The BuckAI Observatory maintains a growing collection of open, freely available resources — handbooks for high-performance computing, course materials for AI in the Earth and environmental sciences, and reusable pipelines for geospatial and remote sensing AI. These resources are developed by BuckAI faculty, students, and collaborators across The Ohio State University, and are intended for the broader research and education community.
Practical, self-contained guides covering tools and workflows used across BuckAI research.
A practical handbook for using high-performance computing — including the Ohio Supercomputer Center — for AI and deep learning research. Covers environment setup, GPU workflows, job submission, common pitfalls, and best practices for training large models on satellite and scientific data.
Turn your research code into a citable, installable, peer-reviewed open-source package. A step-by-step walk-through from repo hygiene to a Zenodo DOI (30 min), pip install + Docker (1–2 days), and a peer-reviewed JOSS paper (2–4 weeks). Includes chapters on mixed open/closed training data, AI-assisted development with JOSS disclosure requirements, and a common-pitfalls appendix.
Open lecture notes, Jupyter notebooks, and lab materials from BuckAI faculty and contributors.
A hands-on course covering machine learning fundamentals — from linear regression through convolutional neural networks — applied to geoscience problems. Interactive Python materials use real datasets spanning seismic analysis, geochemistry, and satellite imagery classification.
A 1-credit, self-guided course that prepares graduate students to run research code on real HPC clusters. Covers SSH setup, shell environments, job scheduling with Slurm, environment management, and right-sizing computational resources, culminating in a reproducible capstone project.
A 25-session open-source lecture series across 9 modules from the NASA Cosmic Origins Program AI/ML Science and Technology Interest Group (AI/ML STIG), co-chaired by BuckAI SAB member Yuan-Sen Ting. Covers deep learning frameworks (PyTorch / JAX), neural network architectures (CNNs, GNNs, Transformers, RNNs), physics-inspired and equivariant networks, generative models (normalizing flows, diffusion, flow matching), large language models and autonomous agents, reinforcement learning, and simulation-based inference. Free and open to everyone — no prerequisites or institutional affiliation required.
Reusable AI and geospatial pipelines, model implementations, and analysis tools — open-sourced for the community.
Acquisition, harmonization, and pre-processing of AI-ready multimodal satellite data cubes — tooling to turn raw multi-sensor satellite imagery into analysis-ready inputs for deep learning.
View on GitHub →A 10 m per-pixel archive of canopy disturbance across Sarawak, Malaysian Borneo (2015–2024, Sentinel-2), classifying each event as bare-soil, canopy, or wet-substrate clearing. Captures selective-logging and plantation-maintenance events that fall below the 30 m Landsat threshold used by Hansen Global Forest Change and JRC Tropical Moist Forest. Includes the labeled archive (194 M pixels), reproducible pipeline code, and an interactive browser dashboard. (Paper under review.)
A two-step deep learning pipeline that estimates shallow-water bathymetry from Sentinel-2 multispectral imagery, trained against airborne LiDAR ground truth. Built on a DeepLabV3+ segmentation backbone with cloud masking (Google Earth Engine) and tidal correction (DTU23).
More pipelines are being prepared for release on the BuckAI GitHub organization.