Recovering rooftop photovoltaic array tilt and azimuth from operational power data

Performance benchmarking and fault detection in distributed photovoltaic systems depend on a theoretical generation reference that requires accurate panel tilt and azimuth as inputs to the irradiance transposition model. In practice, orientation metadata is often missing or derived from design documentation without verification against the physical installation. Data-driven methods that recover orientation from operational power data have been proposed, but no systematic evaluation has examined how irradiance source quality and partial shading affect recovery accuracy. This study evaluated four method combinations on 17 commercially operated Australian rooftop systems with independently verified orientations and documented shading status. For these systems, clear-sky methods gave smaller orientation errors than ERA5 reanalysis methods with both objective functions. On unshaded systems, clear-sky methods achieved median tilt and azimuth errors of 2 ° and 4 ° , while ERA5 methods produced tilt errors of 10 ° to 12 ° and azimuth errors of 48 ° to 62 ° . The clear-sky median azimuth error increased from 4 ° on unshaded systems to 35 ° on shaded systems due to partial shading. A geometric limit on azimuth recovery was identified for systems with panel tilt below 5 ° , where annual plane-of-array irradiance is nearly insensitive to azimuth orientation. For low-tilt rooftop systems in Australia, panel orientation can therefore be recovered from operational DC data without physical inspection. The recovered orientation can improve the reliability of performance benchmarks and fault detection where orientation records are unverified or missing. The approach could support large-scale PV monitoring, where site inspection is impractical.