Jakarta, INTI - The National Research and Innovation Agency (BRIN) held a discussion event featuring a guest lecture, Professor Hironobu Iwabuchi, from Tohoku University, Japan. Held in Bandung on Monday, February 23, 2026, the event discussed cloud and weather monitoring using geostationary meteorological satellites through physics and AI-based retrieval approaches.
The Head of BRIN's Climate and Atmospheric Research Center, Albertus Sulaiman, stated that the use of satellites in predicting weather changes is important to study.
Accurate Weather Monitoring is Needed to Prevent Disasters
"We see the integration of physics and artificial intelligence approaches in satellite data processing as the future direction of atmospheric research. Indonesia, as a maritime and tropical country, really needs fast and accurate weather monitoring technology," said Albertus.
Strengthening satellite data analysis capacity, particularly for monitoring clouds, extreme rain, and water vapor dynamics, is crucial in supporting early warning systems for hydrometeorological disasters.
"This activity provides an important opportunity to expand international collaboration networks and enrich the perspectives of researchers and students in developing innovations in the climate and atmosphere," he said.
The Difference Between AI-based and Physics-based Weather Prediction Technology
Hironobu further explained that the Himawari-8/9 satellite can capture changes in sky conditions and cloud movement. The Himawari satellite does not capture conventional images, but rather uses infrared light. This results in radiation data.
The Himawari satellite generates data such as brightness (albedo), radiant temperature, water vapor absorption, and spatial texture/cloud patterns. Scientists use this data to predict future weather.
Although the predictions are quite accurate, the complex calculations require 100 hours using 10 central processing units (CPUs).
"For a single pixel, this analysis can take 10 milliseconds, and we have to calculate a physics model for each pixel. For a full disk, we have many pixels, around 36 million pixels. This process requires 10 CPUs and is very slow," he explained.
Convolutional Neural Networks (CNN), an AI-based weather prediction technology, use data from the Calipso/Cloudsat satellite as a reference for cloud change patterns to predict the weather. Unlike physics methods that calculate each particle in detail, CNN work by analyzing image patterns of cloud textures and surrounding clouds as supporting data.
"The process required for CNN to provide predictions based on cloud shape or change patterns takes only 40 seconds. Therefore, this represents a significant improvement over previous methods," explained Hironobu.
"Unlike previous methods, which required pixel-by-pixel (PHP) calculations, because CNN rely on cloud patterns and structures, the process time is only 40 seconds," he added.
Conclusion
BRIN’s discussion with Professor Hironobu Iwabuchi emphasized the importance of accurate physics-based and AI-based weather monitoring systems. By using technologies like Himawari-8/9 and faster AI weather prediction technology like CNN, Indonesia can enhance the speed and accuracy of weather predictions, strengthening early warning systems.
Read more: BRIN Develops AI-Based Marine Protein Potential Map with 94.6% Accuracy to Support MBG Program