Forensically useful mid-term and short-term temperature reconstruction for quasi-indoor death scenes

Abstract

While estimating postmortem interval (PMI) ambient temperature plays a pivotal role, so its reconstruction is crucial for forensic scientists. The recommended procedure is to correct temperatures from the nearest meteorological station based on measurements from the death scene; typically applying linear regression. Recently, there have been attempts to use different algorithms, that can improve that correction, for example GAM algorithm. Unfortunately, the improvements are usually a consequence of using more dependent variables than just the temperature from the death scene (e.g. humidity), which is impractical.

This study develops practical new methods to accurately reconstruct ambient temperatures at a death scene, using just temperature measurements. Since the main difficulty preventing practitioners from using the correction protocol more frequently is likely the need to record temperatures on-site for at least several days, we searched for possibilities to shorten the measurement period. For this purpose, we tested two less popular algorithms to achieve this goal. The concurrent regression model (the model from the functional data analysis field) for the mid-term reconstruction (measurements lasting several days) and the functional model based on Fourier expansion for the short-term reconstruction (measurements lasting a few hours).

The algorithms’ performance was tested using data collected in six places: a roof and an attic of a heated building, an unheated garage inside the heated building, an unheated wooden shack, an uninhabited building, and an underground (the data logger was buried about 30 cm below the ground level). We classified these locations as quasi-indoor conditions, contrasting them with typical indoor conditions, where temperatures are nearly constant, and typical outdoor conditions, where there is no heat insulation.

The mid-term model reduced error compared to the linear regression, providing nearly perfect reconstruction for measurement periods longer than six days. More importantly, however, the accuracy of short-term reconstruction was also high. The short-term model closely matched the concurrent regression model’s performance after only four to five hours of measurements.

In practice, both methods are very similar to the standard procedure. The main difference is the change in the algorithm and its implementation. In conclusion, this study demonstrates that correction of weather station temperatures can provide fairly accurate temperature data for use in estimating PMI after only 4-5 h of measurements on a death scene.