________________________________________________________________________________________________________________________________________________
November 2021 - Lee et al. 2021
Applying machine learning methods to detect convection using Geostationary Operational Environmental Satellite-16 (GOES-16) advanced baseline imager (ABI) data
Atmospheric Measurement Techniques (AMT), https://doi.org/10.5194/amt-14-2699-2021
Visible and infrared data from geostationary satellites are available nearly anywhere and in near-real time, but it has been challenging to detect convection using those data due to the lack of vertical information. However, the current operational geostationary satellite, GOES-16 provides data in high spatial and temporal resolutions, and we can easily identify convection by looking at an image loop from GOES-16. In order to mimic how humans identify convection from GOES-16 images, a neural network model with convolution layers is developed. Inputs are five channel 2 reflectance data with two-minute interval and five channel 14 brightness temperature data with two-minute interval, and the model is trained against radar products called Multi-Radar Multi-Sensor (MRMS). Adaptive loss function is tested along with the standard loss function. Results show that the model was able to detect convection based on high reflectance, low brightness temperature, and lumpy textures in reflectance data.
Tags: machinelearning; ML; CNN; GOES16; ABI; convection; precipitation
