Chlorophyll content is a crucial parameter for evaluating forest health and vegetation growth. It is an urgent to accurately estimate chlorophyll content at the regional scale with low cost by using remote sensing techno-logy. In this study, we took Xinping County, Yuxi City, Yunnan Province, as the research area, and used GEDI data as the main information source. Based on the empirical Bayesian Kriging regression prediction (EBKRP) method, we accurately obtained the continuous distribution of the spot characteristic parameters in the unknown space of the study area. Combined with measured data of 52 plots, we used Pearson correlation, random forest (RF) and gradient boosting regression tree (GBRT) to screen the optimal combination parameters. We further established the best estimation model of chlorophyll content of Dendrocalamus giganteus at regional scale by Random forest regression (RFR) and GBRT models. The results showed that EBKRP demonstrated high prediction accuracy and reliability, with R2 values ranging from 0.34 to 0.99, RMSE from 0.012 to 3134.005, rRMSE from 0.011 to 0.854, and CRPS from 965.492 to 1626.887. Different parameter optimization methods yielded slightly different optimal para-meter combinations. Different remote sensing modeling methods showed varying accuracy levels. The GBRT model (R2=0.94, RMSE=0.132, P=91.2%) outperformed the RFR model (R2=0.89, RMSE=0.192, P=89.3%). Using the GBRT model for estimating and mapping the spatial distribution of D. giganteus chlorophyll content, which ranged from 0.22 to 2.32 g·m-2, with an average of 1.36 g·m-2. These results aligned with the actual D. giganteus distribution in the study area, indicating that the GBRT model using GEDI data optimized by EBKRP could be feasible and reliable for estimating forest biochemical parameters, thereby providing effective support for forest health monitoring.
叶绿素含量是评价森林健康状况和植被生长状况的重要参数,利用遥感技术以低成本准确估测区域尺度叶绿素含量是当前亟待解决的问题。本研究以云南省玉溪市新平县为研究区,以GEDI数据为主要信息源,基于经验贝叶斯克里金回归预测(EBKRP)法准确获取研究区内光斑特征参数在未知空间的连续分布,结合52块样地实测数据,采用Pearson、随机森林(RF)、梯度提升回归树(GBRT)3种方法筛选最优组合参数,利用随机森林回归(RFR)模型和GBRT模型建立区域尺度龙竹叶绿素含量最佳估测模型。结果表明:EBKRP法预测精度高,结果可靠,R2值在0.34～0.99,RMSE值在0.012～3134.005,rRMSE值在0.011～0.854,CRPS在965.492～1626.887。参数选优方法不同,选出的最佳组合参数略有差异。遥感建模方法不同,构建的模型精度存在差异,利用GBRT模型构建的遥感估测模型(R2=0.94,RMSE=0.132,P=91.2%),其性能优于 RFR模型(R2=0.89,RMSE=0.192,P=89.3%);选用GBRT模型估测研究区龙竹叶绿素含量,其分布范围为0.22～2.32 g·m-2,平均叶绿素含量为1.36 g·m-2,此研究结果与研究区龙竹分布具有一致性,说明基于EBKRP法优化后的GEDI数据,选用GBRT模型进行森林生化参数估测具有可行性,结果可靠,可为森林健康监测提供有效支持与服务。.
Keywords: EBKRP; GEDI; chlorophyll content; combination parameter optimization; estimation; machine learning.