Predicting Atmospheric Variables in the MERRA-2 Database using Neural Networks

Meteorological research often requires modelling atmospheric variables. With an increase of interest in climate science due to the urgency of climate change, easier access to meteorological data could help accelerate climate mitigation efforts. Specifically, analysis in aviation fields would be impacted greatly with more accurate data. Equations from the International Standard Atmosphere (ISA) are conventionally used to calculate barometric variables but yield only ideal results because they are generalisations of our atmosphere. For applications in aviation and weather research, more accurate values are desired. NASA's MERRA-2 database provides detailed meteorological information about pressure, temperature, humidity, and wind in 3-hour intervals from 1980. However, the data is too large to be feasibly used and inaccessible. Making a mathematical model to summarise it is therefore incredibly valuable. Conventional statistical models fail to represent the complexity of such atmospheric data making machine learning a worthy competitor. This research has successfully explored the possibility of using Artificial Intelligence-based Neural Networks to create a model to predict atmospheric variables from 4-D inputs (latitude, longitude, altitude, and time) to greatly improve upon the standard barometric equations in the ISA.