Nature-based carbon sequestration is one of the most straightforward ways to sequester and to store carbon from the atmosphere. Urban forests hold the promise of optimized carbon storage and temperature reduction in cities. Remote sensing imagery in combination with image processing techniques can identify individual tree location and their sizes, classify trees based on their species, and track tree health. While standard machine learning models like Random Forest and Support Vector Machine can achieve accuracy of 80% in separating trees from other land cover classes, additional deep learning models using noisy labeled data can further improve tree identification maps. We demonstrate such an approach by quantifying the carbon sequestration and urban heat island mitigation for New York City and Dallas.