Creating a gridded global deep ocean temperature and salinity product using machine learning (Ellie Davidson)
This project is using existing data from Argo Floats to assess how accurate a Machine Learning model is when predicting the temperature and salinity of the deep ocean. The machine learning model is trained using data from Deep Argo floats. Then it uses PCA’s of temperature profiles in the upper ocean, along with other features, such as latitude, longitude, and sea surface height, to predict temperature in the deep ocean and create a global gridded product. The predictions created by the random forest and neural network models are then compared to the actual data from the Argo Floats and the World Ocean Atlas. The comparisons are for different latitudes and longitudes as well as time to see if the models’ predictions are within .2 degrees Celsius of the actual data. The difference between that actual data and the predicted data allows for retraining of the machine learning model to make it more accurate.