A Land of Opportunities: Tackling Wildfire Land Degradation in Brazil with the G20 Global Land Initiative and Universitat Politècnica de Catalunya · BarcelonaTech (UPC)

Case Introduction

Who are we?

The G20 Global Land Initiative

The G20 Global Land Initiative at the United Nations Convention to Combat Desertification was launched during the Saudi Arabia G20 Presidency, and it has as ambition to achieve a 50 per cent reduction in degraded land worldwide by 2040. To inspire all stakeholders to collectively deliver on land conservation and restoration outcomes: we showcase success stories; engage the private sector; empower civil society and the public; and share knowledge to build capacity among G20 members as well as interested non-member countries and other stakeholders.

The Universitat Politècnica de Catalunya (UPC)

The Universitat Politècnica de Catalunya, also known as BarcelonaTech, is a pioneering institution in the field of engineering and architectural studies. It has always been at the forefront of academic excellence and innovation, offering a range of programs that encourage creative and critical thinking among students and researchers. UPC collaborates with international initiatives like the G20 Global Land Initiative to provide data and expertise that support global sustainability efforts.

The Problem

The challenge we face as the G20 Global Land Initiative is to raise awareness on the importance of halting land degradation and incentivise investment for land restoration. We expect our Hack Teams to pitch their final product to a panel jury representing members of the government of Brazil and provide a comprehensive visualization product/solution that presents the problem and incentivises for solutions.

In a world where wildfires devastate our landscapes, the battle against land degradation is more critical than ever. The G20 Global Land Initiative stands at the forefront of this battle. We've partnered with the BarcelonaTech University, who provided us with valuable data to tackle this issue. Today we want to take you on a journey with us to Brazil. This year, Brazil holds the G20 Presidency and we have a unique opportunity to collaborate with the Brazilian government in addressing one of their biggest causes for land degradation: wildfires. Extensive wildfire in Brazil leading to widespred land degradation, severely impacting various ecosystems and populations.

Brazil is home to six major biomes, each with its own unique features and biodiversity. These biomes cover a whopping 20 percent of the Earth's natural species. The Amazon, known for its tropical forests and meadows, often takes the spotlight. But Brazil's other biomes like the Cerrado, Caatinga, Pantanal, Atlantic Forest, and the Pampa are equally important. The Cerrado's savanna vegetation supports agriculture, while the Caatinga thrives in fruits and medicinal herbs despite being a semi-arid region. The Pantanal, shaped by rivers, is home to rare flora and fauna. The Pampa integrates the herbaceous Araucária species forests, and the Atlantic Forest exhibits varied climates and ecosystems along its coastlines. Despite their richness, these biomes face significant challenges. Land degradation, caused by forest fires, is an urgent matter that needs immediate attention. These wildfires not only destroy precious ecosystems but also exacerbate land degradation, leading to a cascade of environmental impacts. From soil erosion and habitat destruction to invasive plant species, floods, desertification, air pollution, and water scarcity, the consequences are profound and far-reaching. We've gathered datasets and resources to guide you in developing innovative solutions in locating active wildfires and showcase your results. you must raise awareness on the impact of wildfires to the Government of Brazil. Whether you're tech-savvy or not, you'll have the opportunity to contribute and pitch your ideas to the Brazilian Government. Get ready to use cutting-edge technology and diverse skills to revolutionize wildfire detection and land restoration.

What is the problem to be solved? Which outcome is expected as a final product? Who gets addressed by this case?

A. What is the current problem?

• Extensive wildfire in Brazil leading to widespread land degradation, severely impacting various ecosystems and populations.
• The challenge we face as the G20 Global Land Initiative is to raise awareness on the importance of halting land degradation and incentivize investment for land restoration.
• We expect our Hack Teams to pitch their final product to a panel jury representing members of the government of Brazil and provide a comprehensive visualization product/solution that presents the problem and incentivizes for solutions.

B. What is the expected final product?

• Develop an innovative, interactive visualization product to present the scale of the problem of land degradation in Brazil and present data effectively:

  • There are no constraints on the solution type but it has to be technological and interactive such as an app, website, or dashboard.
  • Ensure the inclusion of features that enhance user engagement and understanding.
  • Work on improving the accuracy of your detection model (see Hackathon Challenges section below).

• The aim of this final product is to:

  1. Accurately represent the extent of wildfire degradation in Brazil.
  2. Raise awareness of land degradation issues.
  3. Incentivize investment in prevention and restoration measures.
  4. Showcase the profitability of technology-driven land restoration solutions.
  5. Explore early warning systems and applications for preventing degradation.

C. Who are the users of this solution?

Main user: Government of Brazil Potential sub users: Entrepreneurs, Environmental Agencies, firefighters, General Public, and residents living near the affected areas.

The Pitch:

What do the partners expect from the participants’ presentation?

Format:

• Each team will have a 3-minute pitch, followed by a 3-minute question and answer session in front of the judging panel.
• Teams will showcase their solutions during this session.
• Teams will have access to a TV screen.

Key Elements:

• Innovative Solution Approach: Demonstrating a novel and creative approach to addressing wildfire detection and land degradation in Brazil.
• Effective Integration of Technologies: Utilizing drones, GIS, population statistics, and other relevant technologies seamlessly within the solution.
• Technical Proficiency: Showing mastery in data collection, analysis, and visualization techniques.
• Practicality and Usability: Ensuring the solution is feasible for real-world implementation, user-friendly for stakeholders with varying technical backgrounds, and reproducible in different contexts.
• Integration of Topic and Impact: Deeply understanding and integrating the given challenge, with a focus on addressing wildfire management and broader environmental issues, emphasizing scalability and adaptability.
• Originality, Creativity, and Vision: Presenting an original and innovative concept with the potential to revolutionize wildfire management practices, considering various stakeholders, applications, and future implications.
• Clear and Effective Communication: Clearly and coherently conveying the solution's significance, features, and benefits to the judges, aligning with other evaluation criteria.

Requirements:

• Technical Setup: Ensure compatibility with the TV screen for presenting the solution.
• Time Management: Adhere to the 3-minute pitch and 3-minute Q&A session time limit.
• Visual Presentation: Utilize visual aids effectively to enhance understanding and engagement.
• Accessible Explanation: Present in a manner understandable to judges with varying technical expertise, avoiding jargon.
• Preparedness for Questions: Be ready to answer inquiries regarding technical aspects, feasibility, and potential impact.
• Engaging Delivery: Maintain the judges' attention through organized, structured, and compelling storytelling.
• Alignment with Evaluation Criteria: Ensure the presentation addresses all evaluation criteria, emphasizing innovation, technical proficiency, feasibility, impact, creativity, and alignment with the given problem statement.

Hackathon Challenges:

In order to encourage innovation and skill enhancement among participating teams, the hackathon is structured in 3 phases:

First Task: Wildfire Identification Challenge

The inaugural task is centered around the critical challenge of wildfire identification. Each participating team will be equipped with a foundational deep learning model, specifically the EfficientNetB0, pre-trained to classify instances of forest fires from a curated dataset dedicated to wildfire detection. This model serves not only as a preliminary tool but also as a standard benchmark to ensure equitable comparison across all entries. Teams are encouraged to enhance this model's accuracy and efficiency by innovating on its architecture or methodologies. Alternatively, should teams choose to develop a model from the ground up, it is imperative to utilize EfficientNetB0 as the baseline framework. This approach guarantees a level playing field, facilitating a fair and objective assessment of each team's inventive solutions tailored to this pressing environmental concern.

Stage One: Initial Model Evaluation

• Teams are first tasked with downloading and deploying the provided baseline EfficientNetB0 model to evaluate its performance on the test wildfire dataset.
• Completing this initial evaluation secures a foundational score of 3 points for each team, acknowledging their competence in working with the provided deep learning model.

Stage Two: Model Improvement

• The crux of the challenge lies in the subsequent improvement of the wildfire detection model.
• Teams are encouraged to refine and augment the baseline model, aiming for substantial advancements in its predictive accuracy.
• The enhanced models will undergo rigorous testing against a concealed wildfire dataset from Brazil.
• The percentage of improvement over the baseline model, as gauged on this hidden dataset, will be the yardstick for success in this stage.
• This improvement percentage, when compared against the enhancements achieved by other teams, will be scaled up and reflected in the final score, with the potential to earn up to an additional 5 points.
• The final tally will be derived by multiplying the relative percentage improvement by 5, thus incentivizing breakthrough innovations in wildfire detection technology.

Deadline for Submission: All submissions for Task 1 must be finalized and submitted by no later than 9 PM on the 21st of March via an email to Ismail El madafri [email protected] with the link to the repository.

Very important and Critical: Ensure the final model submission is in the TensorFlow framework

For a full understanding of Task 1 and its requirements, please consult the detailed documentation provided in our Task 1 Guide.

Task 1: The baseline model: https://colab.research.google.com/drive/1butKK0Q1-jT---732aWQ1XYQoD2rRq1Y?authuser=1

Second Task: Use of satellite image in understanding land degradation context in Brazil

The hack teams are presented with 3 additional sets of data. While most of them are easily accesible online through Google Earth Engine, the rests are furnished to afford hackers the flexibility to cross-reference facts about the country and gain a holistic understanding of the wildfire landscape in Brazil. These will be beneficial in the preparation of task number 3 where hack teams’ will build their visualizations products. While the datasets carry significant volume, their exhaustive utilization is not mandatory. These datasets include:

• A MODIS time series satellite imagery, covering Brazil to evaluate extent of burned areas (i.e. number of hectares, spatial distribution of the burned areas…)
• A MODIS landcover dataset to identify the type of land cover within the burn areas and assess the impact of different land cover types on the severity of wildfires.
• 100m gridded population data to assess the impact on the population.
• Country boundary (share file) of Brazil is provided. If your solution requires projecting the layers in the country map. Please make sure to use the country boundary share file provided. This is to make sure that the administrative boundary of Brazil used is verified by the UN.

Deadline for Submission: All submissions for Task 2 will be at the final pitch. Moreover, in the event that GIS software is utilized for addressing this task, kindly download each dataset from our directory. For any doubts on Task 2, please contact Devashree Niraula, [email protected]

Taks 2: The resources and the recommdended library are publicly available through this link: https://colab.research.google.com/drive/1dOgy7HHjpVgh14KflXeSDxP6J_bS9FIh#scrollTo=np5XRuWmVa-P

Third Task: Vizualisation product

The hack teams need to prepare and present the pitch to convince the Government of Brazil representatives on the importance of this issue of land restoration and propose innovative visualization solutions to prevent and halt land degradation caused by wildfires. We envision a tech innovative solution that is interactive and user friendly. Showcasing the data and providing the key highlights of the problem of wildfire in brazil. Examples of products include: app, platforms, website, dashboard.

Technology:

Which APIs, SDKs, software, or hardware components are vital to get the job done? How will you provide the necessary technology?

For the hackathon focused on fire detection technologies, participating teams will be equipped with essential technological resources to facilitate their development process:

• Google Colab Access: Each team will be granted access to "Pay As You Go - Google Colab" for model training, offering a robust platform with scalable computational resources to support deep learning model development and experimentation.

• Deep Learning Model: Teams will be provided with a pre-built, pre-trained deep learning model specifically designed for detecting and classifying forest fires from images. This model can be used as a starting point or as a benchmark.

• Documentation and Tutorials: To further support the teams, resources on how to use Google Colab effectively, as well as best practices for model training and enhancement, will be available.

• Curated Wildfire Dataset: A curated dataset of wildfire images will be provided to train and test the models, ensuring consistency in evaluation and performance measurement across all submissions.

• GIS/RS Tools: Hackers will have a range of tools at their disposal for tackling the case study, including the recommended Google Collab for GIS/RS tasks and the option to use GIS open software like QGIS for visualizing wildfire occurrences and obtaining rapid basic statistical insights.

• Programming Software: The use of programming software such as Python and R will be supported for accessibility to satellite images, with a focus on analysis rather than visualization alone.

By providing these technological resources, the hackathon aims to challenge participants and equip them with the tools and knowledge necessary to create impactful solutions in the realm of fire detection.

Dataset Description and Access for Hackers

Participants in the hackathon have a variety of datasets at their disposal to aid in the development of their solutions.

Wildfire Dataset Access: The Wildfire Dataset is a key resource and is available through two primary avenues:

• Kaggle Page: You can visit the dataset's Kaggle page to explore and download the data manually.
• Kaggle API: For streamlined access and integration with your Google Colab notebooks, the Kaggle API is recommended. Detailed instructions for using the Kaggle API can be found below in the ressources section.

Datasets: The remaining datasets are essential for comprehensive analysis and are available via the following shared drive link: Datasets on Google Drive.

Dataset Descriptions: To understand the structure, content, and best practices for utilizing these datasets, please consult the provided documentation: Datasets Description Document.

We strongly encourage all hackers to review the dataset descriptions carefully to maximize the potential of their proposed solutions and to align their approaches with the data provided.

Deep Dive Slides:

https://docs.google.com/presentation/d/1bvHDJofG3Njb9ElYIGl1iTbQPvUwySHV/edit#slide=id.p1

Further Information:

To learn more about the G20 Global Land Initiative: https://g20land.org

To learn more about the SOC-STEM Research Group: https://socstem.upc.edu/

To learn more about the UPC's Graphic and Design Engineering Department: https://egd.upc.edu/en/research

Resources:

Task 1: The baseline model: https://colab.research.google.com/drive/1butKK0Q1-jT---732aWQ1XYQoD2rRq1Y?authuser=1

For a full understanding of Task 1 and its requirements, please consult the detailed documentation provided in our Task 1 Guide.

Taks 1: Some resources can be found through this link: https://www.tensorflow.org/tutorials/images

Taks 2: The resources and the recommdended library are publicly available through this link: https://colab.research.google.com/drive/1dOgy7HHjpVgh14KflXeSDxP6J_bS9FIh#scrollTo=6iwRG92A1_zq

Judging Criteria:

To ensure a fair and comprehensive evaluation, we have established a detailed set of judgment criteria. These criteria are designed to assess the innovative aspects, technical proficiency, practicality, and overall impact of your solutions.

The evaluation will be conducted across three tasks, each with its specific focus areas such as accuracy of data, methodology, effective utilization of data sources, and integration of various technologies. Points will be awarded based on the quality and effectiveness of your proposed solutions.

For a complete breakdown of the criteria and the points allocated for each task, please refer to our Judgment Criteria Slide.

We encourage all teams to review the criteria thoroughly to align your presentations and submissions accordingly. Your understanding of these criteria will be pivotal in guiding your project development and can significantly impact your overall score.

We wish all teams the best of luck and look forward to your innovative solutions!

Point of Contact:

For data inquiries :

• Ismail El Madafri, Researcher at the Universitat Politècnica de Catalunya · BarcelonaTech (UPC), Spain, [email protected]

• Devashree Niraula, Research and Outreach Specialist,G20 Global Land Initiative, UNCCD, [email protected]

For visualization, end product, hack case context inquiries:

• Paula Padrino Vilela, Programme Officer, G20 Global Land Initiative, UNCCD, [email protected]

• Song Kim, Communications Coordinator, G20 Global Land Initiative, UNCCD, [email protected]

Prize - the winning team members will each receive:

Winning teams will be awarded with a DJI Mini 2 SE Drone per team along with exclusive G20 GLI polo shirts, caps, water bottles, USB drives, notebooks, and magnets adorned with G20 GLI branding.