Global Air Pollution and Respiratory Health Risk Mapping via GeoAI

// 1. Define Urban Study Area (Example: Northern Italy Industrial Hub / Po Valley)
var roi = ee.Geometry.Polygon([
  [[7.5, 44.5], [12.5, 44.5], [12.5, 46.5], [7.5, 46.5]]
]);

// 2. Load and Filter Sentinel-5P Tropospheric NO2
var startPeriod = '2023-05-01';
var endPeriod = '2023-09-30';

var s5p_no2 = ee.ImageCollection('COPERNICUS/S5P/OFFL/L3_NO2')
  .filterBounds(roi)
  .filterDate(startPeriod, endPeriod)
  .select('tropospheric_NO2_column_number_density')
  .mean()
  .clip(roi);

// 3. Load Population Density from GHSL 2020
var pop = ee.Image('JRC/GHSL/P2023A/GHS_POP/2020')
  .select('population_count')
  .clip(roi);

// 4. Load Meteorological Covariates (ERA5-Land Monthly)
var era5 = ee.ImageCollection('ECMWF/ERA5_LAND/MONTHLY_AGGR')
  .filterBounds(roi)
  .filterDate(startPeriod, endPeriod)
  .select(['temperature_2m', 'u_component_of_wind_10m', 'v_component_of_wind_10m'])
  .mean()
  .clip(roi);

// 5. Load MODIS AOD (Aerosol Optical Depth at 0.47um)
var aod = ee.ImageCollection('MODIS/061/MCD19A2_GRANULES')
  .filterBounds(roi)
  .filterDate(startPeriod, endPeriod)
  .select('Optical_Depth_047')
  .mean()
  .multiply(0.001)
  .rename('AOD_470')
  .clip(roi);

// 6. Calculate Population-Weighted Exposure Raster
// Exposure = NO2 Concentration * Log(Population Count + 1)
var popExposure = s5p_no2.multiply(pop.add(1).log()).rename('Pop_Weighted_Exposure');

// 7. Visualization Setup
Map.centerObject(roi, 7);
Map.addLayer(pop, {min: 0, max: 1000, palette: ['white', 'yellow', 'red']}, 'GHSL Population Count');
Map.addLayer(s5p_no2, {min: 0.00002, max: 0.00015, palette: ['black', 'blue', 'purple', 'orange', 'red']}, 'Sentinel-5P NO2 Mol/m2');
Map.addLayer(popExposure, {min: 0, max: 0.001, palette: ['green', 'yellow', 'orange', 'red', 'purple']}, 'Population Weighted Exposure');

// 1. Stack All Predictors into a Multi-Band Composite Image
var composite = ee.Image.cat([
  s5p_no2.rename('NO2'),
  aod,
  pop.rename('Population'),
  era5.select('temperature_2m').rename('Temp'),
  era5.select('u_component_of_wind_10m').rename('U_Wind'),
  era5.select('v_component_of_wind_10m').rename('V_Wind')
]);

// 2. Define Proxy Target Variable: Integrated Respiratory Risk Score (0-100 Normalized)
// Target combines normalized NO2 exposure, aerosol load, and population vulnerability
var normalizedNO2 = s5p_no2.unitScale(0.00001, 0.00020).clamp(0, 1);
var normalizedAOD = aod.unitScale(0.05, 0.8).clamp(0, 1);
var normalizedPop = pop.log().unitScale(0, 8).clamp(0, 1);

var targetRisk = normalizedNO2.multiply(0.5)
  .add(normalizedAOD.multiply(0.3))
  .add(normalizedPop.multiply(0.2))
  .multiply(100)
  .rename('Respiratory_Risk_Index');

// 3. Sample Training Points across the Region of Interest
var trainingData = composite.addBands(targetRisk).sample({
  region: roi,
  scale: 1000,
  numPixels: 2000,
  seed: 42,
  geometries: true
});

// 4. Train a Random Forest Regression Model
var rfRegressor = ee.Classifier.smileRandomForest(100)
  .setOutputMode('REGRESSION')
  .train({
    features: trainingData,
    classProperty: 'Respiratory_Risk_Index',
    inputProperties: ['NO2', 'AOD_470', 'Population', 'Temp', 'U_Wind', 'V_Wind']
  });

// 5. Predict Respiratory Health Risk across the Entire Spatial Domain
var predictedRiskMap = composite.classify(rfRegressor, 'Predicted_Respiratory_Risk');

// 6. Visualization of Machine Learning Risk Prediction
var riskPalette = ['#1a9641', '#a6d96a', '#ffffbf', '#fdae61', '#d7191c'];
Map.addLayer(predictedRiskMap, {min: 10, max: 80, palette: riskPalette}, 'Predicted Respiratory Risk Index');

// 7. Model Feature Importance Verification
var explanation = rfRegressor.explain();
print('Random Forest Model Details:', explanation);

// Export Predicted Respiratory Risk Map as a GeoTIFF to Google Drive
Export.image.toDrive({
  image: predictedRiskMap,
  description: 'Global_Urban_Respiratory_Health_Risk_2023',
  folder: 'GEE_Health_GeoAI',
  scale: 1000,
  region: roi,
  maxPixels: 1e9,
  fileFormat: 'GeoTIFF'
});

// Export Sampled Health Exposure Table as CSV
Export.table.toDrive({
  collection: trainingData,
  description: 'Respiratory_Risk_Sample_Points_2023',
  folder: 'GEE_Health_GeoAI',
  fileFormat: 'CSV'
});