Open research questions in Hydrological Forecasting Using AI

What the literature leaves open

• Future research should focus on integrating higher-resolution satellite imagery, near real-time monitoring systems, and advanced hybrid artificial intelligence approaches, such as Physics-Informed Neural Networks (PINNs), to enhance the physical realism, scalability, and predictive accuracy of future LULC simulations.

  GEE-integrated ML classifiers evaluation for LULC change detection and CA-Markov-based future prediction in Upper Blue Nile River Basin, Ethiopia · 2026 · DOI

• Limited studies have integrated time series (TS) and machine learning (ML) for estimating agricultural water footprints (AWF) and the agricultural water scarcity index (AWSI).

  Machine learning and time series integration for agricultural water footprints and scarcity index estimation in Northern Nigeria · 2026 · DOI

• Future work will focus on further simplifying the model architecture and optimizing computational efficiency, as well as exploring its extension to larger-scale, multi-site, and cross-regional water quality prediction tasks, in order to enhance its practical value in real-world water environment management and ecological conservation. The generalization capability of the proposed approach for more complex spatially heterogeneous environments or cross-regional water quality prediction tasks remains to be further evaluated and extended in future work.

  A multi-feature water quality prediction method based on adaptive denoising and periodicity-aware residual learning · 2026 · DOI

• 2022) corroborated the ability of ANN to model complex water quality dynamics but did not account for the unique distribution of water quality variability in rural Indian networks.

  Prediction of Residual Chlorine in Water Distribution Network Using Artificial Neural Network (ANN) · 2026 · DOI

• Analysis was limited to a single airport, 12 landfalling typhoons, and a 1-h deterministic lead time, and only near-surface measurements, which cannot rep- resent larger-scale typhoon structure or upstream condi- tions, were employed.

  Application of LSTM neural networks with multivariate numerical analysis to aviation wind gust forecasting · 2026 · DOI

• This learning approach, widely used in surrogate modeling and increasingly adopted in Earth-surface research where obser- vational datasets are sparse, leverages a synthetic, noise-free, and complete dataset to enable robust DL training while avoiding the practical constraints of field data collection (De Melo et al.

  An integrated deep learning framework enables rapid spatiotemporal morphodynamic predictions toward long-term simulations · 2026 · DOI

• In summary, the combination of ML and DL models improved streamflow prediction in basins with limited data and successfully restored missing hydrometeorological data.

  Stream flow prediction utilizing deep learning models in the Lake Abaya-Chamo sub-basin, South Ethiopia · 2026 · DOI

• Future work should focus on im- proving the representation of extreme rainfall events through the integration of additional atmospheric and climatic predic- tors, such as Convective Available Potential Energy (CAPE), Convective Inhibition (CIN), total column water vapor, radar reflectivity, infrared brightness temperature, and regional sea surface temperatures, particularly over the Atlantic Ocean, the Gulf of Guinea, and ENSO-related regions (e.

  Intelligent daily rainfall prediction for early warning using deep learning and satellite data: application to Bouaflé and Zuénoula stations, Ivory coast · 2026 · DOI

• Several limitations deserve acknowledgement. First, all models are trained on data from a single representative grid point per region, which necessarily smooths over sub-regional heterogeneity. Second, the models use same-day meteorological predictors drawn from the NASA POWER reanalysis, which act as effectively perfect predictors. In a true operational setting, predictors would need to come from a numerical weather prediction (NWP) forecast for the lead time of interest, and evaluating model performance when driven by NWP ensemble output is an important next step. Third, climate non-stationarity — particularly the observed intensification of extreme rainfall events in northeast India under ongoing climate change — means that models trained on historical data may underperform as precipitation regimes shift. 301 Future work should explore atmospheric circulation indices, convective instability parameters, and satellite-derived moisture fields as additional predictors. Deep learning architectures may also be worth evaluating, particularly for capturing multi-day persistence structures across stations simultaneously.

  Daily Rainfall Forecasting Across Different Divisions of Assam Using Gradient Boosting with Physically Informed Feature Engineering · 2026 · DOI

• The current method can be improved in future studies by adding satellite observations and remote sensing, which will provide a more thorough spatial representation and enable the detection of exceptional rainfall occurrences. In order to better capture temporal correlations over large distances than recurrent networks, more complex deep learning archi- tectures, including Transformer models and attention, may be investigated. Additionally, the system can be tailored to incorporate real-time early warning in order to mitigate disaster risks in sensitive locations and encourage flood preparedness. Addi- tionally, multi-objective optimization would be useful in achieving a balance between interpretability of the model, computational efficiency, and forecast accuracy, making the framework more practical to use on a broad scale. Author contributions H.F. and S.H. designed and implemented the deep learning framework and performed experiments. M.A.H. contrib- uted to data preprocessing, model benchmarking, and statistical analy- sis. S.S. assisted with theoretical formulation and result interpretation. All authors reviewed and approved the final manuscript. Funding Open access funding provided by NTNU Norwegian Univer- sity of Science and Technology (incl St. Olavs Hospital - Trondheim University Hospital) Data availability No datasets were generated or analysed during the current study.

  A VMD–PSO hybrid framework with machine and deep learning for Aaccurate spatial–temporal daily rainfall occurrence prediction in Pakistan’s diverse climatic regions · 2026 · DOI

• This study proposed a hybrid AVOA-RNN framework for predicting river water quality using the Cauvery River data- set. By integrating advanced imbalance-handling strategies (SMOTE, SMOGN) with African Vulture Optimization for hyper-parameter tuning, the model demonstrated supe- rior predictive capability in both regression and classifica- tion tasks. In classification, the proposed model achieved 97.3% accuracy, 99.52% sensitivity, 97.42% specificity, and a ROC-AUC of 0.97, outperforming comparative approaches such as CNN (93.85%), LSTM (94.92%), RNN (95.73%), GRU (96.38%), WOA-NN (0.82), GA-LSTM (0.80), EMD- (F1 = 0.86), WOA-LSTM (F1 = 0.84), AQPSO-SOFNN Y. et al. Water Science (2026) 40:16 Page 19 of 21 Fig. 11 SHAP-based feature importance (dissolved oxygen (DO ≈ 1.3), BOD (≈ 0.65), pH (≈ 0.45), and TDS (≈ 0.25) dominate, while NO₃, Cl, PO₄, SO₄, turbidity, and FC contribute marginally) Fig. 12 Comparative SHAP analysis for the existing models Y. et al. Water Science (2026) 40:16 Page 20 of 21 and SOFNN-HPS (Accuracy = 0.87). In regression tasks, AVOA-RNN obtained the lowest RMSE (8.3) compared to WOA-NN (9.0), GA-LSTM (9.6), EMD-WOA-LSTM (8.8), AQPSO-SOFNN (8.7), and SOFNN-HPS (8.6), along with a high coefficient of determination ( R2 =0.91). These results confirm that AVOA-RNN not only surpasses tradi- tional and hybrid baselines but also ensures greater ecolog- ical interpretability through SHAP-based feature analysis, highlighting DO, BOD, pH, and TDS as dominant param- eters. The framework thus provides a robust and interpret- able solution for sustainable water quality monitoring and management.

  A new machine learning technique for predicting river water quality using AVOA-RNN · 2026 · DOI

• 1. Cross-river validation: The model can be tested on datasets from other rivers with varying hydrological, climatic, and anthropogenic influences to evaluate its generalizability and robustness across geographies. 2. Integration with real-time sensor networks: Future implementations could integrate real-time IoT sensor data streams to enable continuous monitoring and live forecasting of water quality indices, enhancing responsiveness to pollution events. 3. Policy impact simulation: Coupling the predictive framework with policy simulation models could allow stakeholders to test the environmental impact of regulatory interventions before implementation, thereby promoting more informed decision-making in water governance. Authors’ contributions All the authors are equally contributing to the completion of the manuscript.

  A new machine learning technique for predicting river water quality using AVOA-RNN · 2026 · DOI

• The ANN-RC framework is proposed to enhance physical consistency by explicitly representing sub-seasonal MJO memory; however, the paper does not specify what additional datasets (e.g., upper-level wind fields, sea surface temperature patterns, convective cloud observations) or physical constraints should be incorporated to improve the weak rainfall prediction capability.

  Model Hybrid Artificial Neural Network and Reservoir Computing for Extreme Weather Predicting Based on Madden Julian Oscillation · 2026 · DOI

• While the paper indicates that LSTM-based models require large datasets and high computational costs compared to reservoir computing approaches, the study does not specify the exact dataset size, computational memory requirements, or training time comparisons needed to validate the claimed efficiency advantages of the ANN-RC framework over traditional deep learning methods.

  Model Hybrid Artificial Neural Network and Reservoir Computing for Extreme Weather Predicting Based on Madden Julian Oscillation · 2026 · DOI

• The study focused exclusively on the Aceh region; extension of the ANN-RC hybrid framework to Indonesia's broader and more complex regional climate systems with diverse topography, coastal dynamics, and monsoon patterns requires validation across multiple geographic domains with different climatological characteristics.

  Model Hybrid Artificial Neural Network and Reservoir Computing for Extreme Weather Predicting Based on Madden Julian Oscillation · 2026 · DOI

• The ANN-RC hybrid framework achieved strong temperature prediction (R = 0.99908) but demonstrated significantly weaker performance for rainfall prediction (R = 0.0882), indicating that the current model architecture does not adequately capture the MJO's influence on precipitation in the Aceh region and requires development of more complex models specifically designed for rainfall-MJO relationships.

  Model Hybrid Artificial Neural Network and Reservoir Computing for Extreme Weather Predicting Based on Madden Julian Oscillation · 2026 · DOI

• The study was limited to the 2021 observation period with only nine anomalous days detected, which may not be sufficient for establishing long-term patterns or validating model robustness across multiple years.

  Seasonal Anomaly Detection in the Halda River Using a Multivariate Deep Learning Framework · 2026 · DOI

• The transition toward coordinated multi-physical process constraints represents a shift needed to develop truly physically interpretable hydrological AI models beyond single physical process constraints.

  Integrating multi-task learning with a differentiable physics constrained framework for hydrological forecasting · 2026 · DOI

• The physical constraint currently employed is primarily based on water balance alone, which is insufficient to capture the complex, nonlinear coupling of multiple factors including energy availability and vegetation physiological responses that govern high-ET events.

  Integrating multi-task learning with a differentiable physics constrained framework for hydrological forecasting · 2026 · DOI

• Despite this spike in popularity, the inner workings of ML and DL algorithms are often perceived as opaque, and their relationship to classical data analysis tools remains debated.

  Machine learning and deep learning—A review for ecologists · 2023 · DOI

• The mechanism of transmission and the compartments are not fully elucidated, so models that take into account the regression of the curve of incidence can be a useful tool to predict the epidemic evolution of buruli ulcer (BU).

  ANALYSIS AND FORECASTING OF BURULI ULCER DISEASE WITH ARIMA AND NARNN, A DIDACTIC TOOL APPROACH · 2021 · DOI

• However, interpretable machine learning (ML) models for predicting the TTF of horizontal pressurized storage tanks remain underexplored.

  Explainable Machine Learning for Predicting Time‐to‐Failure of Horizontal Pressurized Storage Tanks in Fire Scenarios · 2026 · DOI

• Conventional deterministic methods for relating seismic response to reservoir properties are limited by the nonlinear and multivariate nature of the seismic–petrophysical relationship, and single-attribute correlations frequently fail to capture the complexity of clastic reservoir systems.

  Machine learning-based seismic attribute analysis for porosity prediction and lithofacies classification in clastic hydrocarbon reservoirs · 2026 · DOI

• foundation and establishes a temporal involving multi-station datasets, higher for resolution time-series observations, multivariate forecasting approaches, and extreme-event-focused rainfall modeling.

  Comparative Analysis of Machine Learning and Time-Series Models for Monthly Rainfall Forecasting: A Case Study of Mumbai · 2026 · DOI

• However, direct GHF measurements remain sparse and unevenly distributed due to drilling limitations and high acquisition costs.

  A physics-guided clustered GBRT ensemble model for geothermal heat flow prediction and its application in China · 2026 · DOI

Most-cited papers in Hydrological Forecasting Using AI

• Deep learning for water quality · Nature Water · 2024 · 259 citations
• Advancing water quality assessment and prediction using machine learning models, coupled with explainable artificial intelligence (XAI) techniques like shapley additive explanations (SHAP) for interpreting the black-box nature · Results in Engineering · 2024 · 166 citations
• Monthly climate prediction using deep convolutional neural network and long short-term memory · Scientific Reports · 2024 · 162 citations
• Evaluation of water quality indexes with novel machine learning and SHapley Additive ExPlanation (SHAP) approaches · Journal of Water Process Engineering · 2024 · 129 citations
• A deep learning interpretable model for river dissolved oxygen multi-step and interval prediction based on multi-source data fusion · Journal of Hydrology · 2024 · 127 citations
• HESS Opinions: Never train a Long Short-Term Memory (LSTM) network on a single basin · Hydrology and earth system sciences · 2024 · 127 citations
• Groundwater level prediction using an improved ELM model integrated with hybrid particle swarm optimisation and grey wolf optimisation · Groundwater for Sustainable Development · 2024 · 121 citations
• Reliable water quality prediction and parametric analysis using explainable AI models · Scientific Reports · 2024 · 119 citations
• A Performance Comparison Study on Climate Prediction in Weifang City Using Different Deep Learning Models · Water · 2024 · 110 citations
• Groundwater Quality Assessment and Irrigation Water Quality Index Prediction Using Machine Learning Algorithms · Water · 2024 · 103 citations

Most recent work

• A comparative analysis of machine learning models for predicting groundwater and surface water in a stressed semi-arid watershed: The Khanmirza case study · Environmental Earth Sciences · 2026
• Modeling CO2 fluxes in coastal wetlands of China using explainable sequence-based deep learning · Ecological Informatics · 2026
• Estimation of Water Quality in Lakes and Rivers Using Remote Sensing and Artificial Intelligence: A Review of Image Processing and Validation Strategies · Limnological Review · 2026
• Correction: Machine learning-based correlation analysis of conventional water quality parameters and composite pollution index in the Luoqing river of the South China Sea Coastal Zone · Frontiers in Marine Science · 2026
• Downscaling of soil moisture in the Wujiang River Basin based on integrated decision tree-based machine learning · European Journal of Remote Sensing · 2026
• Integrating multi-task learning with a differentiable physics constrained framework for hydrological forecasting · Scientific Reports · 2026
• COMPARATIVE ANALYSIS OF PHYSICS-INFORMED AND CONVENTIONAL LSTM AND RNN MODELS FOR TEMPERATURE FORECASTING USING ERA5 REANALYSIS DATA · Journal of Problems in Computer Science and Information Technologies · 2026
• Seasonal Anomaly Detection in the Halda River Using a Multivariate Deep Learning Framework · 2026
• RainSense – A Machine Learning Based Rainfall Prediction System · INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT · 2026
• Satellite-Driven Machine Learning Approaches for Rainfall Forecasting in Support of Smart Irrigation: The Case of Ankara and Samsun · Fırat Üniversitesi Mühendislik Bilimleri Dergisi · 2026

Related topics in Environmental Science