Satellite Based Cloud Motion Forecasting Using Diffusion Model

Read full text Summarise with VIRAT
Authors: Khushi Gulhane Miss. Niramayee Rotwal, Ayush Wankhede Miss. Radhika Kale, Yogita Dhumale
Cited by: 0
References: 8

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Weather forecasting, particularly short-term cloud motion prediction, is essential for climate monitoring, disaster management, aviation safety, and solar energy optimization. Traditional forecasting methods such as Numerical Weather Prediction (NWP) models and optical flow techniques often struggle with high computational complexity, low resolution, and limited ability to capture non-linear cloud dynamics. To address these challenges, this work proposes a novel data-driven approach using Generative Diffusion Models for predicting future cloud positions from satellite image sequences. The proposed model learns the conditional probability distribution of cloud evolution by utilizing preprocessed geostationary satellite imagery. A combination of Convolutional Neural Networks (CNNs) and a UNet-based architecture is used to encode spatial features, while the diffusion process iteratively refines predictions through a denoising mechanism. This enables the generation of realistic, high-resolution, and temporally consistent cloud forecasts. The system effectively captures complex cloud transformations and provides multiple plausible future scenarios, improving uncertainty representation compared to deterministic models. Performance is evaluated using standard metrics such as Structural Similarity Index (SSIM), Mean Squared Error (MSE), and Cloud Motion Error (CME), demonstrating superior accuracy over conventional approaches like ConvLSTM and GAN-based models. Overall, the proposed approach offers a lightweight, efficient, and scalable solution for short-term cloud forecasting, with potential applications in meteorology, renewable energy forecasting, and real-time weather monitoring systems.

Fields of Study & Subjects

Meteorological Phenomena and Simulations Generative Adversarial Networks and Image Synthesis Advanced Neural Network Applications

Publication details

Journal / source
International Journal of Advanced Research in Science Communication and Technology
Pages
468–468
Publication date
27 March 2026
Language
EN