Snow cover is an important component of the Qinghai-Xizang Plateau （QXP） cryosphere， exerting a significant impact on regional and downstream weather and climate through radiative and hydrological processes.However， the simulation of snow cover on the QXP remains systematically biased.Previous studies have focused more on the impact of snow cover parameterization schemes， while giving less attention to the role of meteorological forcing data.In this study， we compare the China meteorological forcing dataset （CMFD） and the high-resolution near-surface meteorological forcing dataset for the Third Pole region （TPMFD）， and simulate the snow cover variations on the QXP during snow seasons in the hydrological years of 2010 -2015 by driving the land surface model SSiB3-SNICAR.The results show that TPMFD substantially improves snow simulation accuracy.The root mean square error （RMSE） of the simulated snow depth is reduced from 0.64 cm to 0.21 cm in winter， and the RMSE of the simulated snow cover fraction is reduced from 6.20% to 4.15%.For snow phenology， the TPMFD experiment simulates a longer snow cover duration in both thick and thin snow regions.TPMFD also enhances the simulation performance of surface energy balance in winter， and the RMSE of simulated surface albedo and land surface temperature is reduced to 0.15 and 3.13 ℃.TPMFD markedly improves simulation accuracy over the western and southeastern QXP by refining precipitation and air temperature inputs， notably through increased winter precipitation below 0 ℃.This enhancement reduces both the negative bias in the east and the positive bias in the western TP.Although the complex topography and snow parameterization schemes contribute to simulation uncertainties， this study still provides an effective reference for the improvement of snow cover simulation on the QXP.

A comprehensive analysis was conducted to examine the processes and impacts of water level variations in Qinghai Lake under changing climatic conditions.The study utilized monthly mean water level data from 1959 to 2017， sourced from the Qinghai Lake Basin， in conjunction with meteorological and climate variables derived from the ERA5 reanalysis dataset developed by the European Centre for Medium-Range Weather Forecasts （ECMWF）.Several large-scale atmospheric circulation indices were also incorporated to investigate their influence on lake dynamics.This integrated dataset enabled a systematic assessment of the dominant climatic drivers and facilitated the development of predictive models to simulate future water level changes.To identify the most relevant influencing factors， the Random Forest （RF） algorithm was employed to perform feature selection and importance ranking.This process allowed for an evaluation of the relationship between feature relevance and model performance.Subsequently， a comparative analysis was undertaken using five machine learning models： RF， Support Vector Machine （SVM）， Multi-layer Perceptron （MLP）， Long Short-Term Memory （LSTM） networks， and Multiple Linear Regression （MLR）.The models were trained and validated to simulate monthly water level fluctuations and to assess the influence of model complexity and temporal learning ability on predictive accuracy.The analysis revealed that key drivers of Qinghai Lake water levels include the North Atlantic Oscillation （NAO）， Atlantic Multidecadal Oscillation （AMO）， Niño 3.4 index， relative humidity at 400 hPa， 450 hPa， and 100 hPa （RH400， RH450， RH100）， precipitation， temperature at 1000 hPa （T1000）， vertical wind velocity at 1000 hPa （W1000）， and longwave radiation （LW）.Among the models tested， the LSTM network exhibited superior performance due to its ability to capture complex nonlinear and sequential dependencies in the data.When the ten most significant features were selected， the LSTM model achieved a Pearson correlation coefficient （R） of 0.95， Nash-Sutcliffe Efficiency （NSE） of 0.96， Normalized Root Mean Square Error （NRMSE） of 0.14， and Kling-Gupta Efficiency （KGE） of 0.87.The MLP model demonstrated the second-best performance， while RF and SVM yielded comparable but slightly lower results.MLR performed the worst， reflecting its limitations in modeling nonlinear and temporal relationships.Projections based on the LSTM model indicate that the water level of Qinghai Lake is likely to rise by approximately 2.55 m between 2017 and 2030.This anticipated increase reflects the continuing influence of climate change and underscores the importance of adaptive water resource management strategies in plateau lake regions.The findings offer a reliable methodological framework for modeling and forecasting hydrological changes in alpine lake systems under future climate scenarios.

As a globally significant ecological security barrier and climate-sensitive region， the Qinghai-Xizang Plateau （QXP） plays a crucial role in regional and global carbon cycles. In recent years， climate warming and ecosystem changes have led to a complex pattern of concurrent strengthening and weakening in the plateau’s carbon sink capacity. To accurately assess the spatiotemporal patterns and mechanisms of carbon dioxide （CO₂） fluxes over QXP terrestrial ecosystems， researchers have established a flux observation network centered on eddy covariance techniques since the 1980s. This network now covers various surface types including grasslands， shrublands， wetlands， and lakes. The integration of satellite remote sensing and machine learning has further facilitated the upscaling from point-based measurements to regional estimates. Studies indicate that alpine meadows serve as stable and strong carbon sinks， with mean annual net ecosystem exchange （NEE） generally below -100 g·C·m⁻²·y⁻¹. In contrast， alpine steppes exhibit weaker sink strength， largely regulated by precipitation. Wetland and marsh ecosystems show high spatiotemporal heterogeneity in CO₂ fluxes， with hydrological dynamics and seasonal freeze-thaw processes driving their shifts between carbon sink and source states. Lakes on the plateau also act as carbon sinks， though the magnitude of CO₂ exchange varies significantly across lakes， reflecting integrated effects of climate， hydrology， and riparian vegetation. At the regional scale， grassland carbon sink strength is notably higher in the eastern and northeastern parts of the plateau compared to the arid west， indicating control by climatic and ecological factors such as temperature， precipitation， and vegetation productivity. High spatiotemporal resolution remote sensing products reveal an overall strengthening trend in the grassland carbon sink over the past several decades， primarily attributed to extended growing seasons and increased vegetation cover under a warming and wetting climate. However， several uncertainties remain， including sparse distribution of flux stations in western and high-altitude regions， large parameter discrepancies among flux models， inadequate representation of carbon emissions in permafrost zones， and a lack of systematic assessment of land use change impacts on CO₂ fluxes. Moreover， interannual variability in ecosystem productivity and soil respiration introduces additional complexity in quantifying the long-term carbon balance of the plateau. Future studies urgently need to integrate multi-source observation technologies-such unmanned aerial vehicles， satellite remote sensing， and subsurface monitoring-to enhance long-term monitoring of land-atmosphere feedbacks and ecological processes， and develop mechanism-based models that couple human activities with ecological responses. These advances will provide a scientific foundation for improving the accuracy of carbon sink estimates on the plateau， understanding its dynamics， and supporting China’s “Dual Carbon” strategy.

The Qinghai-Xizang Plateau （QXP）， recognized as a global climate hotspot and sensitive indicator， profoundly influences regional and global climate and water cycles through complex multi-sphere interactions.This study presents comprehensive review of the current status and data resources of the Qinghai-Xizang’s multi-sphere observation network.Synthesizing multi-source observational data， it comprehensively reviews the key characteristics and mechanisms of the climate “warming and wetting” trend and its multi-sphere responses （atmosphere， cryosphere， hydrosphere， ecosystems）.Key findings reveal an accelerated warming rate， increased precipitation with distinct geographic variations， and subsequent chain reactions including thawing permafrost， melting glaciers， expanding lakes， enhanced vegetation growth， and increased extreme events.However， critical gaps persist in the current observation system， notably sparse coverage in the western Qinghai-Xizang， insufficient multi-sphere synergy， and imperfect data-sharing mechanisms.To address these challenges， future priorities should include expanding western Qinghai-Xizang's monitoring networks， promote low-cost automated instrumentation， enhance multi-source data fusion and model assimilation， and establish secure and standardized data-sharing platforms.This work advances our understanding of the Qinghai-Xizang's climate complexity and provides actionable insights for optimizing climate-environment monitoring systems.

Uneven intra-annual precipitation distribution in arid regions can intensify both drought and flood events， thereby inhibiting ecosystem functioning and water resource management.Under global warming， total precipitation in arid Northwest China （NWC） has significantly increased since the 1980s.Yet， it is insufficient to understand the spatio-temporal variations in its intra-annual distribution patterns.Based on the CN05.1 gridded precipitation dataset， we comprehensively examined the spatio-temporal variations in the temporal homogeneity of precipitation across multiple time scales from 1961 to 2021.As revealed by the Precipitation Concentration Index （PCI） at monthly scale， NWC exhibits pronounced seasonal precipitation patterns， with an average PCI of 17.8.From 1961 to 2021， PCI show a decreasing trend over 89.1% areas of NWC with an overall tendency of -0.29 ·（10a）^-1， especially in the hyper-arid sub-region ［-0.44·（10a）^-1］.The reduction in PCI is primarily attributed to the decreasing proportion of summer precipitation ［-0.6%·（10a）^-1］ and the increasing proportion of winter precipitation ［0.35%·（10a）^-1］， indicating a weakening seasonal concentration of intra-annual precipitation.At daily scale， the long-term changes in the Precipitation Concentration Degree （PCD） and Precipitation Concentration Period （PCP） agree with the monthly-scale PCI.PCD presents a significant decreasing trend ［-0.008·（10a）^-1］， and PCP generally varies insignificant with slight decreasing in some regions ［maximum -1.54 d·（10a）^-1］.On average， 50% of annual precipitation in NWC is contributed by the 11 wettest days.The wettest days contributing to half of annual precipitation decrease at a rate of -1.8 d·（10a）^-1 in humid sub-region but increase at 0.9 d·（10a）^-1 per decade in the hyper-arid sub-region.Our findings provide critical scientific insights in climate adaptation， risk mitigation and strategy making for drought and flood extremes in arid NWC.

As China’s “Central Water Tower” and vital ecological barrier， the Qinling Mountains’ temperature variability plays an important role in regional water conservation， ecosystem stability， and regional climate regulation.To evaluate the performance of statistically downscaled and bias-corrected Global Climate Models （GCMs） dataset （NEX-GDDP-CMIP6） in simulating observed temperature changes and further to project the future temperature variability over the Qinling Mountains， this study analyzes 8 NEX-GDDP-CMIP6 models against the CN05.1 observational dataset.The assessment focuses on the models’ ability to replicate observed annual mean temperature patterns， spatial trends， and temporal variability from 1961 to 2014.Furthermore， future temperature changes under the four Shared Socioeconomic Pathway （SSP） scenarios are projected for the period 2015 -2100.The results demonstrate that 8 models effectively capture the observed spatial pattern， warming trends distribution and interannual variability， with corresponding correlation coefficients of 0.90~0.92， 0.51~0.77， and 0.46~0.57 for 1961 -2014， respectively.The multi-model ensemble mean （MME） outperforms individual models， with correlation coefficients of 0.92， 0.65 and 0.74 for the three metrics.The MME indicates a persistent warming trend over the Qinling Mountains， with the stronger warming under the higher SSP scenarios.The warming trends are projected increase at 0.10 ℃·（10a）^-1 （SSP1-2.6）， 0.26 ℃·（10a）^-1 （SSP2-4.5）， 0.42 ℃·（10a）^-1 （SSP3-7.0）， and 0.57 ℃·（10a）^-1 （SSP5-8.5） for 2015 -2100.Notably， the warming exhibit altitudinal， zonal， and meridional dependencies， intensifying with higher elevation， latitude， and longitude.Relative to the reference period （1995 -2014）， the annual mean temperature is projected to increase by 0.65~0.97 ℃ in the near-term （2021 -2040）， 1.37~2.0 ℃ in the mid-term （2041 -2060）， and 1.39~4.46 ℃ by the end-century （2081 -2100） under the four SSP scenarios.The temperature changes are temporally consistent across the North and South Slopes over the Qinling Mountains and following with the entire regional average.However， the North slope warms more rapidly than the South slope， particularly under high-emission scenarios （e.g.， SSP5-8.5）， where North slope warming accelerates markedly.These findings provide critical insights for climate adaptation and ecological management in the Qinling Mountains.

The weather and climate of the Qinghai-Xizang Plateau （QXP） not only exert profound influences on Asian regional climate but also regulate broader Northern Hemisphere climate patterns.Various global atmospheric reanalysis datasets developed abroad have been extensively utilized to characterize the QXP's climate features.In 2020， the China Meteorological Administration released its first-generation global atmospheric reanalysis product （CRA）， which demonstrated superior performance in regions with dense conventional observations， yet its skill over the QXP remains unclear.In this study， we performed an independent evaluation of CRA's upper-air temperature， wind fields， and relative humidity using 686 high-quality， non-assimilated radiosonde profiles from five stations collected during June-August 2014 in the 3rd Qinghai-Xizang Plateau Atmospheric Scientific Experiment.A non-independent assessment was also performed using routine observations from 21 sounding stations across the QXP and adjacent areas.Results were compared with those from ERA-Interim， ERA5， and JRA-55 reanalysis datasets.Non-independent validation results demonstrate that in the troposphere and lower stratosphere， correlation coefficients of temperature and wind speed between CRA and operational radiosonde observations exceed 0.9.Relative to operational radiosondes， CRA temperature bias of 400 hPa and 500 hPa approaching 0 ℃.The near-surface zonal wind RMSE of CRA is about 2.5 m·s^-1， decreasing gradually with altitude to 1.5 m·s^-1 at 100 hPa.Relative humidity RMSE remains below 20% across all altitude layers.Independent validation results indicate that errors in air temperature and wind speed over the eastern Qinghai-Xizang Plateau are generally smaller than those over the western plateau， whereas relative humidity errors are larger in the east.From 600 hPa to 30 hPa， the mean root-mean-square errors （RMSE） of CRA relative to radiosonde for temperature， zonal wind， and meridional wind are 1.38 ℃， 3.19 m·s^-1， and 3.22 m·s^-1 in the western Qinghai-Xizang Plateau， respectively； and 1.16 ℃， 2.65 m·s^-1， and 2.90 m·s^-1 in the eastern Qinghai-Xizang Plateau， respectively.Errors in the western QXP were slightly larger than in the east and exhibited pronounced diurnal variability.Maximum temperature and relative humidity errors below 600 hPa occur in the afternoon.At 500 hPa， peak relative humidity errors appear in the evening.CRA objectively reproduces the QXP’s vertical structures of temperature， wind speed， and relative humidity.Compared to other reanalysis products， CRA’s relative humidity estimates were closest to radiosonde observations， and wind field errors were slightly higher than those of other datasets but did not exceed 0.4 m·s^-1 on average.

The Ya'an area has a typical "windward slope" and "trumpet" topography， the observation and study of cloud-precipitation in the Ya'an area has always been one of the hot issues of mountain meteorology in China.In this paper， using the observation data of Ka-band millimeter-wave cloud radar， K-band micro-rain radar and DSG4 laser raindrop spectrometer during the autumn rainy season in Ya'an area in 2023， the two types of cloud-precipitation in Ya'an during the autumn rainy period are studied and compared.The results show that （1） the frequency of warm cloud-precipitation （WCP） is higher than that of cold cloud-precipitation （CCP） during the autumn rainy period in Ya'an， but the intensity of precipitation is generally weaker and the cumulative rainfall is less.（2） In terms of vertical structure and macroscopic features， there are obvious differences between WCP and CCP： the radar echo reflectivity factor Z high-frequency region of CCP has a wider distribution and stronger echo， and the magnitude of the slope of Z high-frequency region at different heights reflects the difference between the ice-phase particles' growth mechanism and rate and that of WCP's liquid cloud raindrops' touch-and-go growth rate； CCP's linear depolarization ratio， LDR， increases abruptly near the zero-degree layer due to the melting of ice-phase particles.The mean Doppler velocity V_m has a smaller value in the high-frequency region and the velocity spectral width W has a larger value in the high-frequency region， suggesting a larger mean droplet scale and a wider concentration distribution.In terms of macroscopic parameters， the cloud base and cloud top of the CCP are significantly higher， the cloud layer is thicker， and the distribution of values is more dispersed.（3） In terms of microscopic characteristics， the raindrop spectra of the two types of cloud-precipitation are different with altitude： the concentration of small raindrops in the CCP increases and then decreases with decreasing altitude due to the melting of ice-phase particles and raindrops merging， whereas that in the WCP fluctuates due to the effects of evaporation， merging， water vapor transport， and updraft lifting altitude.The medium- and large-sized raindrop concentrations of both types of cloud-precipitation increase with decreasing altitude on the whole.In terms of the differences in raindrop concentrations at different heights， above 455 m， the concentration of raindrops of all grain sizes in WCP is almost lower than that in CCP； below 455 m， the concentration of small and large raindrops in WCP is higher than that in CCP， and the concentration of medium raindrops is lower.（4） Comparison of the raindrop spectra under different precipitation intensities showed that when the precipitation is weak， the WCP had lower concentrations of raindrops of all particle sizes than the CCP， but when the precipitation reached a certain intensity （rainfall intensity R \ge5 mm·h^-1）， the WCP produced more small raindrops and a small number of larger raindrops.The few extreme values of Z， LDR， V_m and W of WCP are larger than those of CCP at altitudes below 3 km due to the synergistic effect of low-altitude rapids and topographic uplift in the region， which makes the small WCP raindrops repeatedly touch and form medium-to-large raindrops in the re-uplift airflow.

Hongjiannao Lake is the largest desert freshwater lake in China.In recent decades， the area of the lake has sharply decreased.The evaporation of the lake surface is the main factor consuming its water volume.Therefore， this paper aims to reveal the characteristics of evaporation changes and the mechanism of the driving factors.Currently， most studies on Hongjiannao Lake directly use or convert the evaporation data observed at meteorological stations， which have many missing and discontinuous data， and do not qualitatively and quantitatively analyze the meteorological factors influencing the evaporation changes of Hongjiannao Lake.To address these issues， this paper uses the data converted from meteorological stations， calculates the evaporation using the FAO （P-M） formula， and simulates the evaporation using the CLM-LISSS model to obtain the evaporation data of Hongjiannao Lake.Through comparison with the converted evaporation data from meteorological stations， it is found that the evaporation values and correlations simulated by the CLM-LISSS model are closer to the actual situation than the results calculated by the FAO （P-M） formula.The evaporation simulation results based on the preferred model showed that the average annual value of simulated evaporation of Hongjiannao lake from 1980 to 2018 was 1004.56 mm， and the M-K mutation test did not find the mutation year， and the overall trend was significantly upward （3.01 mm·a^-1）.The meteorological factors that have significant positive correlation with evaporation are air temperature， wind speed and downward long-wave radiation， and their correlation with evaporation and their own change trend pass the significance test of 95%.The sensitivity coefficient of evaporation to meteorological factors and the contribution of each meteorological factor to evaporation change were quantitatively analyzed by the formula calculation method and the perturbation analysis method of climate state respectively.The meteorological factors with greater contribution obtained by the two methods were significantly consistent with the correlation， and they were all air temperature， wind speed and downward long-wave radiation.However， the contribution ranking obtained by these two methods is slightly different and the contribution values of each factor are significantly different.This is mainly due to the fact that the change of evaporation is only caused by the change of a single factor， which reduces the influence of other driving factors， and effectively reduces the error between the change value of evaporation trend and the contribution sum of meteorological factors， from 128.40 mm （109.40%） to 56.83 mm （48.42%）.The perturbation analysis of climate state is superior to the formula calculation method in both mechanism and error.The results show that the contribution of meteorological factors to evaporation changes from large to small are downward long-wave radiation （71.47%）， temperature （59.83%）， wind speed （41.00%）， air pressure （1.54%）， downward short-wave radiation （-3.00%） and specific humidity （-22.43%）.

As a powerful deep-learning downscaling technique， convolutional neural networks （CNN） are widely utilized to generate high-resolution precipitation data from low-resolution global climate models through data-driven approaches， playing a critical role in assessing the impacts of climate change at both regional and local scales.This study proposes a CNN-based precipitation downscaling model， the Topography-Guided Attention Network （TGAN）， which downscales coarse-resolution （2°） atmospheric variables to produce high-resolution precipitation fields at 0.1°.The model adopts a Laplacian pyramid as a progressive， multi-level downscaling framework， in which the spatial resolution of precipitation fields is incrementally enhanced and precipitation structures are reconstructed through successive stages.In addition， a topography-guided attention module is incorporated， which leverages an attention mechanism to integrate atmospheric variables with elevation data at corresponding spatial scales.By combining these multi-scale features， the module strengthens the network’s capacity for feature representation and learning， thereby improving the accuracy and reliability of simulated precipitation.Focusing on the middle reaches of the Yellow River， TGAN is trained on daily ERA5 atmospheric variables， GPM IMERG daily precipitation data from 2001 to 2010， together with static elevation data， and validated using data from 2011 to 2020.The results indicate that， compared with a conventional CNN model， TGAN achieves higher accuracy in spatiotemporal precipitation simulations at daily， monthly， and annual scales.At the daily scale， TGAN achieves a lower average root mean square error （5.10 mm·d^-1） and a higher average correlation coefficient （0.42） compared with the conventional CNN model.Additionally， TGAN more accurately captures extreme precipitation events （95th and 99th percentiles） and better aligns with GPM IMERG observations in probability density distribution， particularly for heavy precipitation range.This study further investigates the impact of different loss functions on the downscaling performance of TGAN.Using the RMSE loss function improves overall predictive accuracy but leads to underestimation of extreme precipitation events， whereas the Bernoulli-Gamma loss function， although slightly less accurate overall， more faithfully reproduces extreme precipitation events.Its probability density distributions are highly consistent with both GPM IMERG data and station observations， indicating that the model has an enhanced capability to capture extreme precipitation events， thereby better reproducing the distribution characteristics of precipitation frequency.Overall， by combining the topography-guided attention mechanism with the Bernoulli-Gamma loss function， TGAN demonstrates clear advantages in downscaling precipitation over the middle reaches of the Yellow River， not only improving overall simulation accuracy but also better representing extreme precipitation events， providing a robust and reliable tool for high-resolution precipitation modeling in complex terrain regions.

In order to explore the response of snow cover to climate conditions under the complex topography， based on composite snow cover products on eight days of MOD10A2， daily temperature and precipitation from September 2001 to August 2020 and digital elevation model （DEM） in the Yellow-River source region， the snow cover frequency and percentage was used as an index to study the spatio-temporal variation characteristic of snow cover and their relations with major meteorological elements.The results are as follows： （1） Snow was mainly appeared from October to next May in the Yellow-River resource region， and the higher altitude， the greater snow cover percentage and the longer duration.Snow mainly distributed in high mountainous areas.The yearly accumulative process of snow showed a U-shaped at the altitude of 3500 m and below， the peak appeared in winter， while that transited to inverted W-shaped above the altitude of 3500 m， and the gap between two peaks gradually reduced with the rising of altitude， the appearing time of the maximum peak shifted from spring to autumn.（2） The snow cover percentages on the northern and northerly slopes were more than that on the southern and southerly slopes during the accumulative period， among the maximum was on the northeastern slope， and the minimum was on the southern slope， but the yearly variation characteristics with inverted W-shaped were similar and synchronous on each slope， the maximum and secondary peaks appeared at the end of October and the early of April， respectively.（3） Under the background of warming and humidification， the snow cover percentage during the accumulative period exhibited an overall increasing trend with a rate of about 5.6% per 10-year in the Yellow-River resource since 2001， especially at heights above 4000 m.The significant increase of snow cover area in spring and winter at heights 4000~5000 m contributed greatly to the increase of snow cover percentage during the accumulative period.（4） The average snow cover percentage first increased and then decreased with the decline of temperature and the increase of precipitation during the cumulative period in each altitude zone of the Yellow-River source， and the higher the altitude， the higher the fitting goodness between the two.The consistent increase of precipitation was the key factor contributing to the increase of snow cover percentage in each season of accumulative period since 2001， especially the significant increase of precipitation at high altitude areas in spring.Snow was more dependent on precipitation at high altitude areas.

Snow cover is an important component of the cryosphere.In recent years， climate warming has led to a reduction in snow cover area.This change may cause uneven distribution of water resources and a decline in biodiversity， thereby affecting local life， economic development and ecological environment.Qinghai Lake is the largest inland lake in China.In recent years， its water level has changed rapidly.The runoff into the lake is affected by the snow cover and its changes in the basin.However， the characteristics， changes and causes of the influence of the snow cover in the Qinghai Lake Basin are still unclear.Based on the temperature and precipitation data from the Moderate-resolution Imaging Spectroradiometer （MODIS） daily cloudless 500 m snow area product Dataset and the China Meteorological Forcing Dataset （CMFD）， This paper studies the spatial and temporal variation characteristics and influencing causes of snow cover in the Qinghai Lake Basin.The results show that： （1） There is a good correspondence between the distribution of the average annual snow cover frequency from 2000 to 2020 and the altitude.As the altitude decreases， the snow cover frequency also decreases accordingly.It is simultaneously affected by the average annual temperature and the annual precipitation.Among them， the areas significantly affected by the average annual temperature in a partial correlation are mainly distributed in the northern and eastern parts of Qinghai Lake， and the areas significantly affected by the annual precipitation in a partial correlation are mainly distributed in the middle and upper reaches of the Buha River in the middle of the Qinghai Lake Basin.（2） From 2001 to 2017， precipitation increased in the Qinghai Lake Basin and the Qilian Mountains region.However， due to the increase in the average annual temperature and the decrease in the annual snowfall in the two regions， the snow cover area decreased.（3） The intra-annual variations of snow cover area in the Qinghai Lake Basin and the Qilian Mountains region are relatively similar， both showing a double-peak fluctuation feature.However， the reduction in snow cover area from April to July and the increase from August to January of the following year in the Qinghai Lake Basin are both greater than those in the Qilian Mountains region.From January to March， the snow cover in the Qilian Mountains region decreased， while that in the Qinghai Lake Basin increased， which corresponds well to the lake effect after the melting of Qinghai Lake.

On March 27， 2024， a disastrous gust damaged the terminal building at Kunming Changshui International Airport in Yunnan， China.The role of coupled atmospheric gravity waves in this event is examined using conventional surface meteorological observations， high-resolution mesoscale surface network， upper-air sounding data， ERA5 reanalysis data and FY-2G geostationary satellite imagery.The results show that： （1） After solar heating in the morning， an atmospheric gravity wave originating over the Hengduan Mountains propagated southeastward under the influence of westerly flow.Meanwhile， another gravity wave generated over the Ailao Mountains propagated northeastward.These two gravity waves met near 100.56°E， 25°N about 09:53（Beijing Time）， producing a significant coupling effect.（2） Based on the force analysis within a statically stable atmospheric layer and the initial perturbations induced by orographic lifting of the background flow， the analytical mathematics formula of the gravity wave was derived.Using satellite observations and infrared cloud imagery， the key parameters of each gravity waves including significant amplitude， wavelength， intrinsic frequency， and phase speed-were quantitatively obtained.（3） By harmonic analysis， the coupled wave with its zonal and meridional components were determined.The coupled wave induced substantial vertical motions above the airport： the zonal component generated a downward motion of 18 m·s^-1， contributing positively to the surface winds， while the meridional component produced a weaker upward motion of 1 m·s^-1， exerting a negative contribution.（4） The comprehensive effect of the coupled gravity waves and the DCAPE releasing generated a strong subsiding airflow， which enhanced the downward transport of upper-level jet momentum.This momentum transition was combined with the near-surface horizontal wind field， ultimately resulting in a catastrophic surface gust reaching 86.5 m·s^-1， causing severe structural damage to the main terminal building of Kunming Airport.Quantitative assessment indicates that gravity waves contributed approximately 19.65% to the total wind speed.In contrast， the presence of either gravity wave A or B alone without coupling effect would account for a maximum wind speed contribution of only about 10 m·s^-1.The role of gravity wave coupling is of critical in intensifying subsidence and promoting the extreme low-level wind hazards.

The Yarlung Zangbo Grand Canyon region （YZGC） is an important water vapor transport channel on the Qinghai-Xizang Plateau， and the water vapor transport of the YZGC has an important impact on the weather and climate of the plateau.Based on the ERA5 reanalysis data and CN05.1 precipitation data from 1986-2021， this paper analyzes the water vapor transport characteristics of the YZGC during the summer monsoon and winter monsoon periods and studies its relationship with precipitation in Three-River Source Region （TRSR） during the plateau summer monsoon period.The results showed that： （1） The precipitation in the YZGC was more in the southeast and less in the northwest and showed an increasing trend in time， and the increase in the summer monsoon period was greater than that of the annual average and the winter monsoon period.The water vapor flux of the YZGC mainly comes from the southwest water vapor transport from the Indian Ocean to the Bay of Bengal， followed by the westerly water vapor transport from the southern branch of the mid-latitudes.（2） The south and west boundaries of the YZGC are the water vapor input boundaries throughout the year， and the east and north boundaries are the water vapor output boundaries， and the water vapor transport flux during the summer monsoon is stronger than that during the winter monsoon period.The YZGC is a positive area of net water vapor flux， and the net water vapor income shows a significant decreasing trend.（3） The low-level 1000~600 hPa originates from the Indian Ocean to the Bay of Bengal and the western Pacific Ocean， and the water vapor transport enters the plateau from the southern boundary of the YZGC， and the southern boundary is the main boundary of water vapor input.The upper layer is 600~300 hPa， and the middle and high latitudes are significant westerly water vapor transport， and the water vapor is mainly imported from the western boundary and exported from the eastern boundary.（4） There is a significant positive correlation between the water vapor flux of the YZGC and the precipitation in the central and eastern parts of the TRSR during the summer monsoon.Anomalous southeast water vapor transport from the Bay of Bengal， as well as anomalous southwest water vapor transport from the northern Bay of Bengal and the southwestern side of the plateau， enters through the western and southern boundaries of the YZGC， providing ample moisture to the central and eastern parts of the TRSR.The upward motion of this water vapor in the central and eastern TRSR leads to increased precipitation.

A persistent heavy rainfall event occurred in the Sichuan Basin from July 11 to 13， 2023， with the maximum cumulative precipitation reaching 656.5 mm， matching the historical record， while the second-highest cumulative precipitation reached 650.4 mm， breaking the station's observational record.The torrential rain triggered secondary disasters such as flash floods， landslides， and debris flows， resulting in significant losses.This study utilizes dense surface automatic precipitation observation data， Wenjiang sounding data， ERA5 reanalysis data， and FY-4A satellite Brightness Blackbody Temperature （TBB） data to analyze the atmospheric circulation background and physical mechanisms of this persistent heavy rainfall event.The results indicate that the South Asian High dominated over Sichuan， while the Western Pacific Subtropical High （WPSH） remained stable on the periphery of Sichuan and upper-level trough moved eastward.Continuous moisture transported from the South China Sea and the Bay of Bengal into Sichuan Basin， which provided ample water vapor for the heavy rainfall.Significant low-level warming and moistening in the western Sichuan Basin enhanced atmospheric instability.The large-scale circulation background created favorable conditions for the occurrence of the persistent heavy rainfall.On the evening of July 10， airflow inertial oscillation generated a low-level jet over the eastern Yunnan-Guizhou Plateau， which enhanced wind speed and moisture advection into the basin.The southeasterly flow， obstructed by the western mountains， induced frontogenesis and moisture convergence ahead of the foothills， leading to convective instability.The airflow ascended along the mountain slopes， forming a vertical circulation cell over the basin.Orographic lifting and this vertical circulation jointly triggered convection.The persistent southeasterly flow， mechanically blocked by the terrain， generated northerly winds in front of mountain， which converged with the southeasterly winds and formed a shear line， promoting organized convective development.The trumpet-shaped terrain and canyon topography enhanced precipitation.From July 11 to 12， the intensification of WPSH blocked the eastward movement of the southern branch flow and the trough from the plateau.This process， superimposed with the jet stream transport from the eastern Yunnan-Guizhou Plateau， resulted in enhanced convergence over the southern basin， triggering the development of Southwest Vortex.Latent heat release provided thermal condition for the development of the Southwest Vortex， while divergence and vertical transport supplied the dynamic conditions.Guided by the WPSH’s peripheral flow and influenced by latent heating， the southwest vortex moved westward， triggering vigorous convective activity.In the early hours of July 13， an eastward-moving upper-level trough guided cold air to move southward along the western basin.The convergence of cold and warm air produced intense frontogenesis， triggering precipitation in the pre-frontal warm sector.The intensification of cold air enhanced the development of the cold front and prolonged the lifespan of the Southwest Vortex， resulting in increased rainfall intensity and a significant eastward expansion of the heavy rainfall area.

Stratospheric airships， due to their excellent performance， exhibit great potential in flight missions within the aerospace field.However， the complexity of the stratospheric environment presents significant challenges to flight safety.Therefore， conducting environmental forecasts in advance to mitigate flight risks is crucial for the successful completion of airship missions.Although existing thermodynamic models have provided a theoretical foundation for airship design， their analysis remains insufficient.In this paper， two key parameters， cloud top height and cloud top temperature， are introduced to deeply analyze their effects on the airship's thermal balance.Cloud top height determines the relative position between the cloud layer and the airship， while cloud top temperature directly reflects the thermodynamic state of the cloud layer.Both parameters influence the radiative exchange and thermal balance of the airship.Based on this analysis， the importance and urgency of incorporating these factors into thermodynamic models are emphasized， offering new insights for the optimization of future airship thermodynamic models and revealing the significance of studying cloud layer distribution characteristics.Additionally， the unique advantages of the Qinghai-Xizang （Tibet） Plateau as a natural laboratory are highlighted， and specific data analysis is conducted.This paper analyzes historical meteorological cloud observation data from 2015 to 2020 in the Qinghai-Xizang （Tibet） Plateau region using the CLARA-A3 dataset， focusing on the spatial distribution characteristics， daily mean and extreme values， cloud area proportions， and the correlation analysis between cloud top height and cloud top temperature.The results show that cloud top height exhibits a spatial distribution pattern of lower in the northwest and higher in the southeast， while cloud top temperature shows a trend of higher in the west and lower in the east.During the period from July to September， cloud top height reaches its highest annual value， and cloud top temperature reaches its lowest.The annual variation pattern of the cloud top parameters is also confirmed.Further analysis indicates the presence of extreme meteorological phenomena， such as ultra-high clouds and extremely low temperatures， with daily cloud top heights exceeding 18 km and cloud top temperatures below -83 ℃， which mostly occur between July and September.A Spearman correlation coefficient analysis reveals a moderate negative correlation between the two variables， with a strong negative correlation during the July-September period.This provides important data support for further quantifying the influence of cloud layers on airship performance.The study shows that the potential threat of cloud layers to airships cannot be ignored， particularly during the July-September period， when the deployment preparation for stratospheric airships should focus on the changes in local meteorological cloud parameters.Acquiring cloud observation data in advance and implementing meteorological forecasting preparation are key factors in ensuring the safe flight of airships.This research is the first to apply real meteorological data analysis to airship flight environment assessment， verifying the feasibility of data analysis techniques， and emphasizing the critical role of observational data in model validation， providing a new research perspective for improving the thermodynamic models of stratospheric airships.In the future， with the continuous optimization of dynamic prediction models， the safety of airships in complex meteorological environments is expected to be significantly improved.

To solve the difficult problem of forecasting the location and time of heavy rain under the background of Central Asian low vortex.using conventional observation data， satellite， radar and ERA5 0.25°×0.25° reanalysis data， to analyze the mesoscale characteristics and atmospheric instability， and reveal the convection triggering and maintenance mechanisms from June 30 to July 3， 2024， and find that there are two rainstorm centers in this weather event， whose triggering mechanisms are significantly different.The rainstorm in Bozhou is a short-term heavy rainfall caused by strong convection.It is affected by the eastward movement of short-wave disturbances split from the bottom of the Central Asian trough， the northward movement of warm moist air carried by southwest winds in the middle layer， and the eastward infiltration of cold air guided by east winds in the lower layer.Warm moist air climbs along the cold cushion， and there is a convectively unstable layer in the upper air of Bozhou.Surface mesoscale convergence lines are the main triggering factors， and middle γ scale convective systems continue to generate and merge to enhance middle β scale convective systems to produce heavy rainfall， and the convection has a high-quality heart structure.The rainstorm of Tianshan Mountain area and its northern slope are characterized by mixed rainfall， with long-lasting precipitation.The Central Asian trough weakens and moves eastward， while the southwest airflow at the middle level guides warm and humid air northward.At the same time， the northwest jet stream at the lower level brings cold air southward， leading to the convergence of cold and warm air.Before and in the early stages of precipitation， the atmosphere is affected by convective instability， which accumulates unstable energy for the generation of precipitation.In the later stages， the atmosphere shifts to conditional symmetric instability， which maintains and enhances precipitation.The frontogenesis of cold fronts in the middle and lower layers of the troposphere is the main triggering factor of heavy rainfall.The upward movement of the secondary circulation of the front and the occurrence and development of convection are closely related， and the front disturbance and cumulus convection have a positive feedback effect.Multiple medium β scale convection systems move eastward one after another to produce heavy rainfall， and the convection has a low quality center and is warm cloud precipitation.This study reveals the triggering and maintenance mechanisms of short-term severe convective rainstorms and mixed rainstorms in northern Xinjiang under the background of the Central Asian low vortex by refining their types， providing key references for the refined forecasting of rainstorms in northern Xinjiang.

Soil moisture is the key variable of the land-atmosphere interactions and ecosystem dynamics， so obtaining the soil moisture data with high spatiotemporal resolution is significant for simulating the hydrological process and managing the resource.The study is Maqu County，which is located on the eastern edge of the Qinghai-Tibet Plateau.We choose the SMAP daily soil moisture products and hourly in situ observations from ISMN as the base data.we build a Bi-LSTM model by using the GEE technology to construct the data in temporal scale， and by utilizing the random forest regression to downscale the soil moisture from 9 km to 250 m with the multi-high-resolution data （NDVl， DEM， and LST）.The results show that Bi-LSTM model with a ratio constrained correction can estimate the hourly soil moisture effectively， have a well performance in multi-site validation with a maximum R ² of 0.8735， and the random forest model give us a distribution characterization with more refined spatial scale.In this study， we refine the remotely sensed soil moisture in space-time dimension synchronously， and overcome the limitations of traditional approaches which build model by downscaling spatial scale only.

To further improve the forecast accuracy of strong gusts and mitigate the disaster risks they pose， this study utilizes 24 h forecast 10m wind field data from the European Centre for Medium-Range Weather Forecasts （ECMWF） deterministic model spanning January 2022 to December 2023， as well as concurrent hourly observations of maximum gust speed and direction from ground stations operated by the China Meteorological Administration （CMA）. Focused on 358 national observation stations in Sichuan and its surrounding regions， this study proposes a novel gust parameterization method based on mean wind components following a systematic analysis of the limitations of the traditional gust factor method. Widely adopted in operational forecasting due to its simplicity and strong practicality， the traditional gust factor method exhibits substantial underestimation of strong gusts in Sichuan and adjacent areas. This limitation stems primarily from inherent forecast errors in numerical models， unbalanced distribution of sample data， and failure to account for gust forecast discrepancies induced by variations in mean wind components. To address this underestimation issue， this study develops a gust parameterization method based on the numerical model's 10 m u and v wind forecasts for individual stations. Specifically， observed gust wind directions are first adjusted to align with model-predicted directions， eliminating the interference of directional mismatches on gust speed forecast accuracy. For different sign combinations of the model-predicted 10 m u and v components， gust factors at designatedquantiles are calculated separately for the squared u and v components of direction-adjusted observed gusts and the squared u and v components of model-predicted winds. This approach fully incorporates the influence of differences in 10 m u and v wind components on gust forecasts while reducing biases in strong gust fitting caused by unbalanced sample distribution. Independent sample testing is conducted using data from January to November 2024， with comparative analyses of multiple gust forecast methods. The results demonstrate that three key measures-calculating gust factors based on designated quantiles， constructing a parameterization model using mean wind components， and performing wind direction alignment-can all effectively reduce the underestimation of strong gusts. In the evaluation of metrics such as forecast error and accuracy， the proposed method outperforms other approaches in strong gust forecasting. Notably， the contributions of the first two measures to mitigating strong gust underestimation are greater than those of wind direction alignment. Case validation during a cold air-induced strong wind event in March 2024 further confirms the superiority of the proposed method over the traditional gust factor method. Additionally， a comparative analysis is performed against the 3 hour maximum gust forecast product from the ECMWF deterministic model， a commonly used tool in operational forecasting. The results show that for gusts ≥13. 9 m·s⁻¹， the proposed method achieves higher forecast accuracy than the ECMWF model product. However， the mean error and mean absolute error of the proposed method in this wind speed range are slightly higher than those of the ECMWF model. This is primarily because the 10 m wind field data used in this study are model forecasts at instantaneous time points， which fail to fully capture the hourly variability of wind fields driven by transient meteorological processes. In future research， hourly maximum wind speed and corresponding direction forecast data from mesoscale numerical models will be integrated with the proposed method to further improve the forecast accuracy of strong gusts.

Located in the temperate monsoon climate zone of northeast China， dominated by coniferous and broad-leaved mixed forest the Lesser Khingan Mountains has the largest and most complete primary forest of Korean pine in Asia， which plays an important role in regulating the regional climate.In order to explore the sensible heat flux （H） and latent heat flux （LE） characteristics of the coniferous and broad-leaved mixed forest in the Lesser Khingan Mountains and the regulation effect of environmental， biological factors on H and LE， H and LE at 40 m and 50 m altitude of Wuying National Climate Observatory from 2007 to 2023 were studied by using the eddy correlation method， and a structural equation model was constructed to analyze the influencing factors in detail.The results show that： The interannual variation of H and LE in the coniferous and broad-leaved mixed forest in the Lesser Khingan Mountains showed a decreasing trend， but the variation trend was not completely consistent.The annual average of H and LE at 40 m height was 19.84±1.83 W·m^-2 and 29.39±2.93 W·m^-2， respectively.The annual average of H and LE at 50 m altitude was 22.71±1.29 W·m^-2 and 31.76±1.07 W·m^-2， respectively.The peak of H appeared in April， the secondary peak in October， and the peak of LE appeared in July， in which LE was greater than H from May to September， indicating that the energy conversion was dominated by latent heat exchange in May to September， and by sensible heat exchange in other months.The 30-min scale energy closure rate was 49%， the monthly energy closure rate ranged from 32 to 61%， in which the growing season and non-growing season were 53% and 38%， respectively， and the daily scale energy closure rate was 52%.The structural equation model showed that the heat transfer process of the coniferous and broad-leaved mixed forest in the Lesser Khingan Mountains was mainly limited by energy.Net radiation had a positive effect on H and LE， while air temperature， vapor pressure deficit， soil volumetric water content and leaf area index had opposite effects on H and LE.Climate change has a complex regulatory mechanism on H and LE.The complex interannual variation of H and LE can be partially explained by constructing structural equation models for different years.

The frequent occurrence of heavy precipitation events poses severe challenges to urban disaster prevention and mitigation. High-resolution quantitative precipitation estimation （QPE） products are of great importance for monitoring intense rainfall processes， while X-band phased array radars （X-PAR）， delivering detection data with high spatiotemporal resolution， provide a robust foundation for advanced QPE research. Due to the limitations of traditional radar-based QPE methods， such as insufficient nonlinear mapping capabilities and the lack of dynamic adaptive learning mechanisms， a minute-scale rainfall QPE method based on the CBAM U-Net model is proposed. This model enhances the learning capability for key precipitation regions and feature channels by incorporating the Channel and Spatial Attention Module （CBAM） and optimizes the temporal alignment between radar and ground rain gauge observations by introducing a temporal dimension to the model input. X-PAR data from four radars in Chengdu’s plain area during July-September 2023 -2024 were used to construct training and testing datasets. Additionally， the training process was optimized through the integration of learning rate warm-up and cosine annealing strategies， which accelerated the convergence speed and improved the model’s robustness. Evaluation metrics， including regression indices （CC， R _MSE， R _MAE， R _MB） and classification metrics （POD， FAR， CSI）， were applied to compare the performance of the CBAM U-Net against Attention U-Net， U-Net， and two Z-R methods. Results show that deep learning models significantly outperform traditional Z-R methods in both error metrics and stability. Among the three models， CBAM U-Net achieves the best performance （CC=0.665， R _MSE=0.331 mm·min^-1， R _MAE=44.651%， R _MB=7.324%）， followed by Attention U-Net， while U-Net shows significant underestimation. In multi-threshold classification evaluations， CBAM U-Net achieves CSI of 0.655 and 0.485 at thresholds of 0.34 mm·min^-1（corresponding to 20 mm·h^-1） and 0.67 mm·min⁻¹ （corresponding to 40 mm·h^-1） respectively， which are notably higher than those of the Attention U-Net （0.609， 0.408） and U-Net （0.537， 0.160）. Case studies further verify that CBAM U-Net has a stronger capability to capture temporal features. During the "pulse-type" precipitation event on 10 September 2024， CBAM U-Net accurately identifies the peaks and troughs of precipitation curves and effectively characterizes the temporal variation features of precipitation， performing better than U-Net and Attention U-Net. During the stable precipitation process on 29 September 2024， CBAM U-Net successfully captures the overall trend of precipitation changes but shows insufficient sensitivity to minor fluctuations within the range of 0.1~0.2 mm·min^-1. The CBAM module effectively improves the model's capability to capture temporal characteristics of heavy precipitation via its dual attention mechanism. However， due to the inherent systematic smoothing effect of deep learning models， notable underestimation still occurs for extreme heavy precipitation events with minute rainfall rates ≥1.17 mm·min^-1.

In order to reduce the rate of flight diversion and return caused by low visibility， this study has established a short-term and nowcasting model of low visibility using ground observation data and the upper-air and surface forecast data of the ECMWF at Jingdezhen Airport based on machine learning algorithms.Comparing the evaluation indicators， the results find that the XGBoost and LightGBM machine learning algorithms outperform the SVM machine learning algorithm in nowcasting of the airport low visibility.A detailed comparison of the evaluation metrics was conducted both before and after feature screening in the same machine learning algorithms.The study highlights that feature screening significantly boosts the effectiveness of both models.Furthermore， the SHAP （SHapley Additive exPlanations） method elucidates the contribution of each feature to the LightGBM model's output.The main conclusions are as follows： （1） The machine learning models established by LightGBM and XGBoost perform well in airport low visibility forecasting， with the AUC reaching up to 0.98， and the F1_score for the prediction of current low visibility and the low visibility in the next one hour can reach up to 0.92.（2）Data cleaning and feature screening is conducive to improving the prediction accuracy of the XGBoost algorithm for low visibility in the next hour， according to the principle that "feature engineering in machine learning requires features to be mutually independent".Moreover the LightGBM model with feature screening has a lower false negative rate than the LightGBM model without feature selection when forecasting the current and future one-hour low visibility.For the forecast of the current low visibility， the LightGBM_24_0h model is the best.For the forecast of low visibility in the next one hour， the XGBoost_24_1h model is the best.And feature selection has a greater improvement on the performance of the XGBoost algorithm.（3） The splitting times and SHAP values are used respectively to analyze the feature importance of the LightGBM algorithm model.It shows that under different feature importance criteria， nine features， namely the measured relative humidity， air temperature， wind， sea level pressure at the airport， and the relative humidity at 1000 hPa， vertical velocity and divergence at 925 hPa， and divergence at 850 hPa predicted by ECMWF， are more important for the prediction of low visibility at the airport.And divergence， as an input feature of the machine learning model， can greatly improve the performance of the machine learning model.（4） When explaining feature importance based on SHAP values， the cumulative proportion of the top ten feature importance accounts for 80%.This indicates that in the nowcasting of low visibility at Jingdezhen Airport where fog is the main factor， the LightGBM model can output prediction results according to key forecast factors，.And when forecasting whether the low visibility in the next one hour will continue， more attention should be paid to the changes in 850 hPa divergence， 1000 hPa relative humidity， airport sea level pressure and wind direction.

To evaluate the applicability of air temperature data from three reanalysis datasets （ERA5， JRA-55， and MERRA-2） in the northern part of the Eastern Kunlun Mountains and its different subregions， this study used daily mean temperature observations from 28 national meteorological stations in the northern part of the East Kunlun Mountains during 1980 -2023. The air temperature data from the reanalysis datasets were corrected using a lapse-rate correction method. Four evaluation indicators， namely the correlation coefficient （CC）， mean bias （MB）， root mean square error （RMSE）， and mean absolute error （MAE）， together with the climatic tendency rate， were used to assess the applicability of annual and seasonal air temperature from the three reanalysis datasets at regional and station scales. The results show that： （1） Bilinear interpolation outperformed nearest-neighbor interpolation. After lapse-rate correction， the mean absolute value of MB of the three reanalysis datasets decreased by 0. 92~1. 81 ℃， and the RMSE decreased by 0. 69~1. 34 ℃. （2） After lapse-rate correction， all three reanalysis datasets reproduced the spatial distribution characteristics of mean air temperature in the study area， namely higher values in the east and lower values in the west， with the lowest values in the northeast， although overestimation occurred in some areas. （3） At the regional scale， ERA5 showed the best applicability. JRA-55 ranked second， while MERRA-2 showed the poorest applicability. Only the spring climatic tendency rate of ERA5 was higher than that of the observations； the remaining values were lower than the observed values. All three datasets showed a warming pattern characterized by stronger warming in summer and weaker warming in winter. （4） At the station scale， JRA-55 showed the best applicability. ERA5 ranked second， while MERRA-2 showed the poorest applicability. The annual and seasonal climatic tendency rates of the observations， ERA5， and JRA-55 were mostly concentrated within 0. 30~0. 60 ℃·（10a）^-1， whereas those of MERRA-2 were mainly concentrated within 0. 15~0. 30 ℃·（10a）^-1. Overall， the climatic tendency rates of the three reanalysis datasets were generally lower than those of the observations， and all exhibited a warming pattern of stronger warming in summer and weaker warming in winter. （5） The applicability of the three reanalysis datasets varied among different subregions. ERA5， JRA-55， and MERRA-2 performed best in the eastern agricultural region， the area around Qinghai Lake， and the Qaidam Basin， respectively. All three datasets captured the subregional warming pattern of stronger warming in summer and weaker warming in winter. The observations showed that the warming trend was ranked as follows： eastern agricultural region， where strong warming occurred at Tongde and Datong and the annual and seasonal climatic tendency rates at Tongde were ≥0. 97 ℃·（10a）^-1， followed by the Qaidam Basin， where strong warming occurred at Mangya， and then the area around Qinghai Lake. ERA5 showed warming trends closer to the observations in the area around Qinghai Lake and the eastern agricultural region and thus performed the best， followed by JRA-55， while MERRA-2 showed the poorest applicability. （6） For the northern part of the Eastern Kunlun Mountains， ERA5 should be prioritized for regional climate analysis， whereas JRA-55 should be prioritized for refined studies at the station scale.

To evaluate the performance of numerical weather prediction （NWP） models over complex terrain and improve warm-season （May-October） precipitation forecasting over the Western Sichuan Plateau， this study assesses four models—ECMWF， CMA-GFS， CMA-MESO， and SWC-WARR—using 116 heavy rainfall events during 2022 -2024. Based on REOF regionalization and dominant circulation classification， the models are evaluated with traditional categorical scores， Fractions Skill Score （FSS）， and Method for Object-Based Diagnostic Evaluation （MODE）. Results show that the global model ECMWF has the highest overall TS and FSS scores and the lowest false alarm rate in northwestern Sichuan， but significantly underestimates extreme rainfall intensity. CMA-GFS performs well in intensity forecasts for plateau vortex-type events and accurately locates rainfall for plateau trough-type and plateau shear-line-type events. Among regional models， CMA-MESO has a higher hit rate， while SWC-WARR shows the best spatial pattern matching. Both perform better in the complex terrain of southwestern Sichuan for mesoscale processes such as high-pressure influence-type and dual-high shear-type events. However， they present obvious systematic biases， including overestimated rainfall areas， significant wet biases， and earlier predicted precipitation onset， leading to higher false alarm rates and lower TS scores. The diurnal peak of FSS appears at 11:00 -14:00 （Beijing Time）， caused by the overlap of early regional model forecasts and late global model forecasts with observed afternoon convection. Cross-section analyses indicate that wet biases result from overestimated mid-low-level humidity， excessive orographic lifting， displaced moisture convergence centers， and positional errors of key weather systems. Premature low-level convergence during the convective initiation stage is the main cause of early forecast onset. For operational applications， large-scale processes should refer to ECMWF/CMA-GFS to determine rainfall ranges， while mesoscale convective centers should rely on CMA-MESO/SWC-WARR. It is recommended to shrink the rainfall areas and delay the predicted onset time of regional models by 2~4 hours to fit the observed afternoon convective characteristics over the Western Sichuan Plateau.

To understand the complex wind patterns in Urumqi， which is a gorge city on the northern slope of the Tianshan Mountains， the study set up three Doppler wind lidars at different altitudes along the city's north-south direction. These instruments operated simultaneously throughout the winter season of 2023 to 2024. The reliability of wind lidar data was validated through comparison with synchronous radiosonde data. The vertical structural variations of wind fields in the southern， central， and northern parts of the city were then analyzed in conjunction with air pollution changes. The results revealed that wintertime PM_{2. 5} concentrations in Urumqi exhibited a distinct north-to-south decreasing gradient. The urban area experienced 50 days of lightly polluted or worse conditions， 7 of which were severe， highlighting the acute wintertime pollution challenge. The wind speed and direction data from the Doppler wind lidars exhibited strong correlation （R>0. 95） with the sounding data， with only minor deviations observed. This finding serves to substantiate the reliability of the scanning lidar-derived measurements. The airflow over Urumqi exhibits a distinct three‐layer structure： below 500 m， winds are predominantly northerly （valley breezes） or westerly； between 500 m and 2000 m， southeasterly foehn winds prevail； and above 2000 m， the flow is dominated by northwesterly winds. Shallow foehn winds occur throughout the city， but both the depth and wind speed of foehn airflow exhibited a marked south-to-north attenuation. Valley winds prevailed in Urumqi from 13:00 to 20:00 （Beijing Time， hereafter）. While the northern urban area exhibited the thickest valley airflow （up to 800 m）， both the duration and wind speed of these winds were significantly lower than those observed in the southern and central sectors of the city. During the winter months， the southern urban area exhibited ascending flows， in contrast to the subsiding motions observed in the central and northern sectors. However， all regions demonstrated weak vertical velocities. On days when air quality was optimal， northwesterly winds predominate， with an increase in wind speed with increasing altitude. As levels of pollution increase， the altitude of low-level weak winds （<2 m·s^-1） rises， leading to stagnant near-surface conditions； simultaneously， the valley wind layer weakens in both height and strength， while the foehn flow intensified markedly and its jet core descended. Examination of the low-level wind field showed that severe pollution events are characterized by a 70% occurrence frequency of low near-surface wind speeds. When combined with the intense foehn-driven inversion， this leads to localized pollutant accumulation and trapping.

Under the ongoing warming and humidification trend on the Qinghai-Xizang Plateau， the influence of rainfall on the hydrothermal conditions of frozen soil has become increasingly significant.To investigate the effects of rainfall characteristics on the hydrothermal state of seasonally frozen soil， this study analyzed the responses of frozen soil hydrothermal conditions under varying rainfall regimes based on continuous in-situ monitoring of meteorological variables and soil hydrothermal dynamics in Riwoqê County.The results indicate that the study area is dominated by short-duration （0 h˂H≤6 h） and small-amount （0.1 mm≤P≤9.9 mm） events， accounting for 57.81% and 84.39% of the total， respectively.Under similar rainfall amounts （7.6 mm≤P≤8.1 mm）， short-duration events during the thawing period accelerates cooling of the 0~15 cm soil layer （by 11.13%~81.49%） and reduces heating of the 25~35 cm layer （by 18.18%~85.07%）， while longer events primarily mitigate temperature changes.Short-duration rainfall also produces the largest increase in soil moisture （average 6.90% at 5~35 cm）， though the rates of increase and decrease generally decline with rainfall duration.Under comparable durations （10 h≤H≤13 h）， heavy rainfall induces rapid cooling in the 0~25 cm soil layer and slight warming at 35~45 cm （1.0 ℃ and 0.8 ℃， respectively）， whereas lighter rainfall mainly weakens temperature variation.Small， short-duration events （H≤5 h， P≤5 mm） slowed temperature change with minimal effects below 15 cm， and soil moisture exhibits greatest increases during the complete thawing period （0.59% at 2.5~15 cm） while smallest during complete freezing period.Overall， rainfall typically had a cooling effect， as theoretical rainwater temperatures were consistently lower than surface soil temperatures （hourly temperature difference was -1.57~30.99 ℃， daily average temperature difference was 0.45~15.86 ℃）， particularly from June to August.

极端降水常引发严重的人员伤亡和经济损失， 提高其预报准确率意义重大。鉴于当前数值模式在预报业务中的核心地位， 深入掌握模式性能是提高预报准确率的基础。本研究采用均方根误差（Root Mean Square Error， RMSE）、 空间TS评分（Threat Score， TS）和空间相关系数（Correlation Coefficient， COR）， 评估交互式全球大集合预报系统（THORPEX Interactive Grand Global Ensemble， TIGGE）资料集下中国气象局（China Meteorological Administration， CMA）、 美国国家环境预报中心（China Meteorological Administration， NCEP）、 欧洲中期天气预报中心（European Centre for Medium-Range Weather Forecasts， ECMWF）集合预报对2014 -2024年云南发生的21次极端降水事件的可预报性。结果表明： 总体上模式可预报性NCEP>ECMWF>CMA。集合90%分位值预报（ES90）和集合75%分位值预报（ES75）在3个模式中的可参考性均较高。进一步分析发现ES75（ES90）在降水量级（落区）方面具有更高的可预报性， 其中ECMWF>NCEP>CMA。在预报与实况的空间协同性方面ES75优于ES90。对于中雨、 大雨、 暴雨和大暴雨落区， 可预报性最高的分别是ECMWF-ES75、 ECMWF-ES90、 NCEP-ES90和CMA-ES90。不同模式的可预报性有明显的季节差异， 冬季ECMWF可预报性最高， 其他季节NCEP可预报性最高。不同系统影响下的极端降水事件可预报性也有一定差异， 对于槽前降水型， ECMWF可预报性最高， 对于其他影响系统型， NCEP的可预报性最高。进一步分析发现模式对极端降水事件的可预报性主要依赖于对影响系统的预报能力， 对系统范围和位置的可预报性决定了雨带位置的可预报性， 对系统强度和水汽输送的可预报性决定了降水量级的可预报性。模式对于大尺度系统影响下的降水预报效果要优于较小尺度系统。

This study evaluates ERA5 and MERRA2 reanalysis datasets against OMI and MLS observations to assess their performance in representing the spatial distribution， vertical structure， and temporal evolution of ozone over the Qinghai-Xizang Plateau during 2005 -2023， with an emphasis on dynamic transport within the summer ozone valley core. Both datasets capture the seasonal ozone valley， with minima in summer and maxima in winter. ERA5 generally outperforms MERRA2 in reproducing total column ozone， with biases within ±1 DU， whereas MERRA2 exhibits systematic winter underestimation exceeding -5 DU. Vertically， ERA5 better reproduces the 100~50 hPa core， while differences at the 30 hPa secondary minimum are minor， mainly reflecting variations in formation mechanisms and assimilation strategies. In terms of temporal evolution， ERA5 more accurately reproduces total column ozone trends， whereas MERRA2 better captures concentration trends within the 100~50 hPa core， especially during spring and summer. Zonal transport reduces ozone and strengthens the valley， whereas meridional transport partially compensates； MERRA2 overestimates transport intensity， particularly zonal transport， resulting in lower core concentrations. Divergence analysis reveals decreasing zonal and increasing meridional contributions， producing an overall rise in total transport. This pattern aligns with MLS observations and confirms the dominant role of dynamic processes in maintaining UTLS ozone minima. Overall， both datasets reproduce the ozone field effectively， though they differ in representing core intensity and temporal trends. Mid-to-upper layer biases are influenced by vertical resolution and assimilation coverage， with MERRA2 more sensitive to satellite gaps. Differences also arise from chemical and dynamical coupling that ERA5 better reproduces climatological averages， while MERRA2 more accurately captures local temporal variations. ERA5 is thus preferred for studies of climatological ozone characteristics， whereas MERRA2 is more suitable for analyzing temporal evolution.

Gridded precipitation， potential evapotranspiration （PET）， and land surface temperature （LST） datasets provide valuable alternatives for hydrologic modeling in ungauged areas.However， their performance in specific regions requires further investigation.This study evaluates six gridded precipitation products （CHIRPS， CMORPH， GPM， GSMaP， MSWEP， and PERSIANN）， three PET products （GLDAS， GLEAM， and ERA5-Land）， and one LST product （MOD11A1） in the Taolai River Basin.The effects of different precipitation and PET inputs on actual evapotranspiration （AET） and soil moisture were also analyzed.The results indicate that runoff simulations vary significantly depending on the precipitation product used.GPM and MSWEP perform best， achieving Nash-Sutcliffe efficiency （NSE） values of 0.58 and 0.54， respectively， during the validation period.These two products are reliable alternatives for precipitation sources in data-scarce scenarios.In contrast， the selection of PET products exhibits limited impact on runoff simulations， with the NSE values ranging from 0.61 to 0.74.If runoff simulation accuracy is the primary concern， the GLDAS， GLEAM， and ERA5-Land products are suitable substitutes.Additionally， when surface temperature data are unavailable， the MOD11A1 product achieves an NSE of 0.61 during the validation period.For scenarios with no meteorological observations， combining GPM or MSWEP precipitation data with any of the three PET products and MOD11A1 yields NSE values exceeding 0.53 in both calibration and validation periods.This study confirms that gridded datasets can partially mitigate hydrological simulation challenges in data-scarce regions， while recommending expanded applicability assessments across more basins to enhance reliability.

Accurate retrieval of raindrop size distributions （DSDs） based on dual-polarization radar can provide substantial data for the study of precipitation microphysical properties on a large scale.In order to further improve DSDs retrieval accuracy， this study proposes a new double-moment normalization method based on the sixth and seventh moments （M6M7 method）， comparing it with the third and sixth moments method （M3M6 method） and the constrained Gamma model DSD retrieval method （C-G method） from three perspectives： overall results， different rainfall intensities， and different rainfall particle sizes.Utilizing data from six rainfall events observed by dual-polarization radar and surrounding disdrometers at Heyuan station between May and June 2022， the retrieval results of each algorithm were analyzed.The results demonstrate that during light rain （0 mm·h^-1<R≤5 mm·h^-1）， the M6M7 method exhibits the smallest parameter biases among the three methods.As rainfall intensity increases， biases for most parameters （except for the increase of liquid water content （LWC） and rainfall rate （R）） remain relatively stable， with M6M7 consistently showing the lowest biases across different particle sizes and minimal fluctuation with the increase of particle size.Compared with M6M7 method， the M3M6 method incorporates specific differential phase shift on propagation （K_dp ） for retrieval.Although K_dp is noise-sensitive， it has good quality in heavy rain （R>30 mm·h^-1）， resulting in smaller estimation biases for intense rainfall events and a decreasing trend in bias with larger particle sizes （excluding LWC and R）.For moderate rain （5 mm·h^-1<R≤30 mm·h^-1）， the C-G method shows small median deviations yet significant fluctuations in certain parameters.With the increase of rainfall intensity and particle size， its biases shows a trend of first decreasing and then increase， accompanied by pronounced relative bias instability.Comprehensive evaluation results demonstrate that the M6M7 method consistently maintains median deviations approaching 0 across all DSD parameters， while exhibiting significantly tighter error fluctuation ranges.In marked contrast， both the M3M6 method and C-G method display substantially wider bias variability， with their error distributions spanning broader numerical ranges and demonstrating less stable performance characteristics.The newly proposed M6M7 method technique demonstrates advantages over traditional approaches， exhibiting enhanced comprehensive retrieval capabilities with regard to both accuracy and stability， particularly excelling in light-to-moderate rainfall with consistent accuracy.The M3M6 method proves more effective for heavy rain and storms， while the C-G method demonstrates unstable retrieval characteristics.The final section demonstrates the retrieval performance of the algorithm integrating both M6M7 and M3M6 methods， verifying its capability to further improve raindrop size distribution retrieval accuracy.