Modelling energy losses arising from overridden backtracking in utility-scale photovoltaic plants on slightly undulating terrain: Implementation in the SISIFO tool for performance pre-assessment

Underperformance of utility-scale photovoltaic plants continues to be systematically reported. Based on a detailed analysis over one year of 10–15-minute measurements from over seven thousand single-axis trackers across seven utility-scale photovoltaic plants, this study identifies a previously unaddressed source of such underperformance: tracking irradiation gain during backtracking is lower than expected. Slight terrain unevenness forces tracker controllers to override optimal backtracking angles to avoid inter-row shading, reducing tracker tilt and direct irradiance collection compared to flat-terrain simulations. This leads to a new loss category, here denominated suboptimal backtracking losses, which remain hidden within standard performance indicators and create an unquantified gap between simulated and actual energy yield. To address this, three overridden backtracking models are developed to reproduce real tracker behaviour through (1) direct overriding of the tracking angle, (2) analytical overriding of inter-row spacing, and (3) a virtual ground slope representing unevenness. Implemented in SISIFO, an open-source photovoltaic simulation tool, the models’ estimates better match real performance, allowing estimation of suboptimal backtracking losses and explaining up to 2% of irradiance capture losses relative to conventional flat-terrain approaches. At utility scale, this difference is large enough to affect annual energy yield predictions and plant profitability, quantifying backtracking-related underperformance. • A cause of underperformance in utility-scale one-axis tracking PV plants is revealed. • SCADA data from trackers show intentional overriding of optimal backtracking angles. • Tracking radiation gain is lower than expected due to slightly uneven terrain. • In-plane irradiation and final energy yield losses reach 2%. • Identifies unreported suboptimal backtracking losses and three methods to model them.