Gully Erosion Monitoring & Control System

Inspiration

My inspiration came from a pressing, real-world environmental issue. I witnessed how soil erosion, particularly gully formation, devastates agricultural land and rural livelihoods. Traditional monitoring methods are slow, labor-intensive, and struggle to cover large areas effectively. I was driven by a question: Could we leverage modern technology to give land managers "eyes in the sky" and an "AI brain" to combat this problem faster and smarter? The goal was to transform a reactive, manual process into a proactive, intelligent system for land restoration.

What it does

This system automates the entire workflow of gully erosion management. It first uses satellite imagery and AI to rapidly screen large areas, identifying and mapping erosion gullies. Then, it zooms in with drone data to precisely measure each gully's depth, volume, and slope. Finally, an AI decision engine analyzes all the data to generate actionable repair reports with prioritized actions and specific treatment recommendations—all through a simple, bilingual interface.

How we built it

We built it as a modular three-tier pipeline: Screening Engine: A backend service using Python, GDAL, and a PyTorch-based deep learning model processes satellite imagery for large-scale gully detection. Measurement Core: This module ingests high-resolution drone imagery (RGB and DSM) to perform pixel-level segmentation and 3D morphological analysis. Decision & Interface: A rule-based AI model synthesizes geodata into structured reports. The entire pipeline is served through a containerized web application with a React frontend, providing maps, charts, and reports.

Challenges we ran into

Data Heterogeneity: Creating a robust AI model that performed consistently across different satellite sources, seasonal landscapes, and lighting conditions was difficult. Precision at Scale: Balancing processing speed for continental-scale analysis with the accuracy needed for engineering-grade measurements required significant algorithm optimization. From Pixels to Plans: The biggest hurdle was translating numerical results (e.g., cubic meters of soil loss) into practical, credible conservation advice. This required deep collaboration with soil science experts. System Integration: Seamlessly weaving together diverse components—geospatial processing, AI models, and a web app—into a single, stable product was a major software engineering effort.

Accomplishments that we're proud of

Creating a Complete Loop: We successfully built an integrated "Scan-Measure-Decide" system that closes the loop from detection to solution, which is rare in academic or siloed commercial tools. • Delivering Practical Utility: The system dramatically reduces survey time—from months to days—and provides land managers with clear, data-driven steps to take, moving beyond just analysis. • Design for the User: We're proud of the intuitive, bilingual interface that makes advanced geospatial AI accessible to non-technical stakeholders in the field of land conservation.

What we learned

Interdisciplinary is Key: Solving real-world problems requires merging expertise: remote sensing, software engineering, machine learning, and domain knowledge (soil science). • AI is a Tool, Not a Magic Bullet: The AI's suggestions are only as good as the expert rules and training data behind it. Close collaboration with end-users is essential for validation and trust. • The Devil is in the Data Pipeline: We spent most of our time not on model architecture, but on building robust pipelines for data cleaning, formatting, and processing.

What's next for Gully Erosion Repair Agent

Predictive Power: Integrate weather and climate data to predict erosion risk and simulate the impact of different repair strategies before implementation. • Expanding the Mission: Adapt the core technology to monitor other land degradation issues, such as landslides, desertification, or deforestation. • From Assistant to Autonomy: Explore integration with autonomous machinery (e.g., drones for seeding, smart tractors) to begin closing the loop from digital recommendation to physical intervention in the field. • Building a Community Platform: Develop a shared portal where agencies and researchers can upload data, benchmark models, and collaborate on regional restoration projects.