ABSTRACT

We developed a novel method for retrieving land-surface albedo (LSA) based on 338 groups of meteorological and NPP-VIIRS albedo data. Results showed that the LSA retrieval model calibrated using therandom forest (RF) machine-learning regression algorithm performed well. The RF-based LSA retrieval model explained approximately 84% of the NPP-VIIRS albedo variation (R² = 0.8429; N = 236) and yielded satisfactory validation accuracy (RMSE = 0.0167; MRE = 15.82%; N = 102). The original and retrieved albedo were also compared with the MODIS albedo product, which further validated the accuracy and effectiveness of the approach (RMSE = 0.0091; RME = 8.2%).

摘要

我们开发了一种基于 338 组气象和 NPP-VIIRS 反照率数据反演地表反照率 (LSA) 的新方法。结果表明，使用随机森林 （RF）机器学习回归算法校准的LSA检索模型表现良好。基于 RF 的 LSA 检索模型解释了大约 84% 的 NPP-VIIRS 反照率变化（R ²  = 0.8429；N = 236），并产生了令人满意的验证精度（RMSE = 0.0167；MRE = 15.82%；N = 102）。原始反照率和反照率与MODIS反照率产品进行了比较，进一步验证了该方法的准确性和有效性（RMSE = 0.0091；RME = 8.2%）。