Global Digital Epidemiology: Integrating Search Signals with GeoAI and Earth Observation

// Load global administrative boundaries
var countries = ee.FeatureCollection('USDOS/LSIB_SIMPLE/2017');

// Define temporal range for environmental reduction
var startDate = '2023-01-01';
var endDate = '2023-12-31';

// Load air temperature from ERA5-Land Daily (Convert Kelvin to Celsius)
var era5 = ee.ImageCollection('ECMWF/ERA5_LAND/DAILY_AGGR')
  .filterDate(startDate, endDate)
  .select('temperature_2m')
  .mean()
  .subtract(273.15)
  .rename('mean_air_temp');

// Load Land Surface Temperature from MODIS
var modisLST = ee.ImageCollection('MODIS/061/MOD11A1')
  .filterDate(startDate, endDate)
  .select('LST_Day_1km')
  .mean()
  .multiply(0.02)
  .subtract(273.15)
  .rename('mean_lst');

// Load Population Density from WorldPop
var population = ee.Image('WorldPop/GP/100m/pop')
  .rename('pop_density');

// Combine environmental layers into a single multi-band image
var predictors = era5.addBands(modisLST).addBands(population);

// Aggregate environmental mean values across each country boundary
var countryFeatures = predictors.reduceRegions({
  collection: countries,
  reducer: ee.Reducer.mean(),
  scale: 10000
});

// Attach synthetic Google Trends Relative Search Volume (RSV) health proxy
// (Simulating search interest correlated with climate exposure for demonstration)
var healthDataset = countryFeatures.map(function(f) {
  var temp = ee.Number(f.get('mean_air_temp'));
  var pop = ee.Number(f.get('pop_density'));
  // Synthetic expression representing fever/influenza search volume proxy
  var rsvProxy = temp.multiply(-0.8).add(pop.log().multiply(2.5)).add(50);
  return f.set('RSV_Proxy', rsvProxy);
}).filter(ee.Filter.notNull(['mean_air_temp', 'mean_lst', 'pop_density', 'RSV_Proxy']));

// Print feature sample to Console
print('Processed Global Health Features:', healthDataset.limit(5));

// Visualization setup
Map.setCenter(0, 20, 2);
Map.addLayer(era5, {min: -10, max: 35, palette: ['blue', 'yellow', 'red']}, 'Mean Air Temp (C)');
Map.addLayer(countries, {color: '888888'}, 'Global Boundaries', false);

// Prepare Machine Learning training attributes
var predictorNames = ['mean_air_temp', 'mean_lst', 'pop_density'];
var targetProperty = 'RSV_Proxy';

// Train a Random Forest Regressor inside Earth Engine
var rfRegressor = ee.Classifier.smileRandomForest(100)
  .setOutputMode('REGRESSION')
  .train({
    features: healthDataset,
    classProperty: targetProperty,
    inputProperties: predictorNames
  });

// Evaluate model resubstitution error
var trainedModel = healthDataset.classify(rfRegressor, 'Predicted_RSV');
print('Model Training Diagnostics:', rfRegressor.explain());

// Predict spatial health query risk raster continuously across global land masses
var spatialRiskMap = predictors.select(predictorNames).classify(rfRegressor);

// Apply population mask to exclude uninhabited zones
var maskedRiskMap = spatialRiskMap.updateMask(population.gt(1));

// Display predicted digital epidemiological risk map
var riskVis = {
  min: 20,
  max: 80,
  palette: ['#2b83ba', '#abdda4', '#ffffbf', '#fdae61', '#d7191c']
};

Map.addLayer(maskedRiskMap, riskVis, 'Predicted Health Search Risk Index');

// Export predicted spatial risk map as geotiff
Export.image.toDrive({
  image: maskedRiskMap,
  description: 'Global_Digital_Epidemiology_Risk_Map',
  scale: 10000,
  region: ee.Geometry.BBox(-180, -60, 180, 85),
  maxPixels: 1e13
});

// Export summarized country-level table including actual and predicted search proxies
Export.table.toDrive({
  collection: trainedModel,
  description: 'Global_Digital_Epidemiology_Country_Metrics',
  fileFormat: 'CSV',
  selectors: ['COUNTRY_NA', 'mean_air_temp', 'mean_lst', 'pop_density', 'RSV_Proxy', 'Predicted_RSV']
});