Calibrating autonomous recording units to estimate sound pressure levels using frequency-band univariate models and microphone sensitivity

Anthropogenic sound is widespread across human-dominated and semi-natural landscapes, and many applications require spatially dense exposure estimates that are difficult to obtain with sound level meters alone. Autonomous recording units (ARUs) enable dense acoustic sampling, but audio-derived levels are device- and workflow-dependent and therefore require empirical calibration before they can be compared with meter-referenced sound pressure levels (SPL). Here we derive field-paired octave-band calibration equations that map 60 s ARU-derived band levels to concurrent sound level meter measurements across seven field sound contexts. Using univariate, band-specific models with device-type offsets, unit microphone sensitivity, and meteorological covariates, we show that recorder-derived levels can be translated into meter-referenced octave-band SPL with quantified uncertainty within the sampled domain. Across 10-fold cross-validation, overall bias was −0.07 dB, mean absolute error was 3.77 dB, and root mean squared error was 4.82 dB; 72% of estimates were within ± 5 dB of the meter. Performance varied by device and band, with Song Meter units clustering between 4.37 and 4.66 dB RMSE, smartphones showing larger errors (6.08 dB RMSE), and larger errors at mid-frequency bands and high-SPL observations. We provide coefficient tables, diagnostics, and worked examples for applying the calibration to new 60 s recordings from tested device types processed with the same extraction workflow. The method offers a reproducible route from ARU recordings to meter-referenced octave-band SPL estimates under defined conditions, while retaining explicit limits on extrapolation beyond the sampled devices, weather, sound contexts, and signal-processing pipeline.