Siamak Mehrkanoon

Published: 1 February 2026

Understanding Weather Nowcasting: Why Minutes Matter

Weather prediction is usually associated with forecasts made hours or days in advance.
However, many high-impact weather events such as heavy rainfall, flash floods, or severe storms evolve within minutes. This is where weather nowcasting becomes essential.

In this blog, I share insights from my research on data-driven weather nowcasting, focusing on how modern deep machine learning methods can improve very short-term predictions.

🌧️ What is nowcasting?

Nowcasting refers to weather prediction over very short lead times, typically from 0 to 2 hours ahead. Unlike traditional numerical weather prediction models, nowcasting relies heavily on:

Radar observations
Satellite imagery
High-frequency measurements

The goal is to capture rapidly evolving atmospheric processes that cannot be resolved well by coarse-resolution models.

⚡ Why is nowcasting difficult?

Short-term weather prediction is challenging because:

Precipitation systems evolve nonlinearly
Small errors grow very quickly
Observations are noisy and incomplete

Even a 5–10 minute shift in storm location can significantly change the impact on the ground.

🤖 The role of machine learning

Recent advances in deep learning have made it possible to model spatiotemporal patterns directly from observational data.

In my research project Weather Elements Nowcasting, I explore architectures that combine:

Spatial feature extraction
Temporal attention mechanisms
Multi-source data fusion

The aim is to better capture storm evolution while maintaining physical consistency.

🔬 What this blog will cover

In future posts, I will discuss:

How radar data is prepared for machine learning
Common pitfalls in precipitation nowcasting
Model architectures and design choices
Interpretation of prediction results
Lessons learned from experiments

Some posts will be technical, others more conceptual but all will focus on explaining ideas clearly.

📌 Final note

This blog complements my project page, where full technical details and results are presented.
Here, the focus is on explanation, interpretation, and discussion.

If you are interested in short-term weather prediction, machine learning, or applied atmospheric science, I hope you will find these posts useful.