The emerging field of photoacoustic tomography is rapidly evolving with many new system designs and reconstruction algorithms being published. Many systems use water as a coupling medium between the scanned object and the ultrasound transducers. Prior to a scan, the water is heated to body temperature to enable small animal imaging. During the scan, the water heating system of some systems is switched off to minimize the risk of bubble formation, which leads to a gradual decrease in water temperature and hence the speed of sound. In this work, we use a commercially available scanner that follows this procedure, and show that a failure to model intra-scan temperature decreases as small as 1.5°C leads to image artifacts that may be difficult to distinguish from true structures, particularly in complex scenes. We then improve image quality by continuously monitoring the water temperature during the scan and applying variable speed of sound corrections in the image reconstruction algorithm. While upgrading to an air bubble-free heating pump and keeping it running during the scan could also solve the changing temperature problem, we show that a software correction for the temperature changes provides a cost-effective alternative to a hardware upgrade. The efficacy of the software corrections was shown to be consistent across objects of widely varying appearances, namely physical phantoms, ex vivo tissue, and in vivo mouse imaging. To the best of our knowledge, this is the first study to demonstrate the efficacy of modeling temporal variations in the speed of sound during photoacoustic scans, as opposed to spatial variations as focused on by previous studies. Since air bubbles pose a common problem in ultrasonic and photoacoustic imaging systems, our results will be useful to future small animal imaging studies that use scanners with similarly limited heating units.