Development of an AI system for environmental monitoring
Environmental monitoring covers the condition of air, water, soil, and pollution sources. The AI system integrates data from fixed posts, satellites, and mobile sensors to create a realistic picture of the environmental situation and predict hazardous situations.
Monitoring system components
Atmospheric air monitoring:
- PM2.5, PM10 (fine particles)
- NOx, SO2, CO, O3 (gas pollutants)
- Benzo(a)pyrene and VOCs (volatile organic compounds)
- Meteorological parameters: wind, temperature, humidity, pressure
Monitoring of water bodies:
- Physical: temperature, turbidity, color
- Chemical: pH, dissolved O2, COD, BOD
- Biogenic: nitrates, phosphates, ammonium
- Specific: heavy metals, petroleum products
Soil monitoring:
- Heavy metals (by profile)
- Oil pollution
- Acidity, humus
IoT infrastructure
Stationary posts:
- State: Federal State Budgetary Institution "Hydrochemical Institute", Federal State Budgetary Institution "Central Laboratory of Mechanics"
- Industrial enterprises: mandatory posts of the sanitary protection zone (SPZ)
- Independent: environmental NGOs, smart cities
Low-cost sensor: Low-cost IoT sensors (Plantower PMS7003, SPS30 for PM) allow you to create dense networks:
- Node price: $50-200 vs. $10,000-50,000 for a professional station
- Accuracy: lower, requires calibration at a reference station
def calibrate_low_cost_sensor(low_cost_readings, reference_readings, method='linear'):
LCS calibration using the nearest reference station
if method == 'linear':
model = LinearRegression().fit(low_cost_readings, reference_readings)
return model # apply to future LCS data
elif method == 'rf':
model = RandomForestRegressor().fit(low_cost_readings, reference_readings)
return model
Satellite data:
- Sentinel-5P (TROPOMI): NO2, SO2, CO, O3 - global coverage, 3.5×5.5 km
- Landsat 8/9 + Sentinel-2: surface water, disturbed soils
- MODIS: NDVI, thermal anomalies (fires)
Predictive models
Air quality forecast:
# LSTM + spatial interpolation
# State: 24-hour PM2.5 time series for all stations in the region
# + NWP weather forecast (wind, temperature, atmospheric mixing)
# Output: PM2.5 for the next 24/48/72 hours for each grid
model = StackedLSTM(
input_size=n_stations * n_pollutants + n_meteo_vars,
hidden_size=128,
forecast_hours=72
)
Gaussian plume dispersion models: If the emission source is known, calculate the pollution zone:
- AERMOD / AERSCREEN: US regulatory models
- OND-90: Russian standard dispersion calculation
- ML corrections to deterministic models
Source detection: The inverse problem is to determine the location of the source based on the distribution of concentrations:
- Optimization: minimizing the difference between the observed and modeled field
- Deep Learning: Concentration field image encoder → source coordinates
Alert system
Air quality indices:
- AKI (Atmospheric Quality Index) is a Russian standard
- WHO 2021 Guidelines: PM2.5 < 5 µg/m³ is safe, > 35 is dangerous
- US AQI: 0-500, color coded
Automatic alerts:
- If the maximum permissible concentration is predicted to be exceeded → notification to the population (SMS, mobile application)
- If the maximum permissible emission limit (MPEL) is exceeded → notification to RPN/Rosprirodnadzor
- In case of emergency discharge → Ministry of Emergency Situations + enterprise
Compliance with Federal Law 174 and NDT
Russian legislation:
- Federal Law No. 174 "On Environmental Expertise": requirements for monitoring systems
- Resolution 205 (2019): mandatory automatic control for Category I facilities
- Order of the Ministry of Natural Resources 522: data transfer format in the GIS "Best Available Technologies"
Integration with government systems:
- FGIS "Industry": reporting of environmental impact assessment facilities
- AIS "Industrial Ecology" of Rosprirodnadzor
- Regional GIS for environmental protection
Deadlines: Basic IoT monitoring system + visualization + AQI forecast – 8-10 weeks. A full-fledged platform with source attribution, regulatory reporting, and alerts – 5-6 months.