Job Description

Model Development:
•    Design and implement a hybrid physics-AI spatiotemporal modeling framework for translating satellite-derived Land Surface Temperature (LST) data into high-resolution ambient air temperature maps
•    Develop and train Graph Neural Network (GNN) and Long Short-Term Memory (LSTM) architectures to model complex spatial and temporal dependencies in urban microclimate data, incorporating attention mechanisms, continual learning (e.g., Elastic Weight Consolidation), and feature attribution methods (SHAP, Integrated Gradients).

Data Collection and Validation:
•    Design and execute field validation campaigns across diverse HDB precincts and test-bed sites (NUS campus, SIT Punggol Digital District), coordinating sensor deployment, drone-based thermal measurements, and mobile sensing data from SBS Transit bus networks.
•    Process and integrate high-resolution satellite LST imagery with multi-source urban morphology and sensor data to construct spatiotemporal graphs for model training and inference.

Model Interpretability & Optimisation:
•    Implement and evaluate feature attribution methods (gradient-based saliency, attention analysis, SHAP) to quantify the contribution of urban input features to model predictions, enabling cost-efficient recalibration and targeted data collection strategies.
•    Develop reproducible, modular processing pipelines for integrating multi-modal inputs including satellite LST (Landsat, MODIS, HotSat-1), LiDAR-derived urban morphology features (Sky View Factor, Green View Index, building density), and ground-based sensor time series.

Research & Dissemination:
•    Publish research findings in high-impact peer-reviewed journals and present at international conferences.
•    Contribute to technical reports, tool documentation, user manuals, and outreach materials for government agency partners (HDB, NEA, URA, NParks) to support downstream adoption and policy application.

Tool Development & Stakeholder Engagement:
•    Package the validated model as a deployable Python module and/or QGIS/ArcGIS plugin compatible with HDB’s Integrated Environmental Modeller (IEM); coordinate with project partners (HDB, SBS Transit, A*STAR I2R, SIT) and participate in meetings, workshops, and knowledge-sharing sessions.

Job Requirements

Essential:
•    PhD in Computer Science, Urban Building Science, Environmental Science, Atmospheric Science, Urban Studies, Mechanical Engineering, or a related field with strong quantitative and computational components.
•    Demonstrated experience with Graph Neural Networks (GNNs), spatiotemporal deep learning, or related machine learning methods, with evidence of application to real-world spatial or environmental datasets.
•    Strong Python programming skills, including proficiency with deep learning frameworks (PyTorch or TensorFlow), graph learning libraries (e.g., PyTorch Geometric, DGL), and geospatial analysis tools (e.g., GDAL, rasterio, geopandas).
•    Experience with life-cycle carbon assessment methods and tools.
•    Demonstrated track record of research excellence through peer-reviewed publications and/or conference presentations in machine learning, remote sensing, urban climate, or closely related areas.
•    Experience with remote sensing data processing and geospatial analysis, including handling of satellite imagery products (e.g., Landsat, MODIS, Sentinel) and derived urban morphology features (Sky View Factor, NDVI, LULC classification).
•    Familiarity with urban microclimate dynamics, urban heat island (UHI) effects, and thermal comfort indices (e.g., PET); experience with microclimate simulation tools (e.g., ENVI-met) is advantageous but not required.
•    Ability to work independently and collaboratively within a large multi-institutional team (NUS, A*STAR I2R, SIT), and to communicate technical findings effectively to government agency partners including HDB and SBS Transit.