Estimating photosynthetically active radiation (PAR) is crucial for quantifying ecosystemproductivity through photosynthesis. In particular, providing continuous estimates of direct PAR (PARdir)and diffuse PAR (PARdif) is important for assessing ecosystem photosynthesis. The Breathing Earth SystemSimulator (BESS) has proven effective in estimating global PARdir and PARdif using an artificial neuralnetwork (ANN) that incorporates continuous satellite-based atmosphere products and elevation. However,BESS PAR does not account for topographic correction due to its relatively coarse resolution. The Perez-Driesse model is commonly used for topographic correction by partitioning incoming PAR into fivecomponents: PARdir, anisotropic PARdif, isotropic PARdif, horizontal brightness PARdif, and ground reflectedPARdif. Among these, isotropic PARdif and ground reflected PARdif are strongly influenced by the surroundingvisible sky and terrain fraction. However, the Perez-Driesse model oversimplifies the sky view factor (SVF)and ground-reflected PARdif calculations by considering only the slope at a given point. To address thisissue, we leveraged horizon angles from a digital elevation model (DEM) by scanning azimuth directionsfrom 0° to 360° at 3° intervals to more accurately estimate the SVF for isotropic PARdif, and we accountedfor adjacent terrain slopes and distances for ground-reflected PARdif. We applied this improved SVF Perez-Driesse model to BESS PAR estimates, which combine 2 km-resolution atmospheric products from theGEO-KOMPSAT-2A satellite with the 30 m-resolution Copernicus DEM. Validation with LI-190Rquantum sensor data from rooftop observations demonstrated significant improvements after applyingtopographic correction. For southward orientations, the coefficient of determination (R²) improved from0.52 to 0.78, relative bias (rbias) from 28.17% to –0.78%, mean absolute error (MAE) from 429.3 to 299.4,and root mean squared error (RMSE) from 553.3 to 378.7. For westward orientations, R² increased from–0.36 to 0.55, rbias from –44.63% to 23.80%, MAE from 477.3 to 243.7, and RMSE from 566.3 to 340.0.Analysis of two study sites with different slopes on the Korean Peninsula in 2021 showed that seasonalvariations in shadow ratios reduced PARdir by 40–80% under low solar positions. The rbias of PARdifwas influenced by increased cloud cover and aerosols during summer. The effect of terrain aspect waspronounced in instantaneous topographic corrections, particularly due to differences between aspect andsolar azimuth angle, whereas at the annual scale, north–south aspect differences were more evident. Ourresults showed that Perez-Driesse-based topographically corrected PAR can vary by up to 40% in ruggedterrain, whereas SVF Perez-Driesse-based PAR varies by nearly 20%, leading the significant variation inphotosynthesis estimation.



Keywords : Downwelling shortwave radiation, Photosynthetically active radiation, Topographic correction, Geostationary satellite, Artificial neural network