AgriTwin-GH

Indoor Greenhouse Dataset Generation

Overview

This document describes the methodology and process for generating synthetic indoor greenhouse datasets from outdoor weather data for the AgriTwin-GH project. The datasets simulate passive greenhouse conditions without active environmental control systems, providing realistic baseline data for greenhouse monitoring and digital twin applications.

Generated Datasets

Two hourly indoor greenhouse datasets have been created:

Dataset Location Records Period Size
dindigul_greenhouse_indoor_2024.csv data/processed/ 8,784 Jan 1 - Dec 31, 2024 1.03 MB
dindigul_greenhouse_indoor_2025.csv data/processed/ 8,760 Jan 1 - Dec 31, 2025 1.02 MB

Both datasets contain 24 hourly records per day with no missing values.

Source Data

Input Files

Source Data Structure

The original weather files contain daily outdoor measurements with the following key columns:

Transformation Methodology

Step 1: Data Loading and Standardization

Raw daily weather data is loaded and column names are standardized:

Step 2: Temporal Upsampling to Hourly Resolution

Daily records are expanded to hourly resolution (24 records/day):

  1. Complete Hourly Index Creation: A full hourly datetime index is generated spanning each entire year
  2. Daily Value Distribution: Each daily measurement is distributed across 24 hourly records
  3. Diurnal Variation Addition:
    • Temperature: Sinusoidal pattern with ±3°C variation, peak at 14:00, minimum at 05:00
    • Humidity: Inverse pattern with ±5% variation (clamped 20-100%)
    • Solar Radiation: Distributed using sine pattern across daylight hours (06:00-18:00), zero at night

Mathematical formula for diurnal temperature variation:

temp_variation = 3 × sin((hour - 5) × π / 12)
hourly_temp = daily_temp + temp_variation

Step 3: Day/Night Classification

A binary day_night_flag is computed using actual sunrise/sunset times:

Fallback heuristic (if sunrise/sunset unavailable): Day = 06:00–18:00

Passive Greenhouse Physics Model

The indoor conditions are derived using a passive greenhouse model that simulates natural greenhouse behavior without active climate control.

A) Indoor Temperature (°C)

Indoor temperature accounts for solar heating during day and modest heat retention at night:

Day (solar heating effect):

ΔT = 0.02 × solar_radiation
indoor_temp = outdoor_temp + ΔT

Night (heat retention):

ΔT = 1.5°C
indoor_temp = outdoor_temp + 1.5

Rationale: Greenhouse glazing traps solar radiation during day (proportional to solar intensity). At night, structural thermal mass provides modest temperature elevation above outdoor conditions.

B) Indoor Humidity (%)

Humidity is affected by temperature-driven evapotranspiration and condensation:

Day (drying effect):

indoor_humidity = outdoor_humidity - (0.5 × ΔT) + 0.4

Night (moisture accumulation):

indoor_humidity = outdoor_humidity + 5

Constraints: Clamped between 30% and 100%

Rationale: Daytime heating reduces relative humidity through evapotranspiration. Nighttime cooling increases relative humidity as water vapor condenses on cooler surfaces.

C) Indoor Air Velocity (m/s)

Indoor air movement is reduced compared to outdoor wind:

indoor_air_velocity = outdoor_windspeed × 0.1

Rationale: Greenhouse structure shields interior from direct wind, reducing air velocity to ~10% of outdoor conditions.

D) Indoor CO₂ Concentration (ppm)

CO₂ levels fluctuate with photosynthesis (day) and respiration (night):

Day (photosynthetic depletion):

indoor_CO2 = 400 - (0.05 × solar_radiation)

Night (respiratory accumulation):

indoor_CO2 = 440 ppm

Constraints: Minimum 300 ppm

Rationale: Plant photosynthesis depletes CO₂ during daylight (proportional to light intensity). At night, plant and soil respiration increases CO₂ concentration above ambient levels.

Derived Environmental Features

E) Dew Point Temperature (°C)

Approximation of dew point using simplified Magnus formula:

dew_point = indoor_temp - ((100 - indoor_humidity) / 5)

F) Vapor Pressure Deficit - VPD (kPa)

Critical metric for plant transpiration and disease risk:

SVP = 0.6108 × exp((17.27 × T) / (T + 237.3))  # Saturation Vapor Pressure
AVP = SVP × (RH / 100)                          # Actual Vapor Pressure
VPD = SVP - AVP

Where:

Interpretation:

G) Leaf Wetness Proxy (binary)

Binary indicator of leaf surface moisture based on sustained high humidity:

leaf_wetness_proxy = 1 if (indoor_humidity > 85%) for ≥3 consecutive hours, else 0

Implementation: Rolling 3-hour window checking for humidity >85%

Significance: Extended leaf wetness periods are strong predictors of fungal disease development.

Output Dataset Specification

Column Schema

Column Type Unit Description Range
datetime datetime - Hourly timestamp -
indoor_temp float °C Indoor air temperature 18-43°C
indoor_humidity float % Relative humidity 30-100%
indoor_air_velocity float m/s Air movement speed 0.1-2.6 m/s
indoor_CO2 float ppm Carbon dioxide concentration 300-440 ppm
solarradiation float W/m² Solar radiation intensity 0-250 W/m²
day_night_flag int - Day=1, Night=0 0 or 1
vpd float kPa Vapor pressure deficit 0-5 kPa
dew_point float °C Dew point temperature 10-30°C
leaf_wetness_proxy int - Leaf wetness indicator 0 or 1

Data Quality Guarantees

✅ Temporal Completeness: Exactly 24 records per calendar day
✅ No Missing Values: All columns fully populated
✅ Physical Constraints: All values within realistic bounds
✅ Temporal Continuity: Chronological ordering preserved

Validation Results

2024 Dataset (Leap Year)

Total Records:        8,784 (366 days × 24 hours)
Date Range:           2024-01-01 00:00:00 to 2024-12-31 23:00:00
Indoor Temperature:   19.8 to 42.8 °C
Indoor Humidity:      43.0 to 100.0%
Indoor CO₂:           384.0 to 440.0 ppm
VPD:                  0.00 to 4.38 kPa
Daylight Hours:       4,435 (50.5%)
Leaf Wetness Hours:   864 (9.8%)

2025 Dataset

Total Records:        8,760 (365 days × 24 hours)
Date Range:           2025-01-01 00:00:00 to 2025-12-31 23:00:00
Indoor Temperature:   18.4 to 41.2 °C
Indoor Humidity:      38.3 to 100.0%
Indoor CO₂:           383.9 to 440.0 ppm
VPD:                  0.00 to 4.76 kPa
Daylight Hours:       4,402 (50.3%)
Leaf Wetness Hours:   502 (5.7%)

Usage Examples

Loading the Dataset

import pandas as pd

# Load indoor greenhouse data
df_2024 = pd.read_csv('data/processed/dindigul_greenhouse_indoor_2024.csv')
df_2024['datetime'] = pd.to_datetime(df_2024['datetime'])

# Set datetime as index
df_2024.set_index('datetime', inplace=True)

print(f"Loaded {len(df_2024)} hourly records")

Basic Analysis

# Daily aggregations
daily_avg = df_2024.resample('D').mean()

# Temperature statistics
print(f"Mean indoor temp: {df_2024['indoor_temp'].mean():.1f}°C")
print(f"Max indoor temp: {df_2024['indoor_temp'].max():.1f}°C")

# High VPD stress events
high_vpd_hours = (df_2024['vpd'] > 1.5).sum()
print(f"Hours with high VPD stress: {high_vpd_hours}")

# Disease risk periods
leaf_wetness_hours = df_2024['leaf_wetness_proxy'].sum()
print(f"Hours with leaf wetness: {leaf_wetness_hours}")

Time Series Visualization

import matplotlib.pyplot as plt

# Plot one week of data
week_data = df_2024['2024-07-01':'2024-07-07']

fig, axes = plt.subplots(3, 1, figsize=(12, 8), sharex=True)

# Temperature
axes[0].plot(week_data.index, week_data['indoor_temp'], label='Indoor')
axes[0].set_ylabel('Temperature (°C)')
axes[0].legend()

# Humidity and VPD
ax_vpd = axes[1].twinx()
axes[1].plot(week_data.index, week_data['indoor_humidity'], 'g-', label='Humidity')
ax_vpd.plot(week_data.index, week_data['vpd'], 'r--', label='VPD')
axes[1].set_ylabel('Humidity (%)', color='g')
ax_vpd.set_ylabel('VPD (kPa)', color='r')

# CO2 with day/night shading
axes[2].plot(week_data.index, week_data['indoor_CO2'])
axes[2].fill_between(week_data.index, 0, 500, 
                      where=week_data['day_night_flag']==0, 
                      alpha=0.2, color='gray', label='Night')
axes[2].set_ylabel('CO₂ (ppm)')
axes[2].legend()

plt.tight_layout()
plt.show()

Disease Risk Assessment

# Calculate daily disease risk score
df_2024['disease_risk'] = (
    (df_2024['leaf_wetness_proxy'] == 1).astype(int) * 0.4 +  # Leaf wetness
    (df_2024['indoor_humidity'] > 80).astype(int) * 0.3 +      # High humidity
    (df_2024['vpd'] < 0.4).astype(int) * 0.3                   # Low VPD
)

daily_risk = df_2024.groupby(df_2024.index.date)['disease_risk'].mean()

# Identify high-risk days
high_risk_days = daily_risk[daily_risk > 0.5]
print(f"High disease risk days: {len(high_risk_days)}")

Processing Pipeline

Notebook Reference

Complete implementation details available in:
📓 notebooks/generate_passive_greenhouse_data.ipynb

Pipeline Steps

  1. Load and Standardize - Import CSV, rename columns, parse datetime
  2. Expand to Hourly - Create 24 records/day with diurnal patterns
  3. Add Day/Night Flag - Classify daylight hours using sunrise/sunset
  4. Generate Greenhouse Conditions - Apply passive physics model
  5. Calculate Derived Features - Compute VPD, dew point, leaf wetness
  6. Validate and Save - Quality checks and CSV export

Reproducibility

To regenerate the datasets:

# Open Jupyter notebook
jupyter notebook notebooks/generate_passive_greenhouse_data.ipynb

# Run all cells (or use "Run All" from Cell menu)

The pipeline will:

Scientific Basis and Assumptions

Model Assumptions

  1. Passive Operation: No active heating, cooling, ventilation, or CO₂ enrichment systems
  2. Standard Greenhouse Structure: Single-layer polyethylene or glass covering
  3. Moderate Plant Density: Typical vegetable crop canopy coverage
  4. Natural Ventilation: Passive air exchange through vents/openings
  5. No Irrigation Control: Natural evapotranspiration only

Physical Principles

Limitations

Applications

Suitable Use Cases

✅ Digital Twin Development: Baseline greenhouse behavior modeling
✅ Control System Testing: Benchmarking against uncontrolled conditions
✅ Disease Risk Prediction: Training ML models for pathogen outbreak forecasting
✅ Growth Stage Simulation: Environmental condition correlation with plant development
✅ Energy Analysis: Passive vs. active system comparison
✅ Irrigation Scheduling: VPD-based watering optimization

❌ Active climate control system design
❌ Precise economic cost modeling
❌ Specific crop variety performance prediction
❌ Structural engineering calculations

Changelog

Version 1.0 - February 2026

Generated: February 24, 2026
Author: AgriTwin-GH Data Engineering Team
Contact: arjun-christopher/AgriTwin-GH

This site is open source. Improve this page.