State-space geolocation models feature coupled process (movement) and observation (data likelihood) models to reconstruct fish movement trajectories using electronic tag data. Development of the data likelihood model is therefore a key step in adapting a state-space geolocation model for use with different fish species, geographical regions, or types of electronic data. Here we adapt a discrete hidden Markov model for the geolocation of Pacific spiny dogfish (Squalus suckleyi, n = 154) in the North Pacific Ocean by developing a data likelihood model based on Microwave Telemetry X-tag Pop-up Satellite Archival Tag (PSAT) data. The data likelihood model consists of light-based longitude, light-based latitude, sea surface temperature (SST), temperature-depth profile (TDP), and maximum daily depth. Pacific spiny dogfish tend to occupy coastal waters where small-scale local currents and freshwater inputs make SST and TDP variables difficult to map. To address this issue, we introduce an empirical method for parameterizing SST and TDP likelihoods by calculating root mean square difference between PSAT temperature and depth values recorded at known locations (day of tag deployment and tag pop-up) and mapped values at those locations. For SST observations (n = 85), the difference between measured and mapped values did not vary seasonally or monthly and the overall root mean square error (RMSE) used to parameterize the SST likelihood was 0.9 °C. Likelihood values for SST at known locations were higher for likelihoods parameterized with the empirical value compared to variance specification methods from previous studies. For TDP, measured values differed from mapped values (n = 89) by depth, season, and month. Therefore, RMSE values used to parameterize the TDP likelihood were calculated for each depth bin (n = 27) and month. RMSE values were low (< 1 °C) for all depths during the winter but increased for depths < 100 m during the summer months. Our work provides an example of adapting state-space geolocation models for specific applications. It demonstrates the value of large numbers of tagged animals for parameterizing the data likelihood model in coastal waters as well as flexible data likelihood models with component likelihoods that can be switched on or off depending on geolocation quality.