PORTFOLIO
Technical Skills
Programming Languages
Python
SQL
C++Web Technologies
HTML5
CSS3Libraries & Frameworks
PyTorch
TensorFlowTools
VS Code
GitFeatured Research
Mapping Urban Heat Islands with Convolutional Neural Networks
The Climate Challenge. As metropolitan areas rapidly expand, concrete and asphalt absorb solar radiation, creating dangerous micro-climates known as Urban Heat Islands (UHIs). This project aimed to predict highly-localized temperature spikes using NASA Landsat satellite imagery and historical weather data, allowing city planners to optimize tree-planting initiatives.
Satellite Data Ingestion
The preliminary phase required processing over 500GB of multispectral satellite imagery. I utilized spatial indexing libraries to align the thermal bands with local municipality zoning maps, establishing a baseline correlation matrix between surface materials and thermal retention.
The Baseline Finding: Simple regression models identified a strong correlation, but struggled to account for the cooling effects of irregular wind patterns through building corridors.
The Baseline Finding: Simple regression models identified a strong correlation, but struggled to account for the cooling effects of irregular wind patterns through building corridors.
Python
Pandas Deep Spatial Feature Extraction
To capture the complex spatial relationships of the urban canopy, I implemented a custom ResNet-based Convolutional Neural Network. By feeding both the thermal imagery and elevation models into the network, the model learned to predict temperature variances down to a 30-meter resolution.
The Result: The deep learning architecture outperformed baseline models by 24%, accurately predicting dangerous heat events 48 hours in advance.
The Result: The deep learning architecture outperformed baseline models by 24%, accurately predicting dangerous heat events 48 hours in advance.
PyTorch