Vince Vo

How AI and Machine Learning Are Transforming Remote Sensing Data

Rainforests are home to an astonishing array of plant and animal species, many of which are endangered or undiscovered, while playing a crucial role in maintaining the planet’s climate balance. But they are under threat: the tropics lost 4.1m hectares of primary rainforest in 2022 – the area of Switzerland – an increase of 10% on 2021.  Monitoring forest health and the distribution and abundance of species is crucial for rainforest protection. AI and ML techniques have revolutionised species identification: by training algorithms on large datasets of species occurrences and environmental features, AI can accurately detect and classify different species in aerial and satellite images. This technology aids researchers in monitoring the population dynamics, migratory patterns, and overall health of key rainforest species.  In this blog post, we explore the significant impact of AI and ML on the use of remote sensing data for assessing rainforest conditions.

Technology enabling the impact of AI and machine learning on the use of remote sensing data for assessing the condition/health of rainforests primarily includes:

1. Artificial Intelligence (AI): AI algorithms are designed to mimic human intelligence and perform tasks such as image recognition, classification, and predictive modelling. They are used to analyse and interpret remote sensing data, extract relevant information, and identify patterns and changes in rainforest conditions.
2. Machine Learning (ML): ML algorithms allow systems to learn from data and improve their performance over time. They train models with labelled datasets, enabling them to identify and classify species, detect deforestation, and make predictions about future changes in rainforest health.
3. Remote Sensing Technologies: Remote sensing technologies, such as satellite imagery, aerial photography, LiDAR, and radar (each operating at different scales, sensitivities – and with different costs!) provide the necessary data for comprehensive monitoring and analysis of rainforest ecosystems.

Case study: Rainforest Connection Guardians

In the depths of the forest, one cannot help but be taken aback by the overwhelming auditory experience that engulfs the senses. The essence of the forest lies in its vibrant soundscape, where nature communicates in its own unique way. Recognizing the potential of sound as a vital tool for environmental monitoring, Rainforest Connection, a non-profit tech startup based in San Francisco, ventures into West Sumatra. Collaborating with a local group, they aim to establish a network of Rainforest Connection Guardians, deployed atop towering trees, equipped with solar panels and high-powered microphones. These guardians diligently listen to the symphony of the forest, tirelessly identifying specific auditory cues that may indicate potential threats such as the menacing growl of chainsaws, the echoing blast of gunshots, or the rumble of logging trucks and motorcycles. While visual observation may be hindered by distant hilltops, auditory perception allows us to transcend such limitations.

Advancements in ML machine learning coupled with the field of bioacoustics have unveiled a new era before us. Similar to how the microscope exposed a microscopic realm previously unseen, the utilisation of machine learning in sound analysis enables us to comprehend the intricate messages nature conveys, imperceptible to the human ear. By incorporating GPUs into purpose-built devices, we empower these instruments to conduct sophisticated machine learning algorithms on the fly, analysing incoming audio and making real-time decisions regarding data transmission to the cloud. Sound, being a temporal medium, demands an equal amount of time to listen as it does to record. As a result, the arduous task of auditory analysis is impractical for humans, yet machines excel in this domain. Leveraging existing cellular networks, the captured data is efficiently dispatched to the cloud, processed, and subsequently transmitted back to ground personnel.

It is crucial to recognize that safeguarding even a small expanse of a few square kilometres within the rainforest holds a disproportionately significant impact on mitigating climate change. The efforts of these vigilant guardians, patrolling and protecting these select areas, echo far beyond their immediate vicinity. In fact, the environmental benefits achieved are akin to removing thousands of cars from the road. Thus, the responsibility to develop and employ such technological advancements lies with those who understand the importance of preserving our planet’s heartbeat—the interconnectedness of nature through its symphony of sound. It is an untapped resource that, if harnessed effectively, holds the potential to safeguard our living planet for generations to come. https://rfcx.org/

How can AI and ML assist forest protection?

One of the most significant contributions of AI and ML is their ability to process vast amounts of remote sensing data quickly and accurately. AI can identify and extract information from images, such as vegetation density, deforestation patterns, and changes in land cover. ML models can learn from labelled datasets, improving their accuracy and identifying complex patterns that may elude human observers.

Deforestation poses a significant threat to rainforest ecosystems. AI and ML have introduced powerful tools for early detection of deforestation and illegal logging activities. By analysing changes in land cover and identifying patterns of deforestation, AI algorithms can automatically flag areas for immediate attention.

AI and ML techniques are increasingly being employed in predictive modelling for rainforest conservation planning. By integrating remote sensing data with other environmental variables, such as climate data and topographic information, these models can project future changes in rainforest conditions. This helps scientists, project developers and policymakers make informed decisions on land-use planning, prioritising conservation efforts, and mitigating potential risks such as the impacts of climate change, investments and local land practices on rainforest health.

As technology continues to advance, we can expect further innovations in this field, paving the way for more accurate and efficient rainforest assessment and protection.

 
Deforestation Satellite Images

 

While AI and ML offer immense potential for rainforest assessment, there are several pitfalls to be aware of:

1. Data Limitations: The accuracy and reliability of AI and ML models heavily depend on the quality and quantity of training data. Inadequate or biased datasets can lead to inaccurate predictions and assessments. Ensuring representative and diverse training datasets is crucial to mitigate these issues.
2. Interpretability and Transparency: AI and ML models can be complex, making it challenging to understand the reasoning behind their predictions. Ensuring interpretability and transparency of AI-driven assessments is essential to gain trust from stakeholders and make informed decisions.
3. Ethical & Legal Considerations: The use of AI and ML raises ethical concerns, such as data privacy, data ownership, and potential biases in algorithmic decision-making. It is crucial to address these concerns, ensuring responsible data collection, unbiased algorithms, and transparent processes.
4. Technical Expertise: Implementing AI and ML techniques requires specialised knowledge and skills. Lack of expertise among users and stakeholders can hinder the effective utilisation of these technologies. Adequate training and capacity building are necessary to maximise the benefits of AI and ML in rainforest assessment.
5. Integration and Collaboration: To leverage the full potential of AI and ML, effective integration and collaboration among scientists, conservationists, policymakers, and technology experts are essential. Building interdisciplinary teams and fostering cooperation can lead to more comprehensive and impactful rainforest assessments.
6. Environmental Impact of Technology: It is crucial to consider the environmental impact associated with the deployment of AI and ML technologies because models are trained on huge volumes of data. From the energy consumption of computing infrastructure to the carbon footprint of data centres, sustainable practices should be adopted to minimise the environmental footprint of these technologies.
7. While AI and ML technologies hold great promise for revolutionising rainforest assessment using remote sensing data, it is important to navigate potential pitfalls and issues associated with their implementation. Addressing data limitations, ensuring interpretability and transparency, considering ethical implications, fostering technical expertise, promoting integration and collaboration, and embracing sustainable practices will help unlock the full potential of these technologies while safeguarding the integrity of rainforest conservation efforts.

Case study: Utilising AI to Identify Key Action Areas for Combating Deforestation in the Amazon

Brazilian researchers have utilised satellite images and artificial intelligence to demonstrate that the area requiring action against illegal deforestation in the Amazon could be 27.8% smaller than the current strategy implemented by the federal government. The existing plan, known as the Amazon Plan 2021/2022, focuses on 11 monitored municipalities, ignoring new deforestation frontiers outside these areas.

The researchers’ article, published in Conservation Letters, reveals that the high-priority areas with the highest deforestation rates in the Amazon amounted to 414,603 km2 this year, while the plan targeted an area of 574,724 km2 across the 11 municipalities. This indicates that monitoring efforts could be reduced by 160,000 km2, approximately the size of Suriname.

While the deforestation hotspots identified by the researchers accounted for 66% of the average annual deforestation rate, the 11 municipalities in the plan represented only 37% of the deforestation rate between 2019 and 2021.

The researchers emphasised that the current plan fails to cover new deforestation frontiers outside the targeted municipalities.

Guilherme Augusto Verola Mataveli, the corresponding author of the article and a researcher at INPE, stated, “Using this new approach, we concluded that prioritising areas with higher deforestation rates would be more effective than limiting the monitoring to certain municipalities. This is an important finding, given that the agencies responsible for law enforcement in this case, mainly IBAMA and ICMBio, have had their budgets and staffing steadily whittled down. Some of these deforestation hotspots are in the 11 municipalities, but others are in the vicinity and constitute new frontiers.”

The National Council for Legal Amazonia (CNAL), which oversees the Amazon Plan 2021/2022, responded by stating that the plan aimed to focus on areas where illegal environmental activities had the most significant impact on Brazil’s environmental management, while not neglecting other areas in Legal Amazonia.

The researchers highlighted the importance of INPE, which has been producing satellite-based science and technology for 60 years. They emphasised that the advances in data processing, such as the use of artificial intelligence in planning anti-deforestation measures, are crucial for mitigating environmental problems and establishing a sustainable national development plan.

To determine priority areas, the researchers used a machine learning algorithm called Random Forest, which predicted deforestation hotspots for the following year based on multivariate regressions. The method categorised grid cells across the Amazon into high, medium, or low priority classes. The researchers also stressed that their approach captured areas of increasing deforestation not covered by the current plan and did not rely on geopolitical boundaries like municipalities.

Mataveli warned that prioritising only the 11 municipalities would be insufficient for Brazil to achieve its international commitments, including the commitment to reduce illegal deforestation to zero by 2028. The article called for complementary actions to combat deforestation, including environmental education, identifying and holding accountable those involved in illegal deforestation, promoting projects that support the green economy and forest conservation, and regularising public and Indigenous land holdings.

The researchers intend to include the priority areas identified in the study into the Terra Brasilis platform to make the information accessible to state and municipal governments for practical implementation.

 

Sources:
https://rfcx.org/
https://www.planet.com/markets/forestry/
https://agencia.fapesp.br