Learnijoy
ResearchAI ResearchAI Engineering & DevTools

GenONet Improves High-Resolution Precipitation Nowcasting

Mohammad Kian Golkar, Luciano Alves de Oliveira, Mohammad Khanjani· September 2, 2026 View original

Key takeaways

  • GenONet uses a DeepONet within a GAN for improved precipitation nowcasting.
  • It produces sharper, more physically consistent forecasts over longer horizons.
  • Physics-informed loss and adversarial training enhance model performance.
  • The model shows significant improvements for high-intensity events.

Who benefits

MeteorologyAgricultureDisaster ManagementUrban PlanningLogistics

Summary

GenONet, a novel Spatio-Temporal U-DeepONet, uses a Deep Operator Network as a GAN generator for high-resolution precipitation nowcasting up to 3 hours. It produces sharp, physically consistent forecasts by learning continuous-time dynamics and incorporating a physics-informed loss.

Researchers have introduced GenONet, a new architecture designed for high-resolution precipitation nowcasting, capable of forecasting up to three hours ahead. This model addresses the common issue of blurry forecasts and deteriorating skill over longer horizons in deep learning models by integrating a Deep Operator Network (DeepONet) as a generator within a Generative Adversarial Network (GAN) framework. GenONet's innovative design allows it to learn the continuous-time dynamics of precipitation, ensuring stability and physical consistency in its predictions. Adversarial training, combined with a physics-informed loss regularizer derived from the Moisture Conservation Equation, compels the model to generate sharp, coherent forecasts. Quantitative and qualitative evaluations demonstrate that GenONet outperforms baseline models, especially for high-intensity events and extended lead times, maintaining structural integrity where others degrade.

Why it matters

Accurate, high-resolution precipitation nowcasting is crucial for disaster preparedness, urban planning, and agriculture, enabling better decision-making and impact mitigation for severe weather events.

How to implement this in your domain

  1. 1Evaluate GenONet's potential for integration into existing weather forecasting systems.
  2. 2Pilot GenONet in specific regions prone to severe weather for real-time nowcasting.
  3. 3Collaborate with meteorological experts to fine-tune physics-informed loss functions for local conditions.
  4. 4Develop APIs or interfaces to disseminate GenONet's high-resolution forecasts to relevant stakeholders.

Original post by Mohammad Kian Golkar, Luciano Alves de Oliveira, Mohammad Khanjani

"arXiv:2609.00544v1 Announce Type: new Abstract: High-resolution precipitation nowcasting is critical for reducing the impacts of severe weather but remains difficult because of rapid storm evolution. Deep learning models have shown great promise for this task, but their predictiv…"

View on X

Originally posted by Mohammad Kian Golkar, Luciano Alves de Oliveira, Mohammad Khanjani on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI Engineering & DevToolsAI Research

Subspace Levenberg-Marquardt Algorithms Boost Neural Network Training

This research evaluates subspace Levenberg-Marquardt (LM) algorithms, such as KSLM and HSLM, for training neural networks on regression and classification tasks. These methods address the high computational and memory costs of classical LM, offering more efficient second-order optimization compared to first-order methods like SGD and Adam.

M. Duc HoangSep 2, 2026
AI ResearchAI Engineering & DevTools

Neural Networks Show Varied Conceptual Separation Internally

A study examined "conceptual separation" in CNNs and LLMs, analyzing how internal activations represent concepts. It found that CNNs form coherent representations for familiar concepts, while LLMs show clear separation for distinct domains but collapse distinctions for ambiguous topics.

Jaee Ponde, Roshni Agarwal, Subhashis BanerjeeSep 2, 2026
AI Engineering & DevToolsAI Research

FCCA Boosts Low-Resource LLM Adaptation with Fewer Parameters

Researchers introduced Frozen-Core Cross-Covariance Adaptation (FCCA), a new parameter-efficient fine-tuning method that significantly improves LLM performance with a minimal trainable parameter budget. FCCA achieves results comparable to LoRA and DoRA while optimizing orders of magnitude fewer parameters by carefully selecting fixed core coordinates.

Wentao Ye, Zhanming Shen, Zhiqing Xiao, Yao Ding, Haobo Wang, Gang ChenSep 2, 2026