Ok, I found out why my heatmap plots look so bad. When you run rtl_power the bias-t gets turned off. That turns off the LNA and then you’re just looking at noise from the rtl-sdr dongle front end. As far as I can tell, you can’t turn on the internal bias-t in the rtl-sdr dongle and run rtl_power. You would have to supply an external bias-t voltage to the LNA to turn it on.
Here is a heatmap png where I took out the LNA and just increased the gain in the rtl-sdr dongle to 50 dB. Looks more normal:
You can also tell if the heatmap is going to be reasonable by looking at the lines that heatmap.py puts out when it runs. For the above I got:
loading
x: 4576, y: 60, z: (-34.560000, 4.330000)
drawing
labeling
saving
Here there is a z min to max of about 39 dB. On the previous (noise) plots, z had a min to max of about 3 dB.