Multimodal Physics‑Informed AI for Air Quality

The Methodist Church in Britain; London Borough of Southwark; On-site; GBP 45,000 - 65,000; Full time

Job summary

The University of Surrey invites applications for a Research Fellow position within the FMAI4Air project. The role focuses on designing and pre-training a physics-informed, multimodal foundation model to predict air pollution and related health effects using satellite, sensor, meteorological and emissions data.

You will embed physical constraints, explore super-resolution techniques, and adapt models for wildfire early-warning and urban mitigation.

Qualifications

• Designing and training large-scale DL architectures (Transformers, Mamba or similar)
• Multimodal data fusion across time-series, imagery and tabular data
• Developing and deploying web applications to operationalise models
• Track record of high-quality research in AI or environmental science venues

Responsibilities

• Design and pre-train a multimodal, physics-informed foundation model for air quality prediction
• Fuse satellite imagery, sensor, meteorological and emissions data
• Evaluate health impacts of air pollution
• Embed physics constraints into neural network loss functions
• Apply super-resolution and downscaling to data for street-level insights
• Pretrain and fine-tune models on HPC infrastructure
• Collaborate with academic, industry and government partners
• Publish research in top-tier venues and support dissemination activities

Skills

Transformers

Multimodal data fusion

PyTorch

TensorFlow

Web app development

HPC deployment

Physics-informed neural nets

Calculus

Linear algebra

Research dissemination

Tools

HPC clusters

AI@Surrey

Isambard-AI

Job description

The University of Surrey invites applications for a Research Fellow position within the FMAI4Air project. The role focuses on designing and pre-training a physics-informed, multimodal foundation model to predict air pollution and related health effects using satellite, sensor, meteorological and emissions data.

You will embed physical constraints, explore super-resolution techniques, and adapt models for wildfire early-warning and urban mitigation.