The precision of ERA5 reanalysis datasets and their applicability in the mountainous regions of central China are essential for weather forecasting and climate change research in the transitional zone between northern and southern China. This study employs three months of continuous measurements collected from a high-precision remote sensing platform located in a representative mountainous valley (Xinyang city) in central China, spanning December 2024 to February 2025. Our findings indicate that both horizontal and vertical wind speeds from the ERA5 dataset exhibit diminishing deviations as altitude increases. Significant biases are observed below 500 m, with horizontal mean wind speed deviations ranging from −4 to −3 m/s and vertical mean wind speed deviations falling between 0.1 and 0.2 m/s. Conversely, minimal biases are noted near the top of the boundary layer. Both ERA5 and observations reveal a dominance of northeasterly and southwesterly winds at near-surface levels, which aligns with the valley orientation. This underscores the substantial impact of heterogeneous mountainous terrain on the low-level dynamic field. At an altitude of 1000 m, both datasets present similar frequency patterns, with peak frequencies of approximately 15%; however, notable discrepancies in peak wind directions are evident (north–northeast for observations and north–northwest for ERA5). In contrast to dynamic variables, ERA5 temperature deviations are centered around 0 K within the lower layers (0–500 m) but show a slight increase, varying from around 0 K to 6.8 K, indicating an upward trend in deviation with altitude. Similarly, relative humidity (RH) demonstrates an increasing bias with altitude, although its representation of moisture variability remains insufficient. During a typical cold event, substantial deviations in multiple ERA5 variables highlight the needs for further improvements. The integration of machine learning techniques and mathematical correction algorithms is strongly recommended as a means to enhance the accuracy of ERA5 data under such extreme conditions. These findings contribute to a deeper understanding of the use of ERA5 datasets in the mountainous areas of central China and offer reliable scientific references for weather forecasting and climate modelings in these areas.

The performance of monthly prediction of extreme precipitation from the BCC-CPSv3-S2Sv2 model over the Yellow River Basin (YRB) using historical hindcast data from 2008 to 2022 was evaluated in this study, mainly from three aspects: overall performance in predicting daily precipitation rates, systematic biases, and monthly prediction of extreme precipitation metrics. The results showed that the BCC-CPSv3-S2Sv2 model demonstrates approximately 10-day predictive skill for summer daily precipitation over the YRB. Relatively higher skill regions concentrate in the central basin, while skill degradation proves more pronounced in downstream areas compared to the upper basin. After correcting model systematic biases, prediction skills for total precipitation-related metrics significantly surpass those of extreme precipitation indices, and metrics related to precipitation amounts demonstrate relatively higher skill compared to those associated with precipitation days. Total precipitation (TP) and rainy days (RD) exhibit comparable skills in June and July, with August showing weaker performance. Nevertheless, basin-wide predictions within 10-day lead times remain practically valuable for most regions. Prediction skills for extreme precipitation amounts and extreme precipitation days share similar spatiotemporal patterns, with high-skill regions shifting progressively south-to-north from June to August. Significant skills for June–July are constrained within 10-day leads, while August skills rarely exceed 1 week. Further analysis reveals that the predictive capability of the model predominantly originates from normal or below-normal precipitation years, whereas the accurate forecasting of extremely wet years remains a critical challenge, which highlights limitations in capturing mechanisms governing exceptional precipitation events.

Accurate recognition of micrometeorological conditions along overhead transmission lines is essential for the safe operation of power grids. However, meteorological cues such as snowfall, haze, and rainfall are often distributed across an image at different spatial scales and may exhibit similar visual appearances. To address these challenges, this study proposes MMRNet, a ConvNeXt-based multimodal model for meteorological condition recognition. First, a Scale-Aware Channel Mixing (SACM) module is designed to model spatially dispersed features. Second, a Cross-Scale Transformer Fusion (CSTF) module is introduced to improve semantic consistency across multiple scales. Finally, a Numeric-Integrated Visual Enhancement (NIVE) module fuses time-series meteorological measurements with visual features. A dataset containing 3950 transmission-line images and their corresponding sequences of five hourly meteorological observations was constructed for evaluation. Experimental results show that each module improves recognition performance and that their combined use achieves the highest accuracy. The complete model achieves 96.33% accuracy with 21.55 million parameters and 2.92 GFLOPs, balancing recognition performance and computational cost. Independent field evaluation achieved 96.0% recall for low-quality images, with recognition metrics decreasing by 0.5–0.8 percentage points on images meeting the quality requirements relative to the original validation results.