Radar detection of subsurface ice on Mars has been widely debated in part because the dielectric signature of ice, as deduced from the dielectric constant, can be confused with dry-silicate-rich materials. To identify the ice dielectric signature, it is crucial to estimate the imaginary part of the dielectric permittivity inferred from the dielectric attenuation after removing the scattering loss. Unfortunately, the latter remains poorly quantified at both Mars Advanced Radar for Subsurface and Ionospheric Sounding (MARSIS) and shallow subsurface radar SHARAD frequencies. To address this ambiguity, we conducted multiple-frequency ground-penetrating radar and resistivity investigations in well-characterized temperate permafrost in Fairbanks, Alaska. The area shows several geomorphologic similarities to midlatitude and high-latitude terrains on Mars. This approach allowed us to quantify the dielectric and scattering losses in temperate permafrost over the 10 to 1000 MHz frequency band. At 20 MHz, our results suggest an average dielectric loss rate of 0.25 +/- 0.03 dB/m, whereas the corresponding average scattering loss rate is 0.94 +/- 0.37 dB/m. The scattering loss was found to represent similar to 69% of the total signal attenuation. Considering this result and the study by Heggy et al. (2006a) in volcanic environments, we revised the interpretation of the attenuation coefficient calculated from SHARAD data over the Deuteronilus Mensae region and Amazonis Planitia; we then used the reevaluated dielectric loss to estimate the imaginary part of the dielectric permittivity. Our results suggest that even if Deuteronilus Mensae deposits and the Vastitas Borealis Formation may have similar dielectric constants, their imaginary parts are different. This implies that the two regions have different bulk compositions, with the former being ice-rich sediments and the latter being nonconsolidated volcanic deposits