CO2 inverse modeling accuracy

Question

I used an ERA5 database and data from a weather ground station.

I want to estimate CO2 emission rates from known sources using a combination of weather station data, mobile CO2 sensors and databases that are available to the public such as ERA5. What would be the best computational approach?

I have tried, in Python: the Gaussian method, the combined optimization method Ant Colony Optimization and Interior Point Method( recommended by a research paper I found) and the Nelder Mead optimization method to compare to. It delivers wildly different results. I don't understand why.