Mapping long-term and high-resolution global gridded photosynthetically active radiation using the ISCCP H-series cloud product and reanalysis data

: Photosynthetically active radiation (PAR) is a fundamental physiological variable for research in the ecological, agricultural, and global change fields. In this 23 study, we produced a 35-year (1984 ‒ 2018) high-resolution (3 h, 10 km) global gridded 24 PAR dataset using an effective physical-based model. The main inputs of the model 25 were the latest International Satellite Cloud Climatology Project (ISCCP) H-series 26 cloud products, MERRA-2 aerosol data, ERA5 surface routine variables, and MODIS 27 and CLARRA-2 albedo products. Our gridded PAR product was evaluated against 28 surface observations measured at seven experimental stations of the SURFace 29 RADiation budget network (SURFRAD), 42 experimental stations of the National 30 Ecological Observatory Network (NEON), and 38 experimental stations of the Chinese 31 Ecosystem Research Network (CERN). Instantaneous PAR was validated against 32 SURFRAD and NEON data; mean bias errors (MBE) and root mean square errors 33 (RMSE) were, on average, 5.8 W m -2 and 44.9 W m -2 , respectively, and correlation 34 coefficient ( R ) was 0.94 at the 10 km scale. When upscaled to 30 km, the errors were 35 markedly reduced. Daily PAR was validated against SURFRAD, NEON, and CERN 36 data, and the RMSEs were 13.2 W m -2 , 13.1 W m -2 , and 19.6 W m -2 , respectively at the 37 10 km scale. The RMSEs were slightly reduced when upscaled to 30 km. Compared 38 with the well-known global satellite-based PAR product of the Earth's Radiant Energy 39 System (CERES), our PAR product was found to be a more accurate dataset with higher 40 resolution. This new dataset is now


44
Plants rely on chlorophyll to absorb solar radiation in the visible wavelength range 45 (400-700 nm) for photosynthesis (Huang et al., 2020), and sunlight in this band is 46 commonly referred to as photosynthetically active radiation (PAR). Thus, PAR is the available. 94 Alternatively, satellite-based methods can be used to map spatially continuous 95 PAR, but compared to SSR, little attention has been paid to PAR estimation using  emerged to estimate PAR from regional to global scales with different satellite sources. 105 However, LUT-based methods are more vulnerable to various uncertainties due to their 106 "black-box" nature, and they are also difficult to port across different satellite platforms. 107 In contrast, parameterization methods do not rely on satellite platforms. 108 Essentially, they comprise a simplification of the radiative transfer processes, and thus 109 require various land and atmospheric products from satellite retrievals as inputs to 110 estimate PAR. To some extent, the accuracy of these methods depends on the accuracy 111 of the input data. On the other hand, the uncertainty of parameterization methods comes 112 mainly from the treatment of clouds; this is because the clear-sky part of the method is 113 relatively mature with uncertainty less than 10% compared with the rigorous radiative  global PAR products are either too coarse in spatial resolution to meet refined analyses, 133 too low in temporal resolution to reflect daily variations, or too short in time series to 134 meet the demand of climate change studies. As a result, a high-resolution long-term 135 global gridded PAR product is urgently needed in the scientific community.

136
In this study, a high-resolution 35-year global gridded PAR product was developed 137 using an effective physical PAR estimation model, driven mainly by the latest high-138 resolution ISCCP H-series cloud products, the aerosol product of the Modern-Era 139 Retrospective analysis for Research and Applications, Version 2 (MERRA-2) reanalysis 140 data, and water vapor, surface pressure, and ozone amount products of the ERA5 141 reanalysis data. We also evaluated the performance of our PAR product using in-situ 142 observations measured across three experimental observation networks in the United States and China, and compared its performance with another common global satellite 144 product. The rest of the article is organized as follows. In Section 2, we introduce the 145 method used to map the global gridded PAR product. The input data for estimating the 146 global gridded PAR product, and the in-situ data for evaluating the performance of our 147 estimated global gridded PAR product are described in Section 3. Section 4 presents 148 the validation results of our global gridded PAR product and compares this with the 149 well-known satellite-based global PAR product of CERES. Section 5 describes data 150 availability, and our summary and conclusions are given in Section 6. The algorithm used to map global gridded PAR in this study was the  The inputs of the PAR algorithm mainly include aerosol optical depth, cloud 164 optical depth, water vapor, ozone amount, surface albedo, and surface air pressure.   To produce a long-term (from 1984 to 2018) high-resolution global gridded PAR 181 product using the PAR algorithm presented above, we used input data from four 182 different sources.

183
The first source of input data was the latest level-2 H-series pixel-level global 184 (HXG) cloud products of the ISCCP, here referred to as ISCCP-HXG; these were 185 publicly available, spanned the period July 1983 to December 2018, had a spatial 186 resolution of 10 km, and a temporal resolution of 3 hours. The ISCCP-HXG cloud 187 products were produced by a series of cloud-related algorithms based on global gridded 188 two-channel radiance data (visible, 0.65 μm and infrared, 10.5 μm) merged from 189 different geostationary and polar orbiting meteorological satellites. We must bear in 190 mind that the 3-hour ISCCP-HXG cloud products denote instantaneous data at a given 191 moment every three hours, not a mean of 3 hours. We used four variables from the 192 ISCCP-HXG cloud products; these were cloud mask, cloud top temperature, and the AOD product against 793 AERONET stations worldwide, and also compared with 220 other aerosol products. It was found that the averaged RMSE for the MERRA-2 AOD 221 at 550 nm was about 0.126, which was generally lower than those of other aerosol 222 products.

223
The third source of input data was the routine weather variables of the ERA5 224 reanalysis data, which mainly included total column ozone, total column water vapor, 225 and surface pressure, with a spatial resolution of 25 km and a temporal resolution of 1 226 hour. Total column ozone and total column water vapor were used to calculate the 227 transmittance due to ozone absorption and water vapor absorption, respectively.

228
Surface pressure was used to calculated the Rayleigh scattering in the atmosphere. To 229 maintain consistency with the spatial resolution of the ISCCP-HXG cloud product, 230 these three routine weather variables of the ERA5 reanalysis data were re-sampled to 231 10 km.

232
The fourth source of input data was albedo data from the MODIS MCD43A3  its accuracy at 30 km spatial resolution is clearly higher than that of the CERES product.
324 Table 1 shows the accuracies of our estimated instantaneous PAR at different

335
The instantaneous PAR was also evaluated against the 42 NEON stations ( Figure   336 3 and Table 2). The performance against NEON was slightly worse than that against all scales from 10 km to 110 km. More importantly, the spatial resolution of our PAR 344 product (10 km) is much finer than that of the CERES PAR product (100 km).

345
Due to the significant improvement when our estimated PAR was upscaled to 30 346 km spatial resolution, we used a 3 × 3 spatial window to smooth the raw PAR to derive to ≥ 30 km, our daily PAR product performed slightly better than that of CERES. Validation results for our estimated daily PAR against in-situ data collected at 372 SURFRAD are shown in Figure 5 and respectively.

378
Validation results for our estimated daily PAR against NEON are shown in Figure   379 6 and fact that the quality of PAR observations at CERN is slightly worse than that at 407 SURFRAD and NEON, but further evidence is required to support this speculation. PAR is also shown. The spatial pattern of our ISCCP-ITP PAR product is quite 431 consistent with that of the CERES PAR product, whose spatial resolution was far 432 coarser than that of our PAR product. There were some finer patterns that the CERES 433 PAR product could not distinguish, but our PAR product could clearly capture. This  data to produce the final PAR product.

473
Our estimated PAR product was also compared with the CERES PAR product; we 474 found that the accuracy of our estimated PAR product at the original scale (10 km) was 475 generally comparable to, or higher than, that of the CERES PAR product. When it was 476 upscaled to ≥ 30 km, the accuracy advantage of our product over the CERES PAR 477 product became more evident. Another clear advantage of our PAR product was the 478 increased spatial resolution it offered compared to the CERES PAR product. We expect 479 that our PAR product will contribute to the future understanding and modeling of the 480 global carbon cycle and ecological processes. In future work, we will attempt to 481 separate the components of direct and diffuse PAR from the total PAR because light use 482 efficiency is mainly controlled by diffuse PAR.