Suppose I have access to a network of weather station that measure many variables in near-real time. I want to produce an interpolated product with temperature, humidity, pressure, etc.

The first (easy) way of doing this would be to use a classical interpolation method : Nearest neighbour, natural neighbour, Inverse weighted distance, Kriging.... All these methods use some a-priori mathematical and statistical knowledge to derive the best approximation of the variable over a grid. However, they all lack physical knowledge.

I would like to do the same but using a lightweight assimilation technique. Instead of using a classical method I'd give an ideal package every information I have in a certain moment (for example not only temperature measured at stations but also satellite measurements, radar measurements, altitude, sondes measurements...) and get back the best physical approximation of the atmosphere at the surface.

This is formulated exactly as a typical NWP assimilation method, but I want to run it with less variables and to get the conditions only at the surface. I know that these methods can be really expensive so I was wondering if there's any way to do this in a lightweight manner, ideally with a Python package. The final goal is to have kind of a synoptic analysis of temperature, humidity, precipitation, etc.

Thanks