Mitigating climate change and estimate carbon stock of forest from space using artificial intelligence
In the tropics, forest loss has exceeded forest gain, leading to a net greenhouse gas emission that impacts global climate change. One of the simplest natural solutions to address climate change is, therefore, to protect and expand forests. However, stopping deforestation and forest degradation enough to limit warming to 2 degrees Celsius, the threshold beyond which massive and destabilizing climate events will be inevitable, requires biomonitoring of the carbon in remote rainforests, which can be expensive or impossible due to difficulty with access and poverty of indigenous people (which live near and/or manage more than 40% of all remaining protected lands).
In this project, as part of Stanford's Program for Disease Ecology, Health and the Environment, our research team combines remote sensing data (multi-spectral satellite imagery, radar imagery, and LiDAR) with machine learning and computer vision algorithms to establish a fast, automatic, and cost-efficient generalized AI framework to accurately estimate aboveground carbon density (ACD), at fine-grained resolution, in remote tropical forests. The study area to be focused first is forested areas in Borneo Indonesia.
We are developing a robust AI system that would allow transparency in monitoring carbon stocks and deforestation in real-time, which could enhance access to carbon markets for small indigenous communities performing climate mitigation projects that reduce deforestation. This project is open source and still ongoing, we will soon update the code and data used for our fist study site Borneo.
The data for this project are the satellite imagery for Borneo from the LANDSAT and Planet; attributes include 16-bit raw pixel values and labels, as well as LiDAR data for Borneo taken in 2014 by NASA CMS. We are currently working on collecting more remote sensing data, including Aster, ICESAT, and Sentinel imagery.
The scripts are mostly in Python and R.
August 20, 2020