arXiv subjects

Joan Creus-Costa

Publications and source records attributed to Joan Creus-Costa.

3 recordsLinked to original sources

arXiv2026

The Design and Performance of Meteorological Sensors for WindBorne Global Sounding Balloons

WindBorne Systems has developed a constellation of long-duration atmospheric balloons to collect meteorological data across the globe, filling gaps in current in-situ data collection methods. Each Global Sounding Balloon (GSB) is capable of flying for weeks or months and performing dozens of soundings while measuring pressure, temperature, humidity, and GNSS-derived position, altitude, and wind velocity. This data is transmitted to ground via satellite, processed, and made available within minutes of being collected. The current meteorological sensor package has remained largely unchanged since mid-2024 and has flown on thousands of GSBs totaling over one million hours of flight time. Here we present the design and performance of this sensor package. The custom readout architecture and housing allow for data collection across nearly all in-flight conditions while minimizing sources of bias and noise. Uncertainty is characterized via sounding reproducibility studies and in-house calibration of pressure, humidity, and temperature sensors. The calibration and data processing procedures have been optimized and validated by comparison with external datasets. We present external validation in the form of 1) side-by-side radiosonde launches performed in collaboration with the Center for Western Weather and Water Extremes at the Scripps Institution of Oceanography, which show agreement within expected uncertainty limits, and 2) intercomparison studies with European Centre for Medium-Range Weather Forecasts Reanalysis v5, which show an aggregate root mean square difference of: Geopotential height -- 14 m; Pressure -- 0.36 hPa; Temperature -- 0.91 K; Wind speed u -- 2.45 m/s; Wind speed v -- 2.50 m/s; Relative humidity -- 13%.

physics.ao-ph↗

arXiv2025

WeatherMesh-3: Fast and accurate operational global weather forecasting

We present WeatherMesh-3 (WM-3), an operational transformer-based global weather forecasting system that improves the state of the art in both accuracy and computational efficiency. We introduce the following advances: 1) a latent rollout that enables arbitrary-length predictions in latent space without intermediate encoding or decoding; and 2) a modular architecture that flexibly utilizes mixed-horizon processors and encodes multiple real-time analyses to create blended initial conditions. WM-3 generates 14-day global forecasts at 0.25-degree resolution in 12 seconds on a single RTX 4090. This represents a >100,000-fold speedup over traditional NWP approaches while achieving superior accuracy with up to 37.7% improvement in RMSE over operational models, requiring only a single consumer-grade GPU for deployment. We aim for WM-3 to democratize weather forecasting by providing an accessible, lightweight model for operational use while pushing the performance boundaries of machine learning-based weather prediction.

cs.LG↗

arXiv2019

SURREAL-System: Fully-Integrated Stack for Distributed Deep Reinforcement Learning

We present an overview of SURREAL-System, a reproducible, flexible, and scalable framework for distributed reinforcement learning (RL). The framework consists of a stack of four layers: Provisioner, Orchestrator, Protocol, and Algorithms. The Provisioner abstracts away the machine hardware and node pools across different cloud providers. The Orchestrator provides a unified interface for scheduling and deploying distributed algorithms by high-level description, which is capable of deploying to a wide range of hardware from a personal laptop to full-fledged cloud clusters. The Protocol provides network communication primitives optimized for RL. Finally, the SURREAL algorithms, such as Proximal Policy Optimization (PPO) and Evolution Strategies (ES), can easily scale to 1000s of CPU cores and 100s of GPUs. The learning performances of our distributed algorithms establish new state-of-the-art on OpenAI Gym and Robotics Suites tasks.

cs.LG↗