Low-cost sensors have the potential to create dense air monitoring networks that help enhance our understanding of pollution exposure and variability at the individual and neighborhood-level; however, sensors can be easily influenced by environmental conditions, resulting in performance inconsistencies across monitoring settings. During summer 2020, 20 low-cost particulate sensors were deployed with a reference PM2.5 monitor in Denton, Texas in preparation for calibration. However, from mid to late-summer, dust transported by the Saharan Air Layer moved through the North Texas region periodically, influencing the typical monitoring pattern exhibited between low-cost sensors and reference instruments. Traditional modeling strategies were adapted to develop a new approach to calibrating low-cost particulate sensors. In this study, data collected by sensors was split according to a novel dust filter into dust and non-dust subsets prior to modeling. This approach was compared with building a single model from the data, as is typically done in other studies. Random forest and multiple linear regression algorithms were used to train models for both strategies. The best performing split-model strategy, the multiple linear regression models split according to dust and non-dust subsets (combined R2 = 0.65), outperformed the best performing single-model strategy, a random forest model (R2 = 0.49). …