Understanding the role of climate change and catchment characteristics in hydrological activity is important for the efficient use of water resources. In this study, a Budyko framework suitable for non-steady conditions was used to assess the impacts of climate change and catchment characteristics on the long-term changes in annual and seasonal runoff in the Second Songhua River (SSR) basin during the last 30 years. Based on the analysis of the hydro-meteorological series of the SSR, the runoff in the SSR basin showed a non-significant increasing trend. The hydro-meteorological elements changed abruptly in 2009, and the study period was divided into a baseline period (1989–2009) and a disturbed period (2010–2018). Runoff increased during the disturbed period compared to the baseline period, with a significant increase in spring runoff in the upstream area and summer runoff in the downstream area. The attribution analysis results indicated that the annual runoff was mainly affected by climatic factors, and 66.8–99.6% of yearly runoff changes were caused by climate change. Catchment characteristics had little effect on yearly runoff but significantly affected seasonal runoff. The catchment characteristics affecting runoff were mainly increased water withdrawal, changes in snowfall, degradation of permafrost, and changes in reservoir operation. This study provides a basis for further understanding the intra-annual runoff variability for SSR and other similar rivers.

Analyzing the development trend of non-conventional water resources and identifying the main influencing factors is the initial step toward rapidly increasing the utilization and allocation of these resources in a rational and scientific manner. This will help relieve pressure on water resources and improve the ecological environment. This study introduces the concept of comparison testing and employs advanced Dematel and Random Forest models to identify two sets of optimal indicators from a pool of nine. Based on the two best indicator sets, three prediction models—BP neural network, Particle Swarm Optimization-optimized BP neural network, and Genetic neural network—were used to forecast the future potential of non-conventional water resource use in Heilongjiang Province. The findings reveal that economic indicators are the most significant factors influencing Heilongjiang Province’s utilization of non-conventional water resources. The findings of this study help us understand the extent of development in utilizing non-conventional water resources.

To lessen the strain on Harbin’s water resources and enhance the environment, it is crucial to analyze the key determining factors for the use of unconventional water resources in the city and to allocate unconventional water reasonably among various industries in the planning year. In this study, a back-propagation neural network (BP neural network) model is used to predict the potential for unconventional water resource utilization in the planning year (2025), a gray correlation analysis model is used to evaluate water-using industries, and finally, an unconventional water resource allocation scheme for the study is used to determine the main influencing factors and determine the weights of key indicators. The findings demonstrate a strong correlation between Harbin’s level of investment and construction, economic efficiency, cost, level of water demand, and social factors, as well as a low level of utilization of unconventional water resources throughout the city.

Accurate precipitation prediction is critical for water security and disaster mitigation, yet remains challenging due to atmospheric complexity and class imbalance in rainfall data. This study introduces an integrated "architecture-feature-augmentation" framework to address these limitations. Through systematic comparison of CNN-LSTM and Trans-former architectures, we identify a fundamental trade-off: CNN-LSTM demonstrates higher enhanceability, achieving 80% recall for heavy rainfall when combined with phys-ics-informed augmentation, while Transformer shows superior inherent sensitivity (75% recall) but greater vulnerability to data distribution shifts. Feature engineering benefits are model-specific, significantly improving CNN-LSTM but often introducing redundancy for Transformer. Notably, oversampling techniques like SMOTE achieve peak F1 scores but with substantial generalization gap (ΔF1 > 0.47), indicating overfitting risks, whereas physics-informed augmentation proves more reliable. We establish a principled decision framework: for robust predictions, use CNN-LSTM with physics-informed augmentation; for peak performance where risks are tolerable, employ CNN-LSTM with SMOTE. These findings provide scientific guidance for extreme weather preparedness and water resource management.

Accurate precipitation forecasting is paramount for water security and disaster mitigation, yet it remains formidable due to atmospheric stochasticity and the inherent class imbalance in rainfall datasets. This study proposes an integrated “architecture-feature-augmentation” framework to circumvent these limitations. Through a systematic evaluation of CNN-LSTM and Transformer architectures, we delineate distinct performance profiles: The Transformer model, when coupled with feature engineering and physics-informed augmentation, yields a peak F1-score of 0.1429, marking the optimal configuration for harmonizing precision and recall. Conversely, CNN-LSTM demonstrates superior robustness in extreme event detection, consistently maintaining high recall rates (up to 0.90) across diverse scenarios. We identify feature engineering as a critical performance modulator, substantially bolstering CNN-LSTM’s baseline metrics while enabling the Transformer to realize its maximum predictive capacity. Although synthetic oversampling techniques—such as SMOTE and GAN—effectively extend the detection range for heavy precipitation, physics-informed augmentation provides the most consistent performance gains, particularly in multi-class contexts. We conclude that the Transformer, augmented by physical constraints, is the optimal candidate for high-precision requirements, whereas CNN-LSTM, integrated with synthetic augmentation, offers a more sensitive alternative for early warning systems prioritizing recall. These findings provide empirical guidance for advancing extreme weather preparedness and strategic water resource management.

As global climate change intensifies, understanding drought mechanisms is crucial for managing water resources and agriculture. This study employs the Standardized Precipitation–Actual Evapotranspiration Index (SPAEI), Standardized Runoff Index (SRI), and Standardized Soil Moisture Index (SSMI) to analyze meteorological, hydrological, and agricultural droughts in the lower Songhua River basin. The PLUS model was used to predict future land types, with model accuracy validated using four evaluation metrics. The projected land cover was integrated with CMIP6 data into the SWAT model to simulate future runoff, which was used to calculate future SRI. Drought events were extracted using run theory, while drought occurrence probability and return period were calculated via a Copula-based joint distribution model. Bayesian conditional probability was employed to explore propagation mechanisms. The results indicate a significant increase in multidimensional drought risk, particularly when the cumulative frequency of univariate droughts reaches 25%, 50%, or 75%. Although increased duration and intensity enhance the likelihood of combined droughts, extremely high values cause a decline in joint probability under “OR” and “AND” conditions. Under different climate scenarios, the recurrence intervals of meteorological, hydrological, and agricultural droughts in the lower reaches of the Songhua River exhibit increased sensitivity with severity, demonstrating consistent propagation patterns across the meteorological–hydrological–agricultural system. Meteorological drought was found to propagate to hydrological and agricultural drought within ~6.00 months and ~3.67 months, respectively, with severity amplifying this effect. Propagation thresholds between drought types decreased with increasing intensity. This study combined SWAT and CMIP6 models with PLUS-based land-use scenarios, highlighting that land-use changes significantly influence spatiotemporal drought patterns. Model validation (Kappa = 0.83, OA = 0.92) confirmed robust predictive accuracy. Overall, this study proposes a multidimensional drought risk model integrating Copula and Bayesian networks, offering valuable insights for drought management under climate change.