FarmSwarm: Network-Based Collective Intelligence for Smart Agriculture

An agent-based model (ABM) built with Mesa that simulates farmer decision-making across Morocco's agro-ecological zones. The simulation compares cooperative (SHARED) and individual (INDIVIDUAL) farming strategies using real-world data from FAO, Copernicus, and NASA POWER.

Research Question

Does cooperative knowledge-sharing between farmers improve crop yields and profitability compared to individual decision-making?

Key Features

• 100 paired agents (50 SHARED + 50 INDIVIDUAL) across 6 agro-ecological zones
• 24 crops with FAO EcoCrop suitability and FAOSTAT pricing
• Real cost model: labor (SMAG × days/ha) + fertilizers (NPK by category) + seeds (by category) + mechanization (by category) + water (zone-specific tariffs × crop needs)
• Climate data: NASA POWER temperature and precipitation per zone
• Soil moisture: Copernicus C3S root-zone soil moisture
• Social network: NetworkX-based knowledge propagation (small-world, random, scale-free)
• Interactive dashboard: Solara web interface

Quick Start

pip install -r requirements.txt

Open simulation.ipynb in Jupyter or VS Code. The main simulation parameters are set in the SIM_CONFIG dataclass (see the first code cell):

from dataclasses import dataclass

@dataclass
class SIM_CONFIG:
	n_agents: int = 100        # Number of agents (farmers)
	n_seasons: int = 30        # Number of seasons to simulate
	shared_strategy: str = "both"  # 'zone', 'neighbor', or 'both'
	use_neighbor_graph: bool = True
	# ... other parameters ...

To change the number of agents, update n_agents in the config cell. For example, to run with 200 agents:

cfg = SIM_CONFIG(n_agents=200)
model, df, hist_df = run_sim(cfg)

Run all cells to execute the simulation and export results.

Optional — launch the interactive dashboard (from the folder containing dashboard_solara.py):

cd AgorAI
solara run dashboard_solara.py

Project Structure

├── simulation.ipynb              # Main simulation notebook
├── dashboard_solara.py           # Solara web dashboard
├── requirements.txt
│
├── data/                         # Input data
│   ├── crops.json
│   ├── zones.json
│   ├── climate_profiles.json
│   └── soil_moisture_profiles.json
│
├── scripts/                      # Data preparation utilities
│   ├── fetch_climate_data.py
│   ├── update_crops_faostat.py
│   └── extract_faostat_prices.py
│
├── Extract_morocco_data.ipynb   # Copernicus NetCDF → soil moisture JSON
├── results/                      # Auto-generated outputs
└── logs/

How It Works

Crop Selection

Each farmer scores viable crops using ecological suitability (zone, soil, climate, moisture) plus optional bonuses:

• SHARED: zone-level yields, neighbor knowledge, own experience
• INDIVIDUAL: own suitability and experience only

Economics

Production costs are calculated per hectare using real Moroccan agricultural data:

• Labor: SMAG minimum wage × crop-specific labor days per hectare
• Fertilizer: NPK costs by crop category (cereals, legumes, vegetables, etc.)
• Seeds: Seed costs by crop category
• Mechanization: Equipment costs by crop category
• Water: Zone-specific tariffs × crop water requirements

SHARED farmers get +20% price premium, 30% cost reduction, and 5% post-harvest loss vs 15% for individuals. Profit = (yield × price − total_cost) × land_size.

Learning

After each season, SHARED farmers update zone-level knowledge and propagate results through the social network. Over 30 seasons, cooperative strategies converge on better crop choices.

Configuration

Parameters are defined in the SIM_CONFIG dataclass inside the notebook. Key settings: n_agents, n_seasons, shared_strategy, use_neighbor_graph, suitability weights, and cooperative economics.

Data Sources

Dataset	Source
Crop suitability & ecological requirements	FAO EcoCrop Database
Crop prices	FAOSTAT Producer Prices (Morocco, 2024)
Labor costs	Morocco SMAG (Minimum Agricultural Wage, Decree 2.23.993)
Fertilizer costs	OCP Group domestic prices
Seed costs	SONACOS Morocco seed market prices
Mechanization costs	Regional agricultural cooperatives survey data
Water tariffs	ORMVA (Offices Régionaux de Mise en Valeur Agricole)
Climate data	NASA POWER temperature and precipitation
Soil moisture	Copernicus C3S root-zone soil moisture
Zones & agricultural data	Morocco agricultural ministry data

References

Framework

• Mesa: Kazil, J., Masad, D., & Crooks, A. (2015). Mesa: An Agent-Based Modeling Framework. SciPy Proceedings. https://doi.org/10.25080/Majora-7b98e3ed-009

Data

• FAO EcoCrop: Ecocrop Database, FAO. http://ecocrop.fao.org/ecocrop/srv/en/home
• NASA POWER: NASA Langley Research Center, Prediction Of Worldwide Energy Resources (POWER) Project. https://power.larc.nasa.gov/
• Copernicus C3S Soil Moisture: Soil moisture gridded data from 1978 to present. Climate Data Store. https://cds.climate.copernicus.eu/datasets/satellite-soil-moisture

Data Preparation

Scripts in scripts/ were used to build the JSON files in data/. Re-run only when updating source data:

python scripts/fetch_climate_data.py --start 1991 --end 2020
python scripts/update_crops_faostat.py

Copernicus soil moisture: see Extract_morocco_data.ipynb.

Troubleshooting

Issue	Solution
ModuleNotFoundError	Run from project root; restart kernel
Dashboard won't start	pip install solara solara-plotly
Missing data	Ensure all .json files exist in data/

License

See LICENSE.