Photosynthetically active radiation decomposition models for agrivoltaic systems applications

•Seven stand-alone decomposition models for solar radiation are applied to PAR.•Comparison and re-parameterization are performed against measured PAR values in three locations in Sweden.•The best-performing decomposition model is Yang2 with nRMSE of 23.75% in Norunda.•For agrivoltaic studies, Starke model is suggested if satellite data cannot be retrieved.•EMOS approach is explored and provides further accuracy (nRMSE of 23.02% in Norunda) Decomposition models of solar irradiance estimate the magnitude of diffuse horizontal irradiance from global horizontal irradiance. These two radiation components are well known to be essential for predicting the performance of solar photovoltaic systems. In open-field agrivoltaic systems (i.e., the dual use of land for both agricultural activities and solar power conversion), cultivated crops receive unequal amounts of direct, diffuse, and reflected photosynthetically active radiation (PAR). These uneven amounts depend on where the crops are growing due to the non-homogenous shadings caused by the presence of the installed solar panels (above the crops or vertically mounted). It is known that, per unit of total PAR, diffuse PAR is more efficient for canopy photosynthesis than is direct PAR. For this reason, it is essential to estimate the diffuse PAR component when agrivoltaic systems are being assessed, in order to properly predict the crop yield. Since PAR is the electromagnetic radiation in the 400–700 nm waveband that can be used for photosynthesis by the crops, several stand-alone decomposition models typically used to split global horizontal irradiance are selected in this study to decompose PAR into direct and diffuse. These models are applied and validated in three locations in Sweden (Lanna, Hyltemossa and Norunda) using the coefficients stated on the models’ original publications and locally fitted coefficients. The results showed weaker performances in all stand-alone models for non-locally fitted coefficients (nRMSE ranging from 27% to 43%). However, performances improve with re-parameterization, with a highest nRMSE of 35.24% in Lanna. The Yang2 decomposition model is the best-performing one, with the lowest nRMSE of 23.75% in Norunda when applying re-estimated coefficients. Country level sets of coefficients for the best-performing models (Yang2 and Starke) are given after parameterization using combined data for all three locations in Sweden. These Sweden-fitted models are tested and show an nRMSE of 25.0