Hi Quincy,

CO2 concentration /is/ dependent on air temperature, have a look at this paper.
Also, the CO2 sensors used (Winsen MH-Z19) are not very accurate, the datasheet states +/- 50ppm + 5% of reading.
The datasheet also states that the reading is “temperature compensated”…one never knows how good that works inside a device that generates additional heat.
It is a cheap sensor after all and one cannot expect laboratory-grade precision from them.
Additionally, that sensor performs some magic self calibration (that could explain the sudden drop to 450 ppm) and can also be calibrated by the MCU.

HTH & Cheers,
Wolferl