Mangrove Protection in Tanzania – An Omdena Approach

The impact of deforestation in Tanzania serves as a powerful reminder of the urgent need for action against climate change. Despite boasting Africa’s third-largest forest cover, the country is losing its trees at an alarming rate, ranking as the fifth-highest in deforestation globally.

This widespread and uncontrolled cutting down of forests harms Tanzania’s environment and fuels climate change, which impacts us all. However, there’s a beacon of hope – Nature-Based Solutions.

With local partners, Omdena worked on two projects in the region, tackling deforestation head-on, but with a unique twist: community-driven impact. By leveraging cutting-edge AI for deforestation analysis, the project aims to empower local communities to become active participants in protecting their environment.

In this article, we will delve into this innovative approach and discover how Tanzania can harness the power of AI and community collaboration to unlock a sustainable future, driven by positive environmental, social and technology impact.

Deforestation in the Rufiji Delta

The Rufiji River delta, the largest tidal mangrove wetland on Africa’s eastern coast, spans an impressive 54,500 hectares of protected Mangrove-Rufiji Forest Reserves. This dynamic landscape, characterized by coastal swamps, narrow creeks, and scattered mangrove thickets, presents a formidable challenge for comprehensive monitoring through field visits alone. However, by harnessing the power of Earth Observation (EO) and Artificial Intelligence (AI), local communities can now receive timely warnings about unforeseen events occurring in the region, enabling them to respond effectively to potential threats and changes in this unique ecosystem.

Omdena Partners with the UK FCDO in Tanzania

Omdena is collaborating with the UK Foreign Commonwealth and Development Office (FCDO) and local partners in Tanzania to upskill local AI engineers and develop an AI solution for mangrove conservation in the Rufiji delta. This partnership focuses on AI education, creating innovative AI solutions for societal challenges, and deploying these solutions in real-world scenarios. Over six months, the initiative will enhance the AI capabilities of hundreds of local engineers, contribute to job creation in the sector, and support local entrepreneurs and startups in developing their AI solutions. For more details, visit Omdena’s website.

Project Objectives

Omdena projects are always ambitious, and the Tanzanian deforestation monitoring project is not different, both parts had the same high level objectives:

Omdena Project Structure

Omdena recently completed two projects focused on the Rufiji Delta. The first was an Innovation Challenge where 50 machine learning engineers explored different options to monitor canopy changes from space and predict the type of activity associated with the observed changes. The second project was a Talent Project, where a small team conducted a deep dive on one of the challenge’s findings and built it out into a minimum viable product (MVP) or web application.

The Innovation Challenge’s main finding was that the Normalized Difference Vegetation Index (NDVI), using either the Sentinel-2 satellite constellation or the PlanetScope data from the NICFI program, was a good indicator of canopy disturbance. However, the lack of access to ground truth data limited the team’s ability to build an end-to-end application. The Talent project took these findings and modified the input data to include optical images from Sentinel-2, Landsat 7 & 8, and Synthetic-aperture Radar (SAR) data from Sentinel-1 to create a more stable data pipeline, with Google Earth Engine as our cloud computing solution. Cloud cover is a significant problem in the region, and by combining three data sources, more valid data points are included in the analysis.

Protecting Mangroves Forest in Tanzania’s Rufiji Delta

Deforestation and Land Degradation are important factors in the Rufiji Delta, besides being an unique ecosystem (with seven genera of mangrove trees) the nature reserve also functions as a natural carbon sink and a nature based protection zone for tsunamis and heavy winds and high seas. The two projects are aiming to provide the local communities a practical tool to monitor this system by combining Earth Observation and Artificial Intelligence.

Local, Data-Driven, Forest Management

The Innovation Challenge aimed to create an AI system for detecting deforestation in Tanzania’s Rufiji Delta. However, limitations in reliable deforestation data and imbalanced forested and deforested areas hindered the development of a robust Deep Learning model. Pre-trained models from other forests also proved unsuccessful.

The project pivoted to using the well-established Normalised Difference Vegetation Index (NDVI), which correlates with vegetation health and can be computed from satellite data. This approach was integrated into a web application, allowing users to identify deforested areas and mark locations for conservation and restoration efforts.

The proposed solution is scalable across Tanzania due to its lower resource requirements compared to other AI methods. Scaling country-wide would involve expanding the region of interest, acquiring cloud-free satellite imagery from two time points, and redesigning the architecture to manage multiple projects. This country-wide application could empower local communities to monitor and manage Tanzania’s forests, ultimately preventing further deforestation.

Refining the results

The Talent project aimed to provide Tanzania’s National Carbon Monitoring Centre (NCMC) with a new data layer – a “Near-Real-Time” canopy disturbance indicator. This indicator would issue weekly or bi-weekly warnings when a contiguous area of the canopy was disturbed.

The project leveraged global datasets from the Global Forest Watch (GFW) and the Global Mangrove Watch (GMW) to validate the model results, in the absence of ground truth data. By combining these three datasets and considering local conditions, the resulting layer showed good agreement with the GFW/GMW canopy disturbance indicators.

The indicator’s concept was based on analyzing the greenness of the canopy (from optical images) and the forest’s structure (from radar reflectance) over a known forest area (using Hanzen canopy data). Each 30x30m pixel was assigned a likelihood value for disturbance by comparing the current signals to measurements from the same location 36 months prior. Deforestation activity manifests in two ways:

1. A drop in canopy greenness (from vegetation to bare soil)
2. Removal of the typical radar back-bounce (forest structure)

When a contiguous area of 10 cells is affected by the combination of these two changes, a warning is issued.

These warnings could help the NCMC optimize the deployment of their limited number of field crews, as the Rufiji Delta is a vast area to monitor.

Reaching the Communities

The two projects operated independently, but since the Talent Project began after the Innovation Challenge with the same objectives, it could focus directly on the data most beneficial for generating insights and key process indicators to successfully identify deforestation or abrupt canopy change. Both projects had the same eight-week sprint time, but the Talent Project’s much smaller team would have required more time to test different models and select the most valuable datasets.

Another similarity between the projects was the lack of ground truth data, which is crucial because both projects developed web-based applications as Minimal Viable Products (MVPs). Although both versions could have been “live,” the teams stopped short of creating public/open applications due to the inability to test and adequately validate the methods. In the project’s next phase, it would be beneficial to emphasize the end-user more and better understand local needs.

Presently, Omdena is establishing a Local Chapter to democratize AI usage and is considering relaunching the Deforestation Project in the Rufiji Delta.

Lessons learned

When starting an ambitious project the outcomes are unpredictable, and both projects learn some valuable lessons and insights for future projects:

Demo Application

To showcase the capabilities of the AI-based canopy disturbance monitoring system, the teams developed a web-based demonstration application. This application consolidates the results of the Nation Carbon Monitoring Centre’s efforts to track deforestation in Tanzania’s mangrove forests.

The demo application features an interactive map of the Rufiji Delta, displaying the locations of detected canopy disturbances. Users can zoom in on specific areas of interest and view detailed information about each disturbance event. The app also allows you to apply a date filter to see the amount of disturbances during a set period.

Zooming in allows the user to see the extent of disturbance caused in a single event, and the severity of the event.

Looking ahead, the application is planned to include an automated alerts system that will notify authorities via email when a canopy disturbance event exceeding a certain threshold is detected. This feature aims to enable rapid response to significant deforestation incidents, allowing local authorities to intervene promptly and minimize further damage to the mangrove ecosystem. By leveraging AI to continuously monitor the Rufiji Delta and trigger alerts when necessary, this system has the potential to greatly enhance the effectiveness of conservation efforts in the region.

Conclusion

The Omdena approach is a unique combination of working with a truly global community of Machine Learning engineers who are all driven by the intention to do good, and this creates a project ethos of equality, curiosity, and eagerness to learn and experiment.

Key Achievements

Appropriate Technology Selection

Both projects succeeded in selecting the appropriate AI technology for their level of data engagement. The Innovation Challenge used a regression model to fit the multispectral information to a useful index, while the Talent Project built upon this learning to create a more sophisticated model based on the same principles.

Piloting AI-based Systems

Both teams successfully piloted AI-based systems to test their effectiveness in analyzing and monitoring deforestation in the Rufiji Delta area. Although the developed web-based applications are not yet “live,” their combined knowledge base sets the next project up for success.

Establishing a Foundation for Future Work

While the projects faced challenges in meeting certain objectives, such as providing near real-time data to Tanzania’s National Carbon Monitoring Centre, the teams’ efforts have laid the groundwork for future projects to build upon and refine the AI-based deforestation monitoring systems.

Developing a Demo Application

The teams created a user-friendly web app to make the AI-based canopy disturbance monitoring system more accessible. The interactive platform displays deforestation data on a map of the Rufiji Delta, showcasing AI’s potential in monitoring environmental changes and raising awareness about preserving Tanzania’s mangrove forests.

Lagging in AI and digital transformation? Omdena accelerates your journey with ethical, custom AI solutions built by a global network of experts. Solve real problems, gain insights, and future-proof your business. Unlock your potential with Omdena today.

Want to work with us too?

Let’s see if we are a good fit