Harnessing AI to Monitor and Optimize Reforestation Efforts in Madagascar

Reforestation is a critical solution to combat climate change and restore degraded ecosystems. However, monitoring the success of reforestation efforts can be challenging, especially over large areas. To address this problem, we partnered with Bôndy to develop a machine learning system that uses satellite and drone imagery to track and quantify the progress of their reforestation program in Madagascar. By leveraging advanced AI algorithms, we aimed to provide valuable insights into the effectiveness of reforestation efforts and support data-driven decision making for environmental conservation.

Why we need Reforestation

The world is facing a climate crisis. The global temperature is rising, the ice caps are melting and the oceans are warming up. The Earth’s natural resources are being depleted and we need to take action now before it’s too late. One of the most immediate things that we can do is reforestation which can help mitigate the climate crisis.

According to the World Resources Institute, the world has lost one-third of its forests since the last ice age, and deforestation continues at an alarming rate of about 10 million hectares per year. Reforestation can help reverse this trend and provide numerous benefits. 

In addition to carbon sequestration, reforestation can also improve biodiversity, prevent soil erosion, and provide economic opportunities for local communities.

Why Reforestation is Challenging

1. Harsh Environmental Conditions: Reforestation efforts often take place in areas with challenging environmental conditions, such as drought, extreme temperatures, or poor soil quality. These factors can hinder the growth and survival of newly planted trees, making it difficult for them to establish themselves in the ecosystem.
2. Lack of Resources: Reforestation projects require significant resources, including funding, labor, and equipment. In many cases, these resources may be limited, making it challenging to implement and maintain large-scale reforestation efforts effectively.
3. Pests and Diseases: Young trees are particularly vulnerable to pests and diseases, which can cause significant damage and mortality. Without proper monitoring and intervention, these threats can quickly spread and compromise the success of the entire reforestation project.
4. Competition with Invasive Species: Invasive plant species can outcompete native tree seedlings for resources, such as water, nutrients, and sunlight. This competition can severely impact the growth and survival of the planted trees, making it crucial to monitor and manage invasive species populations.
5. Human Interference: Reforestation sites may be subject to human interference, such as grazing, logging, or land conversion for agriculture. These activities can damage or destroy young trees, undermining the reforestation efforts. Close monitoring is necessary to identify and prevent such interference, ensuring the long-term success of the project.

The Partner

Our partner Bôndy’s incredible initiative towards mitigating climate change through their reforestation program includes planting thousands of trees in Madagascar (one of the biodiversity hotspots) to create social and environmental impact and help the rural population to have more sustainable revenues. Bôndy’s community-based reforestation program in Madagascar consists of 5 regions, spanning a range of climate zones and land use from agroforestry to mangroves.

The Goal

This Omdena challenge aimed to develop an AI algorithm using satellite and drone imagery to monitor the success of reforestation efforts, focusing on the critical first 5 years of tree growth and survival. By analyzing imagery data, the algorithm would assess the health and progress of newly planted trees, providing insights to ensure the long-term success of the reforestation initiative.

The project brought together AI experts, environmental scientists, and forestry specialists to create a robust and scalable solution that could be applied to reforestation projects worldwide, contributing to the fight against climate change and the preservation of our planet’s ecosystems.

Our Approach

Project Pipeline

Data Collection

The following data sources were used in the project.

Satellite Data Sources

Planet satellite monthly data

Norway’s International Climate & Forests Initiative (NICFI) mosaics contain both monthly and biannual collections. (Biannual collections are generated every 6 months). For this challenge, the team used monthly data.

Sentinel 2 satellite cloud-free L3A data

Level 3A products for Sentinel-2 are monthly, cloudless, surface reflectance syntheses.

Drone Image Sources:

These were the drone images provided by Bondy.

Bondy collected data

Meteorological Data:

ERA-5 data

ERA5-Land is a reanalysis dataset providing a consistent view of the evolution of land variables over several decades at an enhanced resolution compared to ERA5.

Field Data

Bondy collected field information: This is the field information provided by Bondy about their 150 fields and their outlines. The data was provided in the form of KMZ files for 5 regions.

Data Preprocessing and Analysis

The team used Python libraries like Rasterio for satellite data processing and visualization. QGIS with KMLTools was used to clean field data and extract valid reforestation site polygons.

Python’s PIL library extracted drone image metadata, while OpenDroneMap software created orthomosaics, elevation models, and 3D textured models. Ground Sample Distance (GSD) was also obtained, which is essential for estimating tree counts from images.

Applying Super-Resolution to Images

The team experimented with pre-trained super-resolution models, particularly the SRCNN model from Dong et al., using an implementation by WarrenGreen. They began by applying transfer learning, then trained the model on high-resolution drone images from Bôndy. Subsequently, they retrained the model using medium-resolution images and evaluated its performance on lower-resolution ones.

The model accepts 400×400 images, so larger images were cropped, augmented, and used for training. To create lower-resolution training images, they scaled the images down and then back up. After training, the model was converted to the ONNX format to enable faster inference and can be applied as a convolutional filter to transform images into their super-resolution counterparts.

Vegetation Indices

The three vegetation indices namely Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Modified Soil Adjusted Vegetation Index Values (MSAVI2) were calculated. The indices calculation is done using a python library called rasterio.

Time Series for Meteorological Data

Three environmental parameters – temperature, precipitation, and evaporation – were chosen to monitor the impact of the environment on the health of trees in 3 regions of Madagascar. Time series models for each parameter were built using the FaceBook prophet library, with data from ERA5-Land hourly data (1950-present) computed using the python library xarray at the closest grid point to the parcel.

Modeling

UNet++ architecture is used in training. It is an architecture for semantic segmentation based on UNet. The architecture is formerly used in medical images. The reason we chose to use UNet++ in this challenge is that it outperforms the UNet by using connected nested decoder sub-networks enhancing extracted features processing. EfficientNet is used as backbone also known as feature extractor in our model.

Postprocessing

The model outputs raw masks showing pixel-wise tree locations. To get the approximate tree count and vegetation percentage, the outputs must be processed using the ground sampling distance (GSD) in cm and average tree size in cm². While the model can’t detect individual trees, it can detect tree patches and estimate the tree count, albeit with a potentially large error margin. However, this can still provide insights into vegetation changes over time. Calculating vegetation percentage and approximating tree numbers involved basic mathematical operations.

Dashboard Deployment

The main objective of the dashboard is to identify the number of trees in a given region for a given drone image. To detect and estimate the number of trees UNet++ model was integrated into the dashboard. The vegetation indices time series and meteorological parameter forecasting time series were also integrated into the dashboard. The dashboard was dockerized by creating a docker container.

Results and Insights

After experimenting with deep learning architectures, the team selected Unet++ for model training. This robust semantic segmentation model can identify areas with planted trees. While it may not detect every small sapling, it will be useful as the trees mature. The model can help detect and estimate tree counts in a given parcel.

Model performance was evaluated using 4 metrics: IoU (Intersection over Union), Precision, Recall, and Loss.

As we trained the model, the IoU increases over the number of batches/steps on training data. Similar trend is seen on test data.

The loss function tells how good your model is in predictions. If the model predictions are closer to the actual values the Loss will be minimum and if the predictions are far away from the original values the loss value will be maximum. Our model shows a decrease in loss function as we train the model.

As we train the model more, the precision increases with an increase in batch size but for the test, it is a bit bumpy.

Usually, recall should increase during the training. But for some reason from batches 10 to 25 our model has seen a very unusual decrease but we kept on training and the trend started showing up from batches 25 onwards. Although the recall for the test is a bit bumpy, it’s showing an increasing trend. The recall improved over the test set.

Dashboard

From the drone image, the model can predict tree patches, the approximate number of trees, and vegetation health (as a percentage). The model can also predict the best precision, best loss, or best recall. The user can also adjust ground sampling distance, tree size, and confidence threshold.

The vegetation indices page lets users choose from NDVI, NDWI, and MSAVI2 for a particular date and particular region. It also shows the time series data for that index.

 

Meteorological data shows the time series forecasting for temperature, precipitation, and evaporation for 3 regions of Madagascar.

Key Achievements

The partnership between Bôndy and Omdena’s machine learning team yielded several notable achievements in advancing reforestation monitoring capabilities in Madagascar:

1. Developed Machine Learning Model for Reforestation Monitoring: Successfully built and trained a machine learning model to track and quantify reforestation progress in Madagascar using satellite and drone imagery. The model can predict tree patches, estimate tree counts, and assess vegetation health from drone images.
2. Created Interactive Monitoring Dashboard: Developed a user-friendly dashboard that visualizes model predictions, allows customization of parameters like ground sampling distance and confidence thresholds, and displays vegetation indices and meteorological time series data for different regions of interest.
3. Integrated Multiple Data Sources: Leveraged a combination of satellite imagery, drone imagery, and meteorological data to provide a comprehensive view of reforestation efforts over time. This multi-modal approach enables more robust monitoring and analysis.
4. Established Methodology for Scaling Monitoring Efforts: Provided recommendations for collecting valid geometry parcels, automating drone flight paths, establishing ground sampling points, and exploring advanced imaging technologies. These steps lay the groundwork for scaling and optimizing the reforestation monitoring methodology.

Future scope

The achievements of the Bôndy-Omdena partnership have laid a strong foundation for expanding and enhancing reforestation monitoring capabilities in Madagascar and beyond. To build upon this progress, several key areas of focus have been identified for future development:

Potential Applications in Other Industries

1. Precision Agriculture: Adapt the machine learning model to assess crop health, estimate yield, and detect disease or pests, helping farmers optimize resources and improve productivity.
2. Urban Forestry Management: Use the technology to assess urban tree health, identify areas needing maintenance, and quantify environmental benefits like carbon sequestration and air pollution reduction.
3. Wetland and Coastal Restoration: Monitor the progress of restoration projects, such as mangrove reforestation or seagrass bed rehabilitation, by analyzing vegetation establishment and water quality changes.
4. Mine Site Rehabilitation: Utilize the model to monitor revegetation efforts, detect soil erosion or instability, and ensure compliance with environmental regulations.
5. Wildfire Recovery Monitoring: Track natural regeneration, assess the effectiveness of post-fire restoration, and identify areas requiring additional interventions like erosion control.
6. Habitat Conservation and Restoration: Assess the health and extent of critical habitats to inform conservation strategies, prioritize restoration efforts, and evaluate habitat management practices.
7. Carbon Offset Project Verification: Verify the success and environmental benefits of carbon offset projects involving reforestation or afforestation, ensuring the integrity of carbon offset markets.