Downscaling Photosynthetically Active Radiation Based on Two-Step Conversions With Consideration of Vegetation Forms

High spatial resolution photosynthetically active radiation (PAR) data are essential for carbon cycle studies. While the downscaling methods have been proven to be a reliable method to obtain fine-resolution data. However, the existing single-step downscaling suffers from the drawbacks of large spatial resolution span and lack of near-surface parameter texture details. Additionally, vegetation forms cannot be ignored in the PAR downscaling process, and the optimization of regression kernels remains an essential step to solve the data redundancy issue. Therefore, this study proposes a two-step downscaling model for PAR (PTDM) to estimate high spatial resolution PAR by considering vegetation forms with highly heterogeneous. The PTDM establishes regression windows based on vegetation forms to adapt to the vegetation heterogeneity and subsequently optimized the regression kernels by using a correlation matrix and importance ranking to construct a high-spatial-resolution PAR. Field verification results demonstrate that the correlation coefficient (R) exceeded 0.90, the bias was within 8 W/m2, the root-mean-square error (RMSE) was below 10 W/m2, and the mean absolute percentage error (MAPE) was below 14%. Compared with the existing downward shortwave radiation (DSR PAR) product, the proposed model achieves an R value greater than 0.65, a bias of less than −4 W/m2, an RMSE of under 7 W/m2, and an MAPE below 5%. Furthermore, the 30-m resolution PAR product generated by the model can capture finer spatial details, both for the overall study area and local plots. This study highlights the potential of the PTDM approach in producing high-precision PAR estimates, offering valuable insights for carbon cycle studies and ecosystem modeling.