Climate change is causing permafrost in the Qinghai-Tibet Plateau to degrade, triggering thermokarst hazards and impacting the environment. Despite their ecological importance, the distribution and risks of thermokarst lakes are not well understood due to complex influencing factors. In this study, we introduced a new interpretable ensemble learning method designed to improve the global and local interpretation of susceptibility assessments for thermokarst lakes. Our primary aim was to offer scientific support for precisely evaluating areas prone to thermokarst lake formation. In the thermokarst lake susceptibility assessment, we identified ten conditioning factors related to the formation and distribution of thermokarst lakes. In this highly accurate stacking model, the primary learning units were the random forest (RF), extremely randomized trees (EXTs), extreme gradient boosting (XGBoost), and categorical boosting (CatBoost) algorithms. Meanwhile, gradient boosted decision trees (GBDTs) were employed as the secondary learning unit. Based on the stacking model, we assessed thermokarst lake susceptibility and validated accuracy through six evaluation indices. We examined the interpretability of the stacking model using three interpretation methods: accumulated local effects (ALE), local interpretable model-agnostic explanations (LIME), and Shapley additive explanations (SHAP). The results showed that the ensemble learning stacking model demonstrated superior performance and the highest prediction accuracy. Approximately 91.20% of the total thermokarst hazard points fell within the high and very high susceptible areas, encompassing 20.08% of the permafrost expanse in the QTP. The conclusive findings revealed that slope, elevation, the topographic wetness index (TWI), and precipitation were the primary factors influencing the assessment of thermokarst lake susceptibility. This comprehensive analysis extends to the broader impacts of thermokarst hazards, with the identified high and very high susceptibility zones affecting significant stretches of railway and highway infrastructure, substantial soil organic carbon reserves, and vast alpine grasslands. This interpretable ensemble learning model, which exhibits high accuracy, offers substantial practical significance for project route selection, construction, and operation in the QTP.

The of the Yellow River between its source and Hekou Town in Inner Mongolia is known as the Upper Yellow River Basin. It is the main source area of water resources in the Yellow River Basin, providing reliable water resources for 120 million people. Studying the hydrometeorological changes in the Upper Yellow River Basin is crucial for the development of human society. However, in the past, there has been limited research on hydrometeorological changes in the Upper Yellow River Basin. In order to clarify the four-dimensional spatiotemporal variation characteristics of hydrometeorological elements in the Upper Yellow River Basin, satellite and reanalysis hydrometeorological elements products need to be used. Unfortunately, there is currently a lack of precise evaluation studies on satellite and reanalysis hydrometeorological elements products in the Upper Yellow River Basin, and the geomorphic characteristics of this area have raised doubts about the accuracy of satellite and reanalysis hydrometeorological elements products. Thus, the evaluation study in the Upper Yellow River Basin is an important prerequisite for studying the four-dimensional spatiotemporal changes of hydrometeorological elements. When conducting evaluation study, we found that previous evaluation studies had a very confusing understanding of the spatiotemporal characteristics of datasets. Some papers even treated the spatiotemporal characteristics of evaluation metrics as the spatiotemporal characteristics of datasets. Therefore, we introduced a four-dimensional spacetime of both datasets and evaluation metrics to rectify the chaotic spatiotemporal view in the past. Our research results show that satellite and reanalysis hydrometeorological elements products have different abilities in describing the temporal and spatial distribution and change characteristics of hydrometeorological elements. The difference in the ability of satellite and reanalysis hydrometeorological elements products to describe temporal and spatial distribution and change characteristics requires us to select data at different temporal and spatial scales according to research needs when conducting hydrometeorological research, in order to ensure the credibility of the research results.

Mountains are the water towers of the world, so it is critical to obtain accurate precipitation data for mountainous areas. Due to the complex topography of high mountainous areas, precipitation ground stations are sparse and unevenly distributed in such areas, so precipitation products such as remote sensing and reanalysis products are used to obtain gridded precipitation data for these areas. However, no single precipitation product performs best in all areas of mountainous regions. Therefore, this study first evaluated the performance of 12 precipitation products in estimating precipitation in the Qilian Mountains at the station scale and sub-basin scale, and then compared the performance of precipitation estimates for the Qilian Mountains generated by 8 multimodel averaging methods. The evaluation results for 29 meteorological stations in the Qilian Mountains showed that the China Meteorological Forcing Dataset product was the best-performing precipitation product, while the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Cloud Classification System-Climate Data Record product was the worst-performing precipitation product. The evaluation results for 18 sub-basins showed that at these sub-basins, the WorldClim was the best-performing precipitation product, while the High Asia Refined analysis was the worst-performing precipitation product. Thus, station-scale evaluations may not necessarily be applicable to the basin scale. Multi-model averaging methods effectively improved the accuracy of precipitation estimates both at station scale and at sub-basin scale. The Granger-Ramanathan variant C was the best multi-model averaging method for estimating precipitation at station scale. As the Granger-Ramanathan methods allow negative weights, they are not recommended to interpolate the Granger-Ramanathan weight values of stations to grids. The Bayesian model averaging (BMA) was found to be the most suitable multi-model averaging method for estimating precipitation in the Qilian Mountains by interpolation of weight values of stations to grids. The precipitation estimates generated by BMA show that the mean annual precipitation in the Qilian Mountains from 2001 to 2018 was approximately 336.1 mm, and the annual precipitation during this period increased linearly by 2.4 mm per year.

Global warming increases the frequency and intensity of climate extremes, but the changes in climate extremes over the Antarctic Ice Sheet (AIS) during different periods are unknown. Changes in surface temperature extreme indices (TN10p, TX10p, TN90p, TX90p, CSDI, WSDI, TNn, TNx, TXn, TXx and DTR) are assessed during 2021-2050 and 2071-2100 under SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5, based on the multi-model ensemble mean (MMEM) from the Coupled Model Intercomparison Project Phase 6 (CMIP6). The extreme indices, excluding TXn and DTR, illustrate the opposite trend in the two periods in SSP1-2.6 over the AIS. Generally, the changes in extreme indices reflect the continued warming over AIS in the future, and the warming is projected to intensify in SSP3-7.0 and SSP5-8.5. The variations in the extreme indices exhibit regional differences. The Antarctic Peninsula displays rapid changes in TNn, TXn and DTR. In SSP5-8.5, the magnitudes of all climate index tendencies are greater during 2071-2100 than 2021-2050. The variations in TX10p, TX90p, TN10p, TN90p, WSDI and CSDI are faster in the Antarctic inland than in the other regions over the AIS. However, the decrease in the DTR is concentrated along the AIS coast and extends to the interior region, whereas the increasing trend occurs in the Antarctic inland. In West AIS, TX90p and TN90p rapidly increase during 2021-2050, whereas the rapid changing signals disappear in this region in 2071-2100. The dramatic changes in TNn, TXn and DTR occur at the Ross Ice Shelf during 2071-2100, indicating an increased risk of collapse. For TNx and TXx, the degree of warming in the later part of the 21st century is divided by the transantarctic mountains, and greater changes appear on the eastern side. Generally, Antarctic amplification of TNn, TXn and DTR is observed except under SSP1-2.6. In addition, TNx and TXx amplifications occur in SSP3-7.0 and SSP5-8.5.

Long-term and high-resolution gridded products of precipitation and temperature data are highly important to study the changes in climate and environment under global warming. Considering the uncertainties of these products in mountainous areas, it is necessary to evaluate the data reliability. This study evaluates the performances of the CMFD (China Meteorological Forcing Dataset) and ERA5-Land in simulating precipitation and temperature in the Qilian Mountains over the period of 1980-2018. We use the observation data of 28 basic meteorological stations in the Qilian Mountains to compare with the reanalysis products. Error metrics (the correlation coefficient (CC), the root mean square error (RMSE), the mean absolute error (MAE), and the relative bias (BIAS)) are used to quantify the monthly differences in existence between the observed data and reanalysis data. Our findings indicate that both CMFD and ERA5-Land could well reproduce the spatial distribution of mean monthly precipitation and temperature in the region. A good correlation is found between CMFD and OBS under different amounts of monthly precipitation conditions. The monthly average temperatures of CMFD and ERA5-Land reveal a high correlation with the observed results. Moreover, the CC values of CMFD and ERA5-Land precipitation products are the highest in autumn and the lowest in winter, and the CC values of both CMFD and ERA5-Land temperature products are higher in spring and autumn. However, we find that both reanalysis products underestimate the temperature to varying degrees, and the amount of precipitation is overestimated by ERA5-Land. The results of the evaluation show that the errors in precipitation yielded by CMFD as a whole are distinctly fewer than those yielded by ERA5-Land, while the errors in air temperature yielded by both ERA5-Land and CMFD are nearly identical to each other. Overall, ERA5-Land is more suitable than CMFD for studying the trends of temperature changes in the Qilian Mountains. As for simulation of precipitation, CMFD performs better in the central and eastern parts of the Qilian Mountains, whereas ERA5-Land performs better in the western part of the Qilian Mountains.

The surface seasonal freeze/thaw (F/T) signal detected by passive microwave remote sensing is very important for the water cycle, carbon cycle and climate change research. In this study, we evaluated and analyzed the Soil Moisture Active Passive (SMAP) L3 F/T product, Advanced Microwave Scanning Radiometer 2 (AMSR2) F/T product and Making Earth System Data Records for Use in Research Environments (MEaSUREs) F/T product over different regions in China, including the Genhe area in Northeast China, the Saihanba area in North China, and the Qinghai-Tibet Plateau (QTP) area. The overall accuracy of F/T products assessed with the 5 cm depth soil temperature is 90.38% for SMAP, 90.23% for AMSR2 and 84.73% for MEaSUREs in cold and humid temperate forest climates and the plateau continental climate area (Genhe, Tianjun, and Qumalai) where permafrost is distributed, and 76.64% for SMAP, 83.67% for AMSR2 and 77.37% for MEaSUREs in the cold plateau mountain climate and plateau continental climate area (Saihanba and Chengduo) with frozen ground distributed seasonally, respectively. The overall accuracy is 69.05% for SMAP, 76.5% for AMSR2 and 81.4% for MEaSUREs in the Ngari, Naqu, and Dachaidan regions belonging to arid and semi-arid climates. It can be seen that SMAP and AMSR2 achieve the best performance in the distributed permafrost area, the second-best performance in the seasonal distributed permafrost area, but the worst performance in the areas with arid and semi-arid climate types due to inconsistent F/T signals between water with small changes and temperature with apparent changes during the F/T transition. The MEaSUREs product showed almost the same performance in different regions, indicating that it was less affected by climate types and the distribution of frozen soil than SMAP and AMSR2 products. SMAP F/T product detected by L-band with long penetration and AMSR2 F/T product calibrated with 5 cm soil temperature could represent the 5 cm F/T, but the MEaSUREs F/T product was more likely to describe the surface F/T state due to calibrated with air temperature and the short penetration of 36.5 GHz. In mid-low latitude areas (Tianjun and Qumalai) with a short duration of snow cover days and a fast snowmelt, the effect of snow melting on F/T products was negligible. Moreover, the spring snowmelt affects the three F/T products in Chengduo, but the SMAP product is not affected by the winter snowmelt, whereas the AMSR2 product is affected by the winter snowmelt.

A version of the Community Earth System Model modified at the North Carolina State University (CESM-NCSU) is used to simulate the current and future atmosphere following the representative concentration partway scenarios for stabilization of radiative forcing at 4.5 W m(-2) (RCP4.5) and radiative forcing of 8.5 W m(-2) (RCP8.5). Part I describes the results from a comprehensive evaluation of current decadal simulations. Radiation and most meteorological variables are well simulated in CESM-NCSU. Cloud parameters are not as well simulated due in part to the tuning of model radiation and general biases in cloud variables common to all global chemistry-climate models. The concentrations of most inorganic aerosol species (i.e., SO42-, NH4+, and NO3-) are well simulated with normalized mean biases (NMBs) typically less than 20%. However, some notable exceptions are European NH4+, which is overpredicted by 33.0-42.2% due to high NH3 emissions and irreversible coarse mode condensation, and Cl-, that is negatively impacted by errors in emissions driven by wind speed and overpredicted HNO3. Carbonaceous aerosols are largely underpredicted following the RCP scenarios due to low emissions of black carbon, organic carbon, and anthropogenic volatile compounds in the RCP inventory and efficient wet removal. This results in underpredictions of PM2.5 and PM10 by 6.4-55.7%. The column mass abundances are reasonably well simulated. Larger biases occur in surface mixing ratios of trace gases in CESM-NCSU, likely due to numerical diffusion from the coarse grid spacing of the CESM-NCSU simulations or errors in the magnitudes and vertical structure of emissions. This is especially true for SO2 and NO2. The mixing ratio of O-3 is overpredicted by 38.9-76.0% due to the limitations in the O-3 deposition scheme used in CESM and insufficient titration resulted from large underpredictions in NO2. Despite these limitations, CESM-NCSU reproduces reasonably well the current atmosphere in terms of radiation, clouds, meteorology, trace gases, aerosols, and aerosol-cloud interactions, making it suitable for future climate simulations. (C) 2016 Elsevier Ltd. All rights reserved.

[1] New aerosol modules of global ( circulation and chemical transport) models are evaluated. These new modules distinguish among at least five aerosol components: sulfate, organic carbon, black carbon, sea salt, and dust. Monthly and regionally averaged predictions for aerosol mass and aerosol optical depth are compared. Differences among models are significant for all aerosol types. The largest differences were found near expected source regions of biomass burning ( carbon) and dust. Assumptions for the permitted water uptake also contribute to optical depth differences ( of sulfate, organic carbon, and sea salt) at higher latitudes. The decline of mass or optical depth away from recognized sources reveals strong differences in aerosol transport or removal among models. These differences are also a function of altitude, as transport biases of dust do not always extend to other aerosol types. Ratios of optical depth and mass demonstrate large differences in the mass extinction efficiency, even for hydrophobic aerosol. This suggests that efforts of good mass simulations could be wasted or that conversions are misused to cover for poor mass simulations. In an attempt to provide an absolute measure for model skill, simulated total optical depths ( when adding contributions from all five aerosol types) are compared to measurements from ground and space. Comparisons to the Aerosol Robotic Network (AERONET) suggest a source strength underestimate in many models, most frequently for ( subtropical) tropical biomass or dust. Comparisons to the combined best of Moderate-Resolution Imaging Spectroradiometer ( MODIS) and Total Ozone Mapping Spectrometer ( TOMS) indicate that away from sources, model simulations are usually smaller. Particularly large are discrepancies over tropical oceans and oceans of the Southern Hemisphere, raising issues on the treatment of sea salt in models. Totals for mass or optical depth in many models are defined by the absence or dominance of only one aerosol component. With appropriate corrections to that component ( e. g., to removal, to source strength, or to seasonality) a much better model performance can be expected. Still, many important modeling issues remain inconclusive as the combined result of poor coordination ( different emissions and meteorology), insufficient model output ( vertical distributions, water uptake by aerosol type), and unresolved measurement issues ( retrieval assumptions and temporal or spatial sampling biases).