Streamflow components (surface runoff and baseflow) play distinct regulatory roles in global water cycles and ecosystem stability; however, systematic understanding of their differential responses to climate change remains limited. This study developed a two-stage runoff model based on the generalized proportionality hypothesis. Using the trend-preserving bias-corrected climate data from five regional climate models under RCP4.5 and RCP8.5 scenarios, we systematically projected streamflow component responses across 56 hydrological stations nested in 13 basins of China's Loess Plateau during 2021-2050. Model validation demonstrated satisfactory performance (R2 = 0.898 and 0.902 for surface runoff and baseflow, respectively), compared with the observed data. Under RCP4.5, increased precipitation (ranging from 5.7 mm/10a to 46.3 mm/10a) raised surface runoff at 95 % of stations and baseflow at 80 % of stations. Under RCP8.5, enhanced evapotranspiration (ranging from 13.6 to 18.8 mm/10a) significantly reduced baseflow at 30 % of stations, far exceeding the 11 % for surface runoff. This differential response originated from inherent hydrological characteristics: rapid surface runoff generation triggered by storm rainfall, and slow soil water-to-groundwater transformation processes. The model provides a novel analytical framework for coupled streamflow component assessment and adaptive water resource management under changing environments.

Leaf Area Index (LAI) is a key biophysical descriptor of crop canopies and is essential for growth monitoring and yield estimation. We present a physics-driven machine-learning framework for operational LAI retrieval and end-to-end uncertainty quantification that couples the PROSAIL radiative transfer model with a genetic-algorithm-optimised multilayer perceptron (NN–GA). PROSAIL is sampled across plausible parameter priors and spectra are convolved with Sentinel-2B spectral response functions to build a 30,000-sample training library; a GA is used to globally optimise network weights and biases. Total retrieval uncertainty is decomposed into a simulation component (PROSAIL parameter variability) and a training component (variability across repeated NN–GA trainings) and combined via the law of propagation of uncertainty. The model was developed in Minqin (modelling/testing area; entirely maize) and transferred to Zhangye (transfer/validation area; predominantly maize, with one sunflower plot). Sentinel-2B validation results were RMSE/R2 = 0.44/0.73 (Minqin) and 0.40/0.56 (Zhangye), indicating reasonable cross-site generalisation. The uncertainty split indicates physical-driven contributions of 11.42% and 11.48% and machine-learning contributions of 18.06% and 12.96%, respectively. The framework improves 10 m LAI retrieval accuracy and supplies a reproducible, per-pixel uncertainty budget to guide product use and refinement.

Hydropower dispatching faces challenges such as uneven distribution of water resources, obvious seasonal changes and large fluctuation of power demand. Traditional dispatching methods rely on manual experience and simple mathematical models, and it is difficult to effectively deal with complex and changeable dispatching scenarios. This paper analyzes the role of large model in data integration and analysis, intelligent decision support, knowledge base construction and update, and man-machine cooperation mechanism, and takes a large hydropower station in China as a practical case to show the application practice of large model technology. By constructing a hydrological forecasting model that includes Long Short Term Memory (LSTM) networks, a Convolutional Neural Network (CNN) power demand forecasting model, and a Reinforcement Learning (RL) algorithm for optimizing generator scheduling, the intelligence and precision of hydropower scheduling have been achieved. In the human-machine collaboration mode, human dispatchers work closely with large model systems to jointly complete hydroelectric dispatch tasks, significantly improving dispatch efficiency and economic benefits. The research results indicate that the application of large models has improved the economic benefits and environmental sustainability of hydropower stations, verifying their effectiveness and importance in hydropower scheduling optimization.

Water quality evaluation usually relies on limited state-controlled monitoring data, making it challenging to fully capture variations across an entire basin over time and space. The fine estimation of water quality in a spatial context presents a promising solution to this issue; however, traditional analyses often ignore spatial non-stationarity between variables. To solve the above-mentioned problems in water quality mapping research, we took the Yangtze River as our study subject and attempted to use a geographically weighted random forest regression (GWRFR) model to couple massive station observation data and auxiliary data to carry out a fine estimation of water quality. Specifically, we first utilized state-controlled sections’ water quality monitoring data as input for the GWRFR model to train and map six water quality indicators at a 30 m spatial resolution. We then assessed various geographical and environmental factors contributing to water quality and identified spatial differences. Our results show accurate predictions for all indicators: ammonia nitrogen (NH3-N) had the lowest accuracy (R2 = 0.61, RMSE = 0.13), and total nitrogen (TN) had the highest (R2 = 0.74, RMSE = 0.48). The mapping results reveal total nitrogen as the primary pollutant in the Yangtze River basin. Chemical oxygen demand and the permanganate index were mainly influenced by natural factors, while total nitrogen and total phosphorus were impacted by human activities. The spatial distribution of critical influencing factors shows significant clustering. Overall, this study demonstrates the fine spatial distribution of water quality and provides insights into the influencing factors that are crucial for the comprehensive management of water environments.

The ecosystem services provided by grasslands depend on their biomass or leaf area index (LAI). Under the background of climate change, the impact of preseason climate and extreme weather events, such as high temperature and drought, on the spatiotemporal dynamics of grassland LAI remains unclear. Here, we constructed three interpretable machine learning models (including a Bayesian model, an interpretable neural network model, and a random forest model), to investigate the impact mechanisms of climate factors on grassland LAI in China from 2001 to 2020. The results showed that all three models performed well in simulating LAI (with R2 ranging from 0.540 to 0.963). The random forest model performed the best. Preseason climate was the most important factor driving LAI changes. The increase in preseason temperature, precipitation, and radiation could lead to an increase in grassland LAI. Regarding extreme weather events, heat events and heavy-rainfall events had positive effects on LAI, while frost events and no-rainfall events had negative impacts. CO2 showed a significant fertilization effect. Grazing intensity had a relatively small impact on LAI. The impact of precipitation on LAI was greater in spring than in autumn, whereas the impacts of temperature and radiation were greater in autumn than in spring. This study develops a climate change-adaptive framework for predicting grassland growth dynamics based on machine learning models, which can provide scientific support for grassland ecological protection and adaptive management. The random forest model performed the best in simulating grassland LAI. Preseason climate was the most important factor driving LAI changes. Heat events and heavy-rainfall events had positive effects on grassland LAI. CO2 showed a significant fertilization effect on grassland LAI. The impact of precipitation on LAI was greater in spring than in autumn.

Traditional fault detection methods often rely on manual experience and regular inspections, which are not only inefficient but also difficult to detect potential safety hazards in a timely manner. In the study, the power data was first preprocessed, including data cleaning, denoising, and normalization, and then key features were extracted using signal processing techniques. A hybrid model was constructed using a deep neural network (DNN) framework, combined with convolutional neural networks (CNN) and long short-term memory networks (LSTM), to handle nonlinear relationships and temporal dynamics in power data. In addition, a power knowledge map is constructed, and the map information is combined with the output of deep learning model through attention mechanism to enhance the model's ability to understand and predict fault modes. The experimental results show that the model based on knowledge map is generally higher than the model based on deep learning in the accuracy of fault prediction, showing better robustness and adaptability. Case analysis further verifies the effectiveness of the model, such as successfully predicting transformer overheating fault and transmission line disconnection caused by strong wind. This study provides a new method for intelligent fault prediction of power system, which is helpful to realize early detection and prevention of power system and ensure the stable operation of power system.

In this paper, a method combining deep learning and natural language processing (NLP) technology is proposed. Through the fusion of large model and knowledge map, structured knowledge is extracted from massive and multi-source power system data, and a knowledge map with rich semantic relations is constructed. In this study, BERT model is used to identify entities, extract relationships and identify attributes. Through the steps of knowledge fusion and map construction, the uniqueness of entities and the unified representation of relationships are ensured. At the same time, new relationships or attributes are inferred by using graph neural network (GNN) method to improve the knowledge map. In addition, a hierarchical power system planning decision support system (DSS) is designed, including data layer, knowledge layer, application layer and user layer, supported by large model and AI technology. Through case analysis, the application of the system in power system planning of large cities is demonstrated, including knowledge extraction from technical documents, knowledge fusion, risk assessment and optimization analysis, which significantly improves the accuracy, efficiency and safety of decision-making. This study not only improves the intelligence and automation level of power system planning, but also provides scientific decision support for the planning and operation of power system, which is of great significance to the sustainable development of power industry.

Climate-sensitive alpine permafrost rivers on the Qinghai-Tibet Plateau export pre-aged carbon (C), yet the role of this pre-aged C in greenhouse gas emissions remains unclear. Here, through radiocarbon analysis of riverine CO2, CH4, dissolved organic carbon (DOC), and sediment organic carbon (SOC), we show that DOC and SOC originate primarily from pre-aged C, and it is more readily degraded than modern C due to lower aromaticity, with microorganisms preferentially metabolizing permafrost-derived fractions. Pre-aged C accounts for >60% of both CO2 and CH4 emissions. CO2 derives mainly from pre-aged active layer C decomposition, with increased terrigenous fresh CO2 input entering rivers during the growing season. Notably, ancient permafrost C contributes up to similar to 30.5% of total CH4 emissions. Our findings highlight that ancient C released by alpine permafrost rivers substantially contributes to greenhouse gas emissions, emphasizing the need to improve global C models and formulate climate-resilient mitigation frameworks to support sustainable development.

The government of China has implemented the Southern Shaanxi Disaster Resettlement program since 2011, which aims to address the problems of reduced livelihood resilience, increased livelihood risks, and single-risk management strategies caused by the frequent occurrence of natural disasters. This study considers the specific situation of disaster resettlement in Ankang Prefecture, southern Shaanxi Province, and draws on Quandt’s measurement idea to quantify livelihood resilience at the household scale in terms of five types of capital assets: natural, physical, human, financial, and social. A coarsened exact matching model was used to control confounding factors in the observational data to reduce sample selection bias, and then multinomial logit regression models were used to examine how livelihood resilience affects risk management strategies; moreover, the effects of different indicators of livelihood resilience, relocation characteristics, and follow-up support measures on risk management strategies were analyzed. Results show that livelihood resilience is higher among new-stage relocation, voluntary relocation, and centralized resettlement households, and working outside of the home accounts for the largest proportion of risk management strategies chosen by the sample households. In addition, livelihood resilience and its dimensions and indicators, relocation characteristics, and follow-up support measures have different impacts on risk management strategies. These results have considerable significance in guiding research on risk management strategies at the household scale and can serve as a reference for disaster resettlement in other developing nations and regions.

Maize, a significant global food crop, is essential in agriculture and the economy. The price of maize futures is affected by many factors, and its data is a nonlinear, unstable, and long-term correlation, so it is difficult to predict it accurately. Accordingly, this paper presents a combined model based on Ensemble Empirical Mode Decomposition (EEMD), Convolutional Neural Network (CNN), and Improved Gated Recurrent Unit (IGRU). EEMD decomposes the maize price data to produce multiple Intrinsic Mode Function (IMF) components and residual sequences. Subsequently, the noisy IMF components are removed, and the remaining IMF components are reconstructed into low, medium, and high frequencies. The CNN is tasked with the extraction of eigenvalues for these components. The IGRU improves in two key respects on the original Gated Recurrent Unit (GRU) model. First, it enhances the update and reset gates. Second, it incorporates the Self-Attention (SA) module. Thereby, the model's predictive capabilities are improved. This study uses the primary maize futures trading data from the Dalian Commodity Exchange as the experimental data. A comparative analysis of the EEMD-CNN-IGRU model with seven baseline models shows that it outperforms other models in all evaluation indexes.

The largest disaster reduction and relocation project was conducted in Shaanxi Province, China, in an effort to reduce the threat of natural disasters and preserve the environment. Although the literature has attempted to assess the economic and ecological impacts of the project quantitatively, there is currently a dearth of research on the connection between resource dependence and adaptive capacity at the rural household levels. Using survey data from southern Shaanxi, China, this study evaluated and quantified natural resource dependence (NRD) and household adaptive capacity (HAC) in the context of disaster resettlement. Simultaneously, we explored the differences in NRD and HAC among various groups and relocation characteristics. An ordinary least squares regression model was used to specifically examine the relationship between them. The results indicated that, first, NRD was significantly and positively related to HAC. Second, the dependence of relocated households on energy, food, and income had a significantly positive correlation with HAC. Third, compared to local, involuntary, and scattered resettlement households, the HAC of relocated households, voluntary relocated households, and centralized resettlement households is substantially lower. The aforementioned findings have significant policy implications for rural China and other developing nations, as they can help reduce natural resource dependence and increase adaptive capacity.

In order to obtain power quality disturbance information accurately and efficiently, a power quality disturbance detection method based on lifting wavelet and fast Fourier transform (FFT) is proposed. Firstly, we use Euclidean algorithm to realize db4 wavelet transform, and verify the characteristics and shortcomings of lifting algorithm in processing transient and steady-state disturbance signals. Then, fast Fourier transform is carried out on the reconstructed steady-state component by utilizing the decomposition and reconstruction characteristics of lifting wavelet transform, so as to make up for the defects of steady-state disturbance processing by lifting wavelet and realize accurate and fast detection of transient and steady-state complex disturbance. The mode maximum is used to judge power quality disturbance in advance. The simulation results show that the proposed method can judge and deal with power quality in complex cases involving transient state and steady state, and has higher positioning accuracy and accuracy than traditional wavelet transform, which verifies the accuracy and efficiency of the proposed method for power quality disturbance detection

The frequent occurrence of urban floods (UFs) poses significant threats to citizens’ lives and the national economy. Utilizing machine learning to assess urban flood susceptibility (UFS) provides valuable decision support for UF management. However, the precision of current studies is usually influenced by the variability of temporal factors like extreme rainfall, which limits the accurate identification of urban flood-susceptible regions (UFSRs). To address this issue, we present a novel approach that leverages the spatiotemporal distribution and characteristics of UFS to accurately identify UFSRs. In our case study of the Greater Bay Area (GBA) in China, we employed the Random Forest to assess the spatiotemporal distribution of UFS. We then used the Savitzky-Golay filter to correct UFS data based on the UFS time series from 2011 to 2020. The Theil-Sen median slope, Mann-Kendall test, and Hurst analysis were used to explore the spatiotemporal patterns of UFS. Shapley additive explanation was applied to quantify the contribution of selected variables. Our findings include: (1) UFS in the GBA demonstrates a rising trend, with high susceptibility areas increasing from 6.3 % in 2011 to 7.4 % in 2020; (2) UFSRs, covering approximately 11 % of the GBA, are primarily concentrated in the cities located around the central GBA; and (3) human behavior factors have a more significant influence on UF than natural ones. We believe the presented framework for the accurate extraction of UFSRs provides valuable decision support for sustainable city development.

Urban Flood (UF) and Urban Heat Island (UHI) become prevalent “urban diseases” currently. Previous studies usually regarded UF and UHI as separate issues that requires respective concentration; however, UF and UHI are inextricably connected so that they should be considered together. Hence, this paper proposes a spatiotemporal framework to link UF and UHI by developing an Urban Flood Heat Island (UFHI) index, taking the Greater Bay Area (GBA) in China as the study case. Random Forest (RF) classifier and Urban-Rural Dichotomy were selected to compute the spatiotemporal distribution of UF susceptibility (UFS) and UHI intensity (UHII), respectively. Theil-Sen Median Slope, Mann-Kendall Test, and Hurst analysis were used to explore the spatiotemporal variation pattern of the UF-UHI joint risk. Spatial Durbin Model was used to compute the correlation between UFS and UHII. RF regressor was employed to quantify the contribution of the common driving factors. We found: (1) UFS and UHII are spatiotemporally correlated, thus UHII can be used as a control variable for the refined UFS assessments, (2) the areas with the UF-UHI joint risk of persistently significant increase (∼7.9 %) majorly distributed in the central region, (3) according to the contributions of factors, UF and UHI in the GBA could be effectively alleviated by breaking the continuous impervious surfaces by introducing blue-green structures. This study offers a new perspective for the effective alleviation of UF and UHI, which would be helpful for the sustainable development of cities in developing countries where human and material sources are limited.

Mapping the spatial distribution of artificial grassland for ecological restoration is of great significance for evaluating its secondary degradation and negative consequences, such as nonpoint source pollution of water bodies in the Three-River Headwaters (TRH) region. Because of the numerous challenges faced in obtaining ground training samples caused by adverse natural conditions, inclement cloudy weather and spectral similarity between natural and artificial grassland, commonly used classification or temporal-profile extraction methods have proven ineffective in identifying artificial grasslands. To overcome these challenges, we present a novel artificial grassland detection index for mapping their distribution using optical images with a resolution of 10 ∼ 30 m, along with their corresponding quality control data based on the Google Earth Engine cloud computing platform. The index is calculated using the ratio of the normalized difference vegetation index during the sowing and emergence period and the growth peak period of artificial grassland. A case study was conducted in Maqin County in the TRH region covering an area of 1.35 × 104 km2 from 2017 to 2021. Our proposed method demonstrated high accuracy and achieved a favorable balance between commission errors and omission errors. Over the study period, the average overall accuracy and Cohen's kappa were 96.2% and 0.91, respectively; with average precision, recall, and F1-score of artificial grassland being 89.6%, 99.2%, and 94.0%, respectively. The proposed method exhibited excellent robustness for the critical threshold used, with the average overall accuracy, F1-score, precision, and recall of artificial grassland between 2017 and 2021 consistently exceeding 90% for threshold values ranging from 1.5 to 2.0 throughout the study period. These findings suggest that our proposed method is capable of efficiently and accurately obtaining the detailed spatiotemporal distribution of artificial grassland in the TRH region. Moreover, the method also meets the pressing requirement for the rapid acquisition of detailed spatiotemporal distribution of artificial grassland across the Qinghai-Tibet Plateau.

Machine learning (ML)-based urban waterlogging susceptibility studies suffer from class imbalance, as fewer positive samples are generally available than potential negative samples. Few studies have considered optimizing the results by improving the quality of training samples. To address this issue, we explored effective approaches to reliably increase the numbers of positive samples for such studies. The Synthetic Minority Over-Sampling Technique (SMOTE) and Optimized Seed Spread Algorithm (OSSA), representative of oversampling (synthesizing new samples based on the feature space) and physical (simulating potential inundated area based on the mechanisms of water flow) approaches, respectively, were employed to increase the number of positive samples. Waterlogging in Shenzhen was selected as a case study using eight selected spatial variables. An elaborate experiment was conducted to compare the quality of added samples based on the classifiers' performance and accuracy of waterlogging susceptibility maps (WSMs). The results indicated that (1) the performance of classifiers generated with SMOTE was worse than the original samples, while the use of OSSA improved the trained classifiers, and (2) the accuracy of WSMs was not improved with SMOTE but increased markedly with OSSA. These results may be driven by the diversity of information and features of the added samples. This study indicates the use of SMOTE fails to synthesize reliable samples when applied to waterlogging analysis in Shenzhen, whereas an effective solution for generating reliable positive samples is to use OSSA that simulates the potential submerged regions based on the mechanisms of disaster occurrence and spread.

Remote sensing (RS) models can easily estimate the net primary productivity (NPP) on a large scale. The majority of RS models try to couple the effects of temperature, water, stand age, and CO2 concentration to attenuate the maximum light use efficiency (LUE) in the NPP models. The water effect is considered the most unpredictable, significant, and challenging. Because the stomata of alpine plants are less sensitive to limiting water vapor loss, the typically employed atmospheric moisture deficit or canopy water content may be less sensitive in signaling water stress on plant photosynthesis. This study introduces a soil moisture (SM) content index and an alpine vegetation photosynthesis model (AVPM) to quantify the RS NPP for the alpine ecosystem over the Three-River Headwaters (TRH) region. The SM content index was based on the minimum relative humidity and maximum vapor pressure deficit during the noon, and the AVPM model was based on the framework of a moderate resolution imaging spectroradiometer NPP (MOD17) model. A case study was conducted in the TRH region, covering an area of approximately 36.3 × 104 km2. The results demonstrated that the AVPM NPP greatly outperformed the MOD17 and had superior accuracy. Compared with the MOD17, the average bias of the AVPM was −9.8 gCm−2yr−1, which was reduced by 91.8%. The average mean absolute percent error was 57.0%, which was reduced by 68.2%. The average Pearson’s correlation coefficient was 0.4809, which was improved by 30.0%. The improvements in the NPP estimation were mainly attributed to the decreasing estimation of the water stress coefficient on the NPP, which was considered the higher constraint of water impact on plant photosynthesis. Therefore, the AVPM model is more accurate in estimating the NPP for the alpine ecosystem. This is of great significance for accurately assessing the vegetation growth of alpine ecosystems across the entire Qinghai–Tibet Plateau in the context of grassland degradation and black soil beach management.

The frequent occurrence of wildfires presents a serious threat to human livelihoods and local ecosystems. The use of machine learning (ML) methods to assess wildfire susceptibility can provide decision support for disaster prevention. However, most current ML-based wildfire susceptibility assessments overly focus on spatially evaluating the disaster threat, while ignoring the potential threats of wildfires to local ecosystems. This situation makes it difficult to determine seasonal variations in wildfire susceptibility and limits the value of assessment results. We present a framework to assess wildfire susceptibility and wildfire threats seasonally to local ecosystems. The ecosystem service value (ESV) was used as a proxy for the economic value of an ecosystem, the random forest algorithm was used to evaluate wildfire susceptibility, and the Daxinganling region, the largest forested area in China, was selected as the study area, and the dynamic equivalent coefficient factor method was used to calculate the ESV of each cell. Our main findings were as follows: (a) wildfire susceptibility exhibited obvious disparities in terms of spatial distribution across the four seasons; (b) each ecosystem in the study area faced a different magnitude of wildfire disturbance; and (c) the expected ESV loss (USD 10.8 billion) due to wildfires was much higher than the region's total GDP (USD 2 billion) in 2019. This study was repeatable, and all data required were obtained freely. The methodologies used can be applied directly to other regions. Our study will be of particular interest to developing counties where intensive wildfire monitoring is limited.

With the development of smart grid, smart meters have been widely used, and the Non-Intrusive Load Monitoring (NILM) technology for sensing and identifying electricity load has gradually matured. Aiming at the problem that the characteristics of industrial users’ load status are few and the gap between different industries is large, this paper uses the random forest algorithm to build a load identification model, and at the same time builds an energy-saving potential evaluation index system including three dimensions of economy, technology and management. Based on the load identification results, users can evaluate the energy-saving potential, and output the use strategies and power consumption suggestions of electric equipment.