Satellite derived product benchmarking and empirical model development for estimating photosynthetically active radiation at high latitudes

• PAR estimated using SARAH-3, CERES, and STRÅNG was benchmarked against ICOS data. • Linear models estimated PAR accurately at sites with GHI measurements (nMAE < 8 %). • A spatial regression model predicted PAR well at untrained sites (R² ≥ 0.98). • SARAH-3 was the most reliable satellite source when GHI data was unavailable. • Results support practical PAR estimation for high-latitude agrivoltaic modelling. Photosynthetically Active Radiation (PAR) is a critical parameter for understanding plant growth and optimising agricultural productivity. Accurate estimation and measurement of PAR are essential for various applications, including the design of agrivoltaic systems, which enable dual use of land for solar energy conversion and crop cultivation. Despite its importance, routine measurements of PAR remain scarce globally, creating a significant gap in comprehensive tracking. This study addresses this gap by comparing PAR estimates derived from satellite sources such as CERES and SARAH-3 and the mesoscale model STRÅNG with weather station measurements. In addition, a multiple linear regression model was developed and calibrated for Sweden using data from the Integrated Carbon Observation System network. Seasonal and hourly variations in the PAR to Global Horizontal Irradiance (GHI) ratio were also analysed to understand their dynamic changes over time. The findings indicate that linear models using GHI as the primary predictor for PAR demonstrated high accuracy, with normalised Mean Absolute Error below 8% at all stations, with values such as 4% at Degerö and 3.2% at Norunda. Seasonal variability in the PAR to GHI ratio was observed, particularly during winter months at higher latitudes, where the ratio fluctuated between 0.39 and 0.42 at Degerö. In contrast, the summer period showed minimal variation, with the PAR/GHI ratio remaining stable across locations. Moreover, the spatial regression model, which combined data from different stations, successfully predicted PAR at new sites such as Norunda, achieving an R² of 0.98 to 0.99. Model residuals were within the typical uncertainty of PAR sensors (±5%), confirming remaining deviations are dominated by measurement error rather than modelling uncertainty. This demonstrates the model’s applicability across Sweden, providing a robust and versatile tool for estimating PAR in areas lacking measurements. The linear model reduces the need for extensive PAR measurement campaigns.