A promising start to Tanzania's green future with AI
Monitoring reforestation efforts using machine-learning
Two years ago we learned how trees prevent food shortages and drought through a process called Farmer Managed Natural Regeneration (FMNR). It stands to reason that working with nature by practising sustainable farming strengthens ecosystems in harmony with climate resilient communities.
"Cutting trees is forbidden"
Kongwa District in Tanzania
We are working with World Vision Tanzania (WVT) to monitor FMNR efforts in 5 villages in the Kongwa District of Dodoma Region, the country's capital. Drones are used to create high-resolution maps in demonstration plots to monitor the regreening efforts in each village. The idea is to repeat this exercise every 2 years while training what WVT calls the "local cluster" to collect drone data and more importantly as always to visualize it for local analysis.
One promising technique it to use image classification to identify vegetation within the demonstration blocks and monitor changes over time. We first used a photogrammetry software called Pix4Dmapper which introduced this technique in 2017. This type of machine-learning has gotten better with time and we now use Pix4Dfields to rapidly detect vegetation and other features on the ground as depicted below in a village called Makawa. Use the slider to compare the demo plot in 2023 to the baseline in 2021 (lemongrass).
FMNR in Makawa Village between 2021 - 2023
Another technique we are exploring is to generate vegetation indices like the the Visible Atmospherically Resistant Index (VARI) which is designed to emphasize vegetation in the visible portion of the spectrum. Again, we used Pix4Dfields which can rapidly process drone data on-site even at full resolution.
Makawa Village Demo Plot VARI
Clearly the machine-learning technology is better at detecting vegetation than the VARI. However, given limited computing power and funding the VARI is able to offer some insights on vegetation growth in a given area. The next step is to add statistical information like vegetation density and compare different times. Less soil erosion because of vegetation increase should also lead to less gully formation and this can be analyzed from the drone data. All in all we can now use AI to inform stakeholders on land-use and the progress of interventions like FMNR in a given area.
You can support us by sending us constructive feedback on our work. More importantly, you can support vulnerable communities in places like Kongwa District by funding programs like FMNR and by being a responsible steward of the planet.
