Agent-Based Crop Health Simulation

This project models crop health as a dynamic simulation system instead of a one-time static prediction. Each crop record becomes an agent whose viability changes over time based on support and stress conditions.

• Represents each crop record as an individual crop agent.
• Models growth, risk, and viability as evolving states.
• Balances environmental support factors against stress pressure over time.

The model is based on Agent-Based Modeling, where crop health emerges from interactions between crop agents and environmental, biotic, and management factors.

The project uses Smart Farming Dataset 2024, containing environmental, biotic, water, soil, and management variables used to describe crop conditions.

• Categorical variables were encoded for modeling.
• Continuous variables were normalized to keep feature ranges consistent.
• Raw smart farming measurements were transformed into interpretable simulation inputs.

Raw smart farming variables were transformed into engineered features and aggregated into nine interpretable indicators representing support and stress signals.

• Engineered features include Temperature-Humidity Index, Vapor Pressure Deficit, Growing Degree Days, Soil Moisture Deficit, and nutrient ratios.
• Feature aggregation reduces raw variables into interpretable crop-condition signals.
• The i1-i9 indicators act as the input layer for the simulation model.

The ABM workflow uses a layered structure that maps aggregated indicators into condition states, behavior states, inference outputs, and final crop viability prediction.

The simulation connects growth outlook and immediate risk into crop viability, then updates long-term viability gradually over time.

• Go represents expected growth under current conditions.
• Ir represents immediate risk driven by stress and instability.
• Cv combines growth support and risk pressure into short-term crop viability.
• LCv smooths crop viability over time instead of changing instantly.

The model updates crop viability over time using long-term dynamic adaptation, allowing crop health to evolve gradually instead of changing instantly.

• At each time step, i1-i9 indicators are processed into condition and behavior states.
• Growth Outlook and Immediate Risk are computed from the model layers.
• Crop Viability is calculated from growth outlook balanced against immediate risk.
• Long-term viability updates gradually over the simulation horizon.

The ABM simulation was extended into a NetLogo 3D environment to make crop-agent behavior easier to observe visually. Instead of only reading time-series outputs, the simulation shows crop agents changing health states across a field-like environment.

• Built a 3D agent-based simulation view using NetLogo.
• Represented crop units as agents inside a simulated field environment.
• Used visual state changes to show how crop health evolves over time.
• Connected the simulation concept to the same ABM logic used in the crop viability workflow.

The simulation outputs are interpreted using crop health states and growth-risk relationships, making the model easier to explain than a black-box score.

• High growth and low risk indicate a healthy crop state.
• Low growth and low risk indicate a moderate crop state.
• High growth and high risk indicate a warning state.
• Low growth and high risk indicate a critical state.

The project uses visualizations to inspect crop health patterns, feature relationships, and global system behavior across simulated agents.

• Generated time-series plots for simulation behavior.
• Used feature-to-viability analytics to inspect relationships between indicators and outcomes.
• Visualized global behavior patterns to understand stability and convergence.

The repository is organized into a three-stage notebook workflow covering preprocessing, feature aggregation, and simulation results.

• The workflow separates data preparation, feature aggregation, and modeling into reproducible stages.
• Core model logic is implemented in src/model.py.
• Visual outputs are generated and saved under results/figures/.

The stack combines Python data processing, interpretable simulation logic, agent-based modeling, and Matplotlib visual analytics.

This project emphasizes interpretable simulation and system behavior rather than direct real-world agricultural deployment.

• Model outputs depend on the assumptions and weights defined in the simulation logic.
• The model is predictive in a simulation sense, not a supervised machine learning model trained on real future crop outcomes.
• The system is designed for analytical interpretation, not direct field diagnosis.
• Further validation would require real crop outcome data and expert agronomic review.
• The model simplifies complex crop-environment interactions into structured indicators and update rules.

This project helped me understand how simulation can model crop health as a time-dependent system rather than a static prediction problem.

• Learned how raw environmental variables can be transformed into interpretable indicators.
• Designed a layered model that connects support, stress, risk, and viability.
• Used simulation outputs and visual analytics to explain crop health behavior over time.