Methods for quick and accurate detection and diagnosis of defects in PV systems are increasingly important as the global photovoltaic (PV) capacity continues to grow at a rapid pace. Two of the most used methods for defect detection involve aerial infrared thermography and data analysis of production data. In this work, we combine the two methods to analyze two utility scale PV plants, providing new understanding about the two methods. We report on the percentage and distribution of thermal anomalies of different categories and quantify their relationship with performance on string level. We find that the most important parameter for determining production losses on string level is the number of module substrings containing thermal anomalies. Due to the large variability of the effect of different thermal signatures, as well as uncertainties in the estimate of the string performance, we find no clear correlation between performance and thermal signature category or temperature. However, a whole hot cell is the thermal signature that on average has the smallest impact on the power output on string level. Finally, in our data, the performance of a string of 20 modules with 3 bypass diodes is on average reduced by 1.16 +/- 0.12% per module substring containing thermal anomalies.