This case study explores the collaboration between e-Core and a leading pulp producer, who faced a critical challenge: assessing the potential impact of frost on paper production. Frost poses a significant threat, ranging from minor disruptions to severe losses in cellulose output. Given the unpredictable nature of forest asset management, especially under the influence of climate change, our partnership aimed to develop an advanced frost detection mechanism powered by AI, enabling proactive strategies to mitigate potential losses.
Opportunities
The primary objective of this project was to conduct a proof of concept (PoC) to verify the ability to identify unexpected production changes, specifically those caused by frost. We analyzed data from the Monitora and Zeus systems — tools designed for real-time monitoring and data collection in forest operations — focusing on critical areas such as geographic intelligence and forest asset management, utilizing AI to manage the inherent unpredictability of this field.
Project Stages
Our project adhered to the CRISP-DM (Cross-Industry Standard Process for Data Mining) framework, a widely recognized methodology for data mining and machine learning projects. This structured approach guided us through six key phases, each contributing to the successful development of our AI-driven PoC.
- Business Understanding: We thoroughly understood the business objectives and PoC goals. Collaborating with stakeholders, we defined requirements, set key performance indicators (KPIs), and ensured alignment with the operational context.
- Data Understanding: We assessed the available data to understand its quality and relevance. Although Sentinel-2 satellite images were considered, they were ultimately deemed impractical due to complexities and uncertainties. Key variables, particularly meteorological factors like humidity and minimum temperature, were identified as critical for frost prediction.
- Data Preparation: We focused on ensuring the dataset’s quality, selecting the most relevant columns and applying interpolation to create a comprehensive data area. This preparation was crucial for feeding the model with robust inputs.
- Modeling: In this phase, we built the PoC’s analytical foundation. We selected appropriate algorithms, split data into training and testing sets, and continuously optimized performance using accuracy and recall metrics.
- Evaluation: A thorough analysis of the model’s performance was conducted, ensuring it met the business understanding criteria. Collaboration with the client’s experts helped validate predictions and refine the model as needed.
- Implementation/Delivery: The final phase involved implementing and delivering the PoC’s results. We worked closely with the client to validate and refine outcomes, ensuring the model aligned with both business and technical goals.
Project Results
The project led to significant advancements in frost detection for our pulp sector partner through a series of five experiments. Leveraging AI and machine learning models, we developed a highly accurate classifier that reliably distinguishes frost events from other anomalies. The experimentation process, involving several iterations, was vital in optimizing the model’s performance.
Key Results:
- Enhanced Frost Detection Accuracy: The final model achieved approximately 98% accuracy in identifying frost occurrences, making it highly reliable.
- Refined Feature Selection: Meteorological factors, particularly minimum temperature and humidity, were identified as the most crucial variables for predicting frost, streamlining the model for better performance.
- Consistent Historical Data Performance: The model consistently identified frost events when applied to historical data, avoiding false positives during periods without frost.
- Improved Data Handling: Unsupervised learning in Experiment 5 effectively addressed challenges in labeling large datasets, enhancing the model’s robustness.
- Operational Impact: The AI-driven frost detection model enables proactive decision-making, helping the company mitigate potential losses from frost damage, thereby reducing operational risks and improving resource management.
Advanced AI Solutions
Using precise data classification, our team developed an advanced AI model capable of identifying frosts with notable accuracy. One particular experiment stood out, achieving 98% accuracy in tests and consistently validating the model’s effectiveness with previously unclassified historical data. The chosen Random Forest model, known for its efficiency and ease of training, provides a robust solution for precise frost identification.
Projected Benefits
- Early identification of at-risk areas within forest assets
- Optimized resource allocation based on predictive analytics
- Improved accuracy in weather forecasting, enhancing operational decision-making
- Automation of processes leading to increased efficiency and significant cost reductions
- Strengthened technological innovation, reinforcing the company’s competitive market position
Future Perspectives
In conclusion, the Proof of Concept demonstrates the feasibility of using data from Monitora and Zeus systems, combined with AI and machine learning algorithms, to detect changes caused by frost. This approach significantly enhances forest asset management, particularly in a field as unpredictable as forestry under climate change. Looking ahead, there are opportunities to further refine the model by exploring more complex data characteristics, integrating additional relevant information, and expanding its capacity to recognize a broader range of climate anomalies.
