Smogseer: A Convolutional LSTM for forecasting air quality from Sentinel-5P data

Training and checkpoint datasets S5PL2_5D.nc This is the Sentinel-5P traning dataset for the Smogseer ConvLSTM model. The dataset was created using the xcube Sentinel Hub data store from the Deep Earth System Data Lab. The dataset has the following characteristics: bbox= [68.137207,24.886436,84.836426,34.379713] #WGS84 // lon,lat,lon,lat res = (bbox[2]-bbox[0])/512 # ~3629m date_range = ['2019-01-01', '2023-12-31'] timesteps = '5D' X_val.npy Validation feature data with shape: (74, 1, 291, 512, 6) 74: dates 1: time steps 291: latitudes 512: longitudes 6: Features ['SO2', 'NO2', 'CH4', 'O3', 'CO', 'HCHO'] Y_val.npy Validation target data with shape: (74, 1, 291, 512, 1) 74: dates 1: time steps 291: latitudes 512: longitudes 6: Features ['SO2', 'NO2', 'CH4', 'O3', 'CO', 'HCHO'] smogseer50.keras Model weights for training the ConvLSTM with 50 epochs. smogseer100.keras Model weights for training the ConvLSTM with 10 epochs.