable seasonal variations (Fig. 5). Specifically, the temporal consistency of wind speed and minimum air temperature was stronger from May to October compared with other months, with R² of 0.23 versus 0.14 on average and 0.69 versus 0.60, respectively. In contrast, precipitation (0.26 versus 0.38) and maximum air temperature (0.76 versus 0.82) showed lower consistency during May-October than in other months. Estimation errors, as indicated by rRMSE, rMAE, and Pbias, were generally less than half during May to October compared to the remaining months for all examined meteorological variables, except for solar radiation. Regarding specific variables, precipitation exhibited pronounced seasonal performance variations, with CV of R² exceeding 25% and CVs of estimation errors exceeding 40%. Wind speed displayed substantial variability in R² (CV > 30%) and moderate variation in Pbias (CV = 18%). Air temperature-related factors and solar radiation also showed apparent monthly differences in estimation errors (CV > 30%). These findings indicate that seasonal variations should be considered when using ERA5-Land estimates for meteorological analyses.

Fig. 5.

Monthly variations of (a) R², (b) rRMSE, (c) rMAE, and (d) Pbias indices for multiple meteorological variables from the raw (solid lines) and bias-corrected (dashed lines) ERA5-Land estimates during 1980–2019. Pbias values for the raw precipitation (P) are plotted on a separate light-blue axis due to their distinctly larger magnitude, while Pbias values for wind speed, air temperature, and solar radiation are shown on the left-side axis.

Statistical bias correction of the ERA5-land estimates

The effectiveness of the statistical bias correction procedure in improving ERA5-Land estimates across multiple meteorological variables was evaluated. Figure 6 displays scatter density plots of bias-corrected ERA5-Land estimates against ground gauge observations, along with the corresponding improvements in performance metrics. The results indicate that the constructed correction strategy for each calendar month effectively enhanced the accuracy of ERA5-Land estimates, particularly in reducing systematic errors. Specifically, Pbias for all examined variables was reduced to a negligible level (~ 0%), and absolute errors (indicated by RMSE and MAE) decreased by more than 10% for most variables. The correction procedure also improved the temporal consistency between ERA5-Land estimates and ground observations, with R² increases ranging from 0.3% to 29.5% across different variables. Wind speed and solar radiation exhibited the most notable improvements in R² (increased by 29.5% and 25.8%, respectively), while maximum air temperature showed the largest reductions in RMSE and MAE, both exceeding 30%.

Fig. 6.

Scatter density plots similar to Fig. 3, but using the bias-corrected ERA5-Land estimates. The improvement ratio (IR) of each evaluation metric is indicated in red.

As shown in Fig. 5, the statistical bias correction improved the performance of ERA5-Land estimates in nearly all months, although the magnitude of improvement varied across meteorological variables and seasons. The increase ratio in R² was most pronounced for wind speed, with an average increase of 10.3%, followed by precipitation (4.5%), for which R² increased more than twofold from January to April and November to December compared to May to October. In contrast, the improvements of temperature-related factors and solar radiation were relatively modest. Regarding estimation errors, both rRMSE and rMAE showed substantial reductions for precipitation and maximum air temperature, with average decreases exceeding 25%. The reductions in absolute errors were generally larger from January to April and November to December than from May to October for most examined variables. Pbias was substantially decreased for all examined meteorological variables across almost all months.

The hourly-scale correction procedure, constrained by bias-corrected daily statistics, also modified the intra-day patterns of the raw ERA5-Land estimates (Fig. S3), bringing them closer to the daily-scale statistical characteristics derived from ground observations.

The long-term meteorological variations in the LJRB

Applying the corrected ERA5-Land dataset, long-term interannual and intra-annual variations of meteorological variables from 1980 to 2019 were examined. Figure 7 illustrates the spatiotemporal distributions of the variables analyzed in this study. The climatological mean precipitation in the LJRB is 909 ± 183 mm, with higher accumulations observed in some southern areas (> 1000 mm/yr). Wind speed and solar radiation exhibited similar spatial patterns, both being generally stronger in the southern part of the basin compared to the north (Fig. 7d, g, and j). The annual mean air temperature and accumulated solar radiation exceeded 10 °C and 1.5 10⁵ W/m² in more than 80% and 65% of the basin, respectively.

Fig. 7.

Long-term climatological mean values (a, d, g, and j), change rates (b, e, h, and k), and trends (c, f, i, and l) of precipitation, wind speed, air temperature, and solar radiation in the LJRB during 1980–2019. The trends were categorized as no trend (NT), significant increase (SI), nonsignificant increase (NI), significant decrease (SD), or nonsignificant decrease (ND). The base map is sourced from the Standard Map Service of the Ministry of Natural Resources of China (No. GS(2020)4619) without boundary modifications.

Regarding long-term trends, almost the entire basin experienced a decrease in precipitation from 1980 to 2019, with an areal mean change rate of − 2.4 mm/yr (p < 0.05), and approximately one-third of the basin area exhibited a statistically significant decline (Fig. 7b and c). Wind speed showed an increasing trend in the northwestern areas but decreased in the southeastern areas (Fig. 7e and f). Air temperature increased across the LJRB, with an areal mean change rate of 0.03 °C /yr (p < 0.05), and significant warming trends were observed in nearly the entire basin (Fig. 7h and i). Solar radiation also showed an upward trend, increasing by 1078 W/m² yr (p < 0.05) across the basin (Fig. 7k and l). The long-term interannual variations derived from the corrected ERA5-Land estimates closely matched the ground gauge observations, both in terms of overall trends and magnitudes (Fig. S4).

The changing trends in each calendar month were further analyzed (Fig. 8). Precipitation decreased in most months, particularly in February and November (p < 0.05). Wind speed increased significantly in November and December. Air temperature exhibited positive trends in all months, especially from May to October, with change rates ranging from 0.02 to 0.04 °C /yr (p < 0.05). Solar radiation also increased in most months, except for March and December.

Fig. 8.

Monthly change rates and trends of (a) precipitation, (b) wind speed, (c) air temperature, and (d) solar radiation in the LJRB from 1980 to 2019. Trends were determined using the Mann-Kendall Z statistic and were categorized as SI, NI, SD, ND, and NT.

At the intra-annual scale, the examined meteorological variables exhibited distinct seasonality (Fig. 9). Precipitation in the LJRB was predominantly concentrated from June to September, accounting for over 70% of the total annual amount. Wind speed was higher from February to April, with a monthly mean exceeding 2.5 m/s, but was substantially lower during summer (June to August) and autumn (September to November). Air temperature peaked in summer, while solar radiation was abundant during spring and summer. The intra-annual distribution patterns derived from ground gauge observations and the corresponding ERA5-Land grid data were found to be highly consistent after applying the bias correction (Fig. S5).

Fig. 9.

Boxplots of intra-annual variations in (a) precipitation, (b) wind speed, (c) air temperature, and (d) solar radiation over the LJRB during 1980–2019.

As shown in Fig. 10, the intra-day patterns of precipitation, air temperature, and solar radiation exhibited larger fluctuations from June to September, whereas wind speed showed more pronounced diurnal variations from January to May. Precipitation was primarily concentrated at night; wind speed and air temperature peaked between 13:00 and 18:00 local time; and solar radiation reached maximum values between 11:00 and 13:00 local time.

Fig. 10.

Intra-day variations of (a) precipitation, (b) wind speed, (c) air temperature, and (d) solar radiation across different months during 1980–2019. Time is shown in Local Standard Time.

Overall, the development of clean energy in the LJRB should prioritize the southern part of the basin, where wind and solar resources are more abundant. The complementarity of wind and solar resources is mainly reflected at the seasonal scale. Given that the existing hydropower plants in the LJRB have seasonal or annual regulation capacity, coordinated operation of hydro, wind, and solar power at medium- and long-term scales appears feasible. However, wind and solar resources are highly synchronized at the intra-day scale, both peaking during the daytime, with only limited wind resources available at night. This implies that hydropower will be subject to greater short-term regulation pressure. Such challenges may be further intensified under climate change, as reduced precipitation would decrease reservoir inflows and thereby weaken hydropower’s regulating capacity.

Discussion

Performance and uncertainties of ERA5-Land

ERA5-Land exhibited significant variability in the accuracy of different meteorological variables^36,37. Estimates of air temperature and solar radiation were notably more reliable than those of precipitation and wind speed over the LJRB (Fig. 3). These differences are closely related to the physical parameterizations in the ECMWF numerical weather prediction model. Temperature and solar radiation are primarily governed by relatively stable processes such as radiation, convection, and heat transfer, which can be reasonably simplified in the model³⁸. In contrast, precipitation and wind speed are controlled by highly transient and nonlinear processes that are more difficult to represent accurately with limited parameterizations^2,39.

Topography further contributes to the estimation uncertainties of ERA5-Land. Previous studies have demonstrated the influence of elevation on the accuracy of reanalysis data, particularly in high-altitude regions (e.g., > 1000 m)^40–42. Consistent with these findings, this study indicates that at higher elevations, precipitation and wind speed estimates tend to be more accurate, whereas air temperature estimates, especially minimum air temperature, exhibit more pronounced biases (Fig. S2(a)).

In addition, mismatches between ground gauge and ERA5-Land grid elevations represent another source of uncertainty. In the LJRB, gauge-grid elevation mismatches ranged from 0.43 to 23.12%, with a mean of 9.54% (Table S1). Such mismatches were moderately correlated with larger absolute errors in air temperature (Fig. S2(b)), suggesting that uncertainties arise not only from model physics but also from scale representativeness between point measurements and gridded estimates. By contrast, precipitation and wind speed estimates were less sensitive to elevation mismatches, implying that other factors, such as convective parameterization and local circulation, are more influential.

Taken together, both large-scale elevation gradients and local gauge-grid mismatches contribute to ERA5-Land uncertainties in complex terrain. These results underscore the importance of accounting for topographic influences when applying ERA5-Land in practical studies. Due to the limited number of ground stations sustaining solar radiation records, the role of elevation in shaping solar radiation uncertainties remains to be further investigated. Overall, compared with air temperature and solar radiation, precipitation and wind speed exhibit markedly higher uncertainties, particularly in low-altitude regions, and should therefore receive greater attention in future applications.

Effectiveness and limitations of the statistical bias correction procedure

The statistical bias correction procedure applied in this study proved highly effective in reducing systematic biases (Pbias), bringing them to nearly negligible levels across all variables. Absolute errors (RMSE/rRMSE and MAE/rMAE) also decreased consistently, while temporal consistency (R²) increased to different extents (Fig. 6). The improvements were most evident for systematic biases, whereas gains in absolute error magnitudes and temporal consistency were generally more modest.

Bias correction further enhanced the agreement of long-term variation trends and change rates with ground observations (Figs. S3-S4). Such alignment is critical for quantitative resource assessment and operation planning of clean energy bases, where both accurate magnitudes and reliable long-term dynamics of meteorological drivers are required. Moreover, the corrected estimates reproduced interannual and intra-annual variation patterns consistent with existing studies^43,44, reinforcing that the procedure yields reliable long-term daily estimates.

To extend daily-scale improvements to sub-daily time series, the procedure incorporated an hourly-scale correction step. Although comprehensive validation was constrained by the lack of high-frequency ground observations, indirect evidence supports its plausibility. For air temperature and solar radiation, the hourly sequences showed reasonable agreement with reference information. Specifically, the timing of daily maxima and minima of air temperature followed the expected diurnal cycle commonly recognized in climatology, and solar radiation peaks were closely matched with ground observations (Fig. S6). The corrected intra-day distribution of the examined variables also agreed broadly with the previous analyses^45,46. These findings suggest that uncertainties in intra-day variability of corrected hourly ERA5-Land estimates are likely limited, especially for air temperature and radiation. Nevertheless, uncertainties in wind speed and precipitation at the sub-daily scale remain to be addressed, and rigorous validation with high-frequency data or alternative high-resolution references is required in future work.

Despite the certain improvements achieved, it should be noted that the correction procedure has several limitations. First, while systematic biases were mainly eliminated, residual uncertainties in absolute error magnitudes and temporal variability persist. Second, the current correction strategy was developed using all available gauges collectively, without stratification by elevation-related conditions, which may limit its ability to address the heterogeneous uncertainty characteristics across diverse terrain settings. Third, in topographically complex regions such as the LJRB, the sparse gauge network cannot fully represent local heterogeneity, meaning that correction strategies derived from limited observations may not adequately reflect uncertainty patterns under the topographic conditions absent from the gauge network. These constraints highlight the need to explore more advanced approaches, such as elevation-stratified corrections, multi-source data integration, or hybrid statistical and machine learning techniques, to enhance the robustness and spatial transferability of bias-corrected reanalysis data in complex terrain.

Methodological comparison and applicability

The statistical bias correction procedure developed in this study builds upon regression-based approaches but extends beyond simple scaling. The procedure captures seasonal variability and potential nonlinearities that single-parameter scaling may not adequately address by incorporating month-specific regression relationships with multiple candidate functional forms. Compared with distribution-based methods such as quantile mapping, which generally require long temporal records and relatively dense spatial station networks to ensure reliable distributional mappings, the proposed procedure is more practical in regions with sparse observational coverage. Nevertheless, quantile mapping or hybrid methods may be preferable to replicate higher-order distributional characteristics, particularly for extreme events. It should be noted that the focus of this study was on developing and validating a pragmatic framework for data-sparse scenarios, whereas a comprehensive intercomparison of multiple techniques remains a subject for future work.

Despite being developed and applied in the LJRB, the framework has broader applicability and can guide systematic bias correction in other regions with limited ground observations. The specific correction parameters derived for the LJRB may also serve as references for nearby or climatically and topographically similar areas. While the procedure still inherits uncertainties associated with sparse gauge density and spatial heterogeneity, its operational simplicity and relatively low dependence on dense gauge networks make it a pragmatic means to enhance the feasibility of ERA5-Land estimates in less- or non-gauged areas. This is particularly valuable for supporting applications in clean energy system assessment and planning, where long-term reliable meteorological information is essential.

Conclusions

This study comprehensively evaluated the performance of ERA5-Land in the Lower Jinsha River Basin (LJRB) for four fundamental meteorological variables, i.e., precipitation, wind speed, air temperature, and solar radiation, and developed a statistical bias correction procedure that combines month-specific regression fitting with daily- and hourly-scale adjustments.

Results indicate that ERA5-Land exhibited variable-dependent performance: air temperature and solar radiation were reproduced reasonably well, whereas tprecipitation and wind speed suffered from larger biases and lower correlations with ground observations. After applying the proposed correction procedure, systematic errors were substantially reduced across all variables, along with consistent decreases in absolute errors and improvements in temporal consistency. The corrected ERA5-Land estimates further provided a better representation of intra-annual and interannual variations, and long-term change trends.

In general, this study presents a pragmatic statistical bias correction procedure to enhance the applicability of ERA5-Land estimates in regions with limited observational networks. The procedure is expected to provide more reliable meteorological information for clean energy resource assessment and long-term planning, and to serve as a methodological reference for similar applications in other data-scarce regions. However, considering the limited ground gauge network and the lack of fine temporal resolution records, it remains challenging to address residual uncertainties in complex terrain by integrating additional observations and developing refined correction approaches. The effectiveness of the proposed correction procedure at the hourly scale also requires further quantitative evaluation when adequate observations become available.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

L.Z. wrote the main manuscript text and conceptualization of the research. L.Z. and Z.Y. conducted the main formal analysis. K.H. and W.Z. conducted data curation. All authors reviewed the manuscript.

Data availability

Data cannot be shared openly, but is available on request from the authors. Please contact the corresponding author at luzhang11@outlook.com.

Declarations

Competing interests

The authors declare no competing interests.