Global leaf inclination angle (LIA) and nadir leaf projection function (G(0)) products

Leaf inclination angle (LIA), the angle between leaf surface normal and zenith directions, is a vital parameter in radiative transfer, rainfall interception, evapotranspiration, photosynthesis, and hydrological processes. In the radiative transfer regime, LIA is generally represented by the leaf projection function (G(θ)), which is defined as the average projection ratio of unit leaf area in the illumination or viewing direction θ.

This dataset (CAS-GLA) includes the first global mean LIA (MLA) and nadir leaf projection function (G(0)) products at the spatial resolutions of 500 m and 0.05 degrees. The products are recorded in GeoTIFF format under the WGS-84 geographic coordinate system. The global 500 m MLA product was generated by gap-filling LIA measurement data using a random forest regressor after a series of preprocessing, such as spatial expansion, LIA upscaling, and sample screening. Cross-validation shows that the predicted MLA presents a medium consistency (r = 0.75, RMSE = 7.15°) with the validation samples. The 500 m MLA product was further upscaled to 0.05 degrees by weighting the MODIS 500 m leaf area index. The G(0) product was derived from the MLA by assuming a single-parameter ellipsoidal leaf angle distribution. The G(0) product agrees moderately with high-resolution reference data (r = 0.62, RMSE = 0.15). Note the global MLA and G(0) products mainly represent the typical state during the growing season from 2001 to 2022. The MLA and G(0) products would enhance our knowledge about global LIA and should greatly facilitate remote sensing retrieval and land surface modeling studies.

In the version 1.1, two quality layers were added to 500 m MLA product to represent the quality of input data and the prediction model. The input data quality was denoted by the proportion of high-quality BRDF inversions for each pixel. The prediction model quality was represented qualitatively for each pixel considering whether the MLA was predicted by extrapolating beyond the range of the training samples. In the 500 m MLA product, the three bands sequentially store MLA, input data quality (scale factor = 0.01), and model quality (1: predictions within the range of samples; 0: predictions out of the range of samples).