AI for Weather Nowcasting: Opportunities and Challenges

Abstract

Weather nowcasting, defined as forecasting weather phenomena on time scales from minutes to several hours, is critical for mitigating the impacts of high-impact events such as flash floods, severe convective storms, and extreme precipitation. Traditional nowcasting approaches based on radar extrapolation and convection-permitting numerical weather prediction (NWP) exhibit fundamental limitations in representing rapid storm evolution, convective initiation, and localized extremes at short lead times. Recent advances in artificial intelligence (AI) and deep learning have enabled a new generation of nowcasting systems that learn complex spatiotemporal relationships directly from high-resolution radar, satellite, lightning, and NWP data. This paper provides a comprehensive review of AI-based precipitation nowcasting, covering data sources, model architectures, and evaluation methodologies. We discuss deterministic and probabilistic approaches, including convolutional recurrent networks, encoder–decoder convolutional neural networks, transformers, generative adversarial networks, diffusion models, and emerging physics-informed and hybrid AI–NWP systems. Opportunities such as improved short-lead forecast skill, multi-sensor fusion, probabilistic decision support, and enhanced forecast equity are examined alongside key challenges related to data quality, class imbalance, generalization, interpretability, and operational deployment. Finally, we highlight current research frontiers and methodological trends, outlining open challenges and promising directions for future AI-driven nowcasting systems at the PhD level and beyond.

Introduction

Weather nowcasting—forecasting conditions from minutes up to about six hours ahead—is crucial for mitigating the impacts of hazardous events such as flash floods, severe storms, and extreme rainfall. These phenomena evolve rapidly and are difficult to predict accurately with traditional methods. Radar extrapolation performs well only at very short lead times, while numerical weather prediction (NWP) models struggle with precise timing and location of convective events and cannot fully exploit high-frequency observations. As a result, nowcasting is considered one of the most challenging problems in meteorology.

Recent advances in artificial intelligence (AI) and deep learning offer a powerful alternative. AI-based nowcasting models learn complex, non-linear spatiotemporal relationships directly from radar, satellite, lightning, and NWP data. Systems such as DeepMind’s DGMR, MetNet, PredRNN, RainNet, diffusion models, and hybrid physics–AI frameworks have demonstrated significant improvements over traditional extrapolation and, in some cases, over high-resolution NWP for short lead times (0–2 hours), particularly for convective precipitation and extremes.

The text reviews traditional nowcasting methods, key data sources, and a wide range of AI approaches, including ConvLSTMs, encoder–decoder CNNs, transformers, GANs, diffusion models, graph neural networks, and physics-informed hybrids. A major trend is the shift from deterministic forecasts to probabilistic, ensemble-based predictions that better represent uncertainty and support risk-based decision-making. Multi-sensor fusion and hybridization with physical constraints or NWP fields further enhance robustness and extend useful lead times toward 3–6 hours.

Despite strong progress, significant challenges remain. These include data quality issues, class imbalance and poor representation of extreme events, limited generalization across regions and climates, lack of interpretability, high computational costs, and the absence of standardized evaluation benchmarks. Trust, fairness, and operational integration are also key concerns for real-world deployment.

Overall, the text concludes that AI-based nowcasting has clear societal value and strong potential to improve short-term weather prediction. The most promising path forward is a hybrid approach, combining traditional extrapolation, AI nowcasts, and NWP, with human forecasters playing a central role. Future research should focus on physics-informed learning, uncertainty quantification, transfer learning, open benchmarks, and responsible, equitable deployment.

Conclusion

In practice, the most promising operational paradigm is hybrid: use extrapolation for the first tens of minutes, AI nowcasts for 0 –2 (or 0 –3) h, and gradually transition to NWP beyond that, possibly with AI assisting in blending and post -processing. Human forecasters remain central in synthesizing guidance and managing warnings. The emergence of AI nowcasting has generated both excitement and hype. High -profile results from DGMR, MetNet, NowcastNet and others demonstrate real, substantial gains in short -range precipitation skill [1][2][3][5][13]. At the same time, operational uptake is still in early stages. Most national weather services are cautiously experimenting with AI guidance alongside established methods rather than replacing them. AI-based nowcasting has rapidly progressed from proof -of-concept ConvLSTMs to sophisticated hybrid diffusion and transformer systems. These models leverage large archives of radar, satellite and NWP data to produce detailed, frequently updated, and increasingly probabilistic precipitation forecasts. Evidence from multiple studies indicates that, for 0 –2 h convective rainfall, AI systems can deliver higher skill than both optical -flow extrapolation and high -resolution NWP in many settings [2][3][13]. Physics -informed and hybrid architectures extend useful lead times and improve robustness for extremes [1][2][6][8]. At the same time, AI is not a magic bullet. Data limitations, non -stationarity, class imbalance, generalization issues, interpretability challenges and operational constraints remain active research and engineering problems. To fully realize the promise of AI nowcasting, the community should: 1) Invest in high -quality, multi -sensor datasets and open benchmarks. 2) Embrace hybrid physics -AI designs that respect known dynamics. 3) Develop standardized verification practices, especially for probabilistic products and extremes. 4) Focus on interpretability, user training, and human –AI teaming in forecast offices. 5) Prioritize equity, ensuring that advances benefit data -sparse, vulnerable regions as well as data-rich ones [5]. 6) Design governance frameworks that clarify responsibilities and manage model updates transparently. For PhD -level research, open questions include optimal multi -sensor fusion strategies, foundation -model approaches for high -resolution nowcasting, principled physics -ML coupling, robust uncertainty quantification, continual learning under climate change, and human -centric design of decision -support products. Progress on these fronts could yield AI nowcasting systems that are not only more accurate, but also more trustworthy, interpretable and impactful.

References

[1] Zhang, Y., Long, M., Chen, K. et al. Skilful nowcasting of extreme precipitation with NowcastNet. Nature 619, 526–532 (2023). https://doi.org/10.1038/s41586-023-06184-4. [2] Das, P., Posch, A., Barber, N. et al. Hybrid physics-AI outperforms numerical weather prediction for extreme precipitation nowcasting. npj Clim Atmos Sci 7, 282 (2024). https://doi.org/10.1038/s41612-024-00834-8. [3] Ravuri, S., Lenc, K., Willson, M. et al. Skilful precipitation nowcasting using deep generative models of radar. Nature 597, 672–677 (2021). https://doi.org/10.1038/s41586-021-03854-z [4] S. An et al., “Deep learning for precipitation nowcasting: A survey from the perspective of time series forecasting,” arXiv preprint arXiv:2406.04867, 2024. [5] S. Agrawal et al., “An operational deep learning system for satellite -based high -resolution global nowcasting,” arXiv preprint arXiv:2510.13050, 2025. [6] Leinonen, Jussi & Hamann, Ulrich & Nerini, Daniele & Germann, Urs & Franch, Gabriele. (2023). Latent diffusion models for generative precipitation nowcasting with accurate uncertainty quantification. 10.48550/arXiv.2304.12891. [7] S. Zhao, S. Saha, Z. Xiong, N. Boers and X. X. Zhu, \"Exploring Geometric Deep Learning for Precipitation Nowcasting,\" IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium, Pasadena, CA, USA, 2023, pp. 3760-3763, doi: 10.1109/IGARSS52108.2023.10282387. [8] Wang, R., Fung, J. C. H., & Lau, A. K. H. (2024). Skillful precipitation nowcasting using physical-driven diffusion networks. Geophysical Research Letters, 51, e2024GL110832. https://doi.org/10.1029/2024GL110832. [9] L. Bouchard, “DeepMind uses AI to predict more accurate weather forecasts,” Louis Bouchard AI Blog, Oct. 2021. [Online]. Available: https://www.louisbouchard.ai/deepmind-rain-nowcasting/. [10] S. Pulkkinen et al., “Pysteps: an open-source Python library for probabilistic precipitation nowcasting (v1.0),” Geosci. Model Dev., vol. 12, pp. 4185–4219, 2019. doi: 10.5194/gmd-12-4185-2019 [11] Hunt, Kieran. (2025). Stop using root-mean-square error as a precipitation target! 10.48550/arXiv.2509.08369. [12] X. Shi et al., “Convolutional LSTM network: A machine learning approach for precipitation nowcasting,” in Proc. NIPS, 2015, pp. 802 –810. [13] B. Sønderby et al., “MetNet: A neural weather model for precipitation forecasting,” arXiv preprint arXiv:2003.12140, 2020. [14] Y. Wang et al., \"PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning,\" in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 2, pp. 2208-2225, 1 Feb. 2023, doi: 10.1109/TPAMI.2022.3165153. [15] Nguyen, Tung & Brandstetter, Johannes & Kapoor, Ashish & Gupta, Jayesh & Grover, Aditya. (2023). ClimaX: A foundation model for weather and climate. 10.48550/arXiv.2301.10343. [16] L. Espeholt et al., “Deep learning for twelve hour precipitation forecasts,” Nat. Commun., vol. 13, art. no. 5145, 2022, doi: 10.1038/s41467-022-32483-x. [17] Li, Y., Liu, Y., Sun, R. et al. Convective Storm VIL and Lightning Nowcasting Using Satellite and Weather Radar Measurements Based on Multi-Task Learning Models. Adv. Atmos. Sci. 40, 887–899 (2023). https://doi.org/10.1007/s00376-022-2082-6 [18] L. Espeholt et al., “Skillful Twelve Hour Precipitation Forecasts using Large Context Neural Networks,” arXiv preprint arXiv:2111.07470, 2021. [19] L. Lam et al., “GraphCast: Learning skillful medium -range global weather forecasting,” arXiv preprint arXiv:2212.12794, 2022. [20] H. Bi et al., “Pangu -Weather: A 3D high -resolution model for fast and accurate global weather forecast,” arXiv preprint arXiv:2211.02556, 2022. [21] K. A. Browning and C. G. Collier, “Nowcasting of precipitation systems,” Reviews of Geophysics, 27(3), 345–370, 1989. [22] U. Germann and I. Zawadzki, “Scale-dependence of the predictability of precipitation from continental radar images. Part I: Description of the methodology,” Monthly Weather Review, 130(12), 2859–2880, 2002. [23] X. Shi et al., “Deep learning for precipitation nowcasting: A benchmark and a new model,” in Advances in Neural Information Processing Systems 30 (NeurIPS), 2017, pp. 5617–5627. [24] V. Lebedev et al., “Precipitation nowcasting with satellite imagery,” in Proc. 25th ACM SIGKDD Int. Conf. on Knowledge Discovery & Data Mining (KDD), 2019, pp. 2680–2688. [25] G. Ayzel, T. Scheffer, and M. Heistermann, “RainNet v1.0: A convolutional neural network for radar-based precipitation nowcasting,” Geoscientific Model Development, 13(6), 2631–2644, 2020. [26] M. S. Veillette et al., “SEVIR: A Storm Event Imagery Dataset for Deep Learning Applications in Radar and Satellite Meteorology,” in NeurIPS Datasets and Benchmarks Track, 2020. [27] Z. Gao et al., “PreDiff: Precipitation Nowcasting with Latent Diffusion Models,” in Advances in Neural Information Processing Systems (NeurIPS), 2023. [28] X. Bai et al., “Rainformer: Features extraction balanced network for radar-based precipitation nowcasting,” IEEE Geoscience and Remote Sensing Letters, 19, 1–5, 2022

Copyright

Copyright © 2026 Zeyad Faieq Suleiman Abuhuweidi, Ibrahim Khaled Qasem Farea. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.