The same weather-state model, now a real orbitable atmosphere
The scene is driven by four inputs (temperature T, relative humidity RH, precipitation rate R, wind v). Everything else is computed from them.
Dew point (Magnus): g = ln(RH/100) + 17.62 T/(243.12 + T); Td = 243.12 g/(17.62 - g)
Cloud base (LCL): z = 125 m per C x (T - Td)
Wet-bulb (Stull 2011): Tw = T atan(0.151977 sqrt(RH+8.313659)) + atan(T+RH) - atan(RH-1.676331)
+ 0.00391838 RH^1.5 atan(0.023101 RH) - 4.686035
Phase: snow if Tw <= 0.5 C and R > 0.1 mm/h, otherwise rain
Wind chill (NWS 2001, T <= 10 C, v > 4.8 km/h): 13.12 + 0.6215 T - 11.37 v^0.16 + 0.3965 T v^0.16
Heat index (Rothfusz regression) for T >= 27 C and RH >= 40%
Beaufort: B = (v / 0.836)^(2/3) with v in m/s
Raindrop fall: D = 0.9 R^0.21 mm (Marshall-Palmer median), v = 9.65 - 10.3 exp(-600 D) m/s (Gunn-Kinzer fit)
Visibility: V = 3.912 / sigma (Koschmieder), sigma = 0.15 (1 - RH)^-0.6 + rain/snow extinction
What you see. Cloud altitude follows the lifting condensation level, rain falls at a speed proportional to the computed terminal velocity (scene speed is a visual scale, not to scale), and fog distance follows the computed visibility.
Honest simplifications. LCL assumes a surface parcel lifted dry-adiabatically; the visibility law is an empirical illustration, not a measurement; the thunderstorm preset flashes at random and does not model charge separation; cloud and terrain sizes are stylised. Temperature, humidity and rain are independent sliders, so physically odd combinations are possible (heavy rain at 20% humidity).