Using compressed sensing (CS), stepped Raman optical frequency-domain reflectometry (ROFDR) can be performed on a reduced set of frequencies. We experimentally show that temperature hotspots can be reconstructed from a small subset of the usually required frequency steps when the measured temperature profile corresponds to a known model. In this case, the measurement rate can typically be increased by a factor of 4. CS reconstruction of temperature hotspots is investigated in two different setups. The results, compared to full measurements, indicate that CS is a viable practical technique for ROFDR when the expected temperature variations follow the assumed model and the signal to noise ratio is sufficient.