This study investigates the power prediction and performance of bifacial photovoltaic (PV) systems installed at four demonstration sites, including a fruit farm, a regular field, a rice field, and a salt field. A fence-type bifacial PV system was installed at each site, and power generation was predicted using bifaciality factors (BF) ranging from 0.50 to 0.75. The predicted results were then compared with actual power data collected over the demonstration period. The results revealed that the impact of bifaciality factors on the accuracy of power generation predictions varied by site. At the fruit farm and regular field, higher BF values resulted in larger discrepancies between predicted and measured power generation, with an overestimation of predicted values. This indicates that environmental factors, such as shading and reflectivity, negatively impacted the rear-side power generation of bifacial modules in these locations, suggesting the need for further optimization of BF application in such environments. In contrast, the rice field and salt field exhibited a decrease in prediction error as BF values increased. Notably, the rice field demonstrated the most accurate prediction at a BF value of 0.65, with an error rate of 0%, while the salt field achieved a minimal error rate of 1% at a BF value of 0.75. These results suggest that high-reflectivity environments, such as the salt field, benefit from higher BF values, leading to more accurate power generation predictions.