Evaluation of reverse transposition and separation methods for global tilted irradiance: Insights from high-latitude data

Accurate separation of global tilted irradiance (GTI) becomes important when the measured irradiance is used for quality control or PV simulation purposes, for which the latter often requires global horizontal irradiance (GHI), or diffuse and beam irradiance fractions. This study presents an evaluation of irradiance reverse transposition and separation models for the application with GTI in high latitudes. The evaluation is made based on measured and quality controlled six-second irradiance from latitude 59.53°N, containing GTI at 30°, 40°, and 90° tilt angles, as well as GHI and diffuse horizontal irradiance (DHI). Based on a literature review, two specialized GTI reverse transposition and separation models – GTI-DIRINT and Perez-Driesse – and four GHI separation models were chosen for evaluation. The latter were tested in an optimization loop developed for this study that utilizes existing GHI separation models combined with transposition models for reverse transposition and separation of GTI. Specifically, the separation models Erbs , Skartveit1 , Engerer2 , and Yang4 were tested with Hay & Davies and Perez1990 transposition. The models were investigated using both statistical evaluation metrics and Diebold–Mariano test to compare measured and predicted GHI and DHI. An evaluation with measured data showed that for GTI reverse transposition and separation at high latitudes, the use of the proposed optimization model with Engerer2 in combination with Hay & Davies transposition, or the Perez-Driesse model is recommended. This is based on overall good ranking and low bias of GHI prediction with −2.0 W/m 2 and −2.3 W/m 2 , respectively. • Evaluated two established and one novel model using high-latitude data. • Proposed fmincon optimization model is suitable for application with high-latitude measured data. • Lowest bias: Engerer2 with Hay & Davies (−2.0 W/m 2 ) or Perez-Driesse (−2.3 W/m 2 ). • All evaluated models tend to overpredict GHI during clear sky conditions.