森林植被碳密度遥感反演和校准研究

Remote Sensing Retrieval and Calibration of Forest Vegetation Carbon Density Based on Time-series Data

采用2002—2017年间的地面监测数据和每年3月份的遥感数据,建立参数模型和BP人工神经网络模型,并进行动态分析;采用2002年不同月度、季度数据以及不同年度数据,分析遥感数据时相和估算尺度对反演精度的影响。结果表明:生物量密度、碳储量密度与遥感指标间的线性关系较弱,Pearson相关系数均小于0.60,不同时期生物量密度、碳储量密度存在显著性差异。相较于参数模型,采用BP人工神经网络生物量密度拟合决定系数平均提高了0.0705,碳密度拟合决定系数平均提高了0.0762。辅助GPS精准定位进行时序数据校准,2002—2017年期间四川省森林植被生物量密度、碳密度呈不断增大的趋势,两者动态变化规律具有一致性。遥感数据反演生物量密度和碳密度均存在明显的高估和低估区间,季度数据和年度数据均没有月度数据的估计效果好,不同估算尺度对遥感反演的相对误差具有显著性的影响。

Based on the continuous forest inventory data and remote sensing data from 2002 to 2017,parameter models and BP artificial neural network models were established,and the dynamic analysis was carried out.The effects of time phase and estimation scale of remote sensing data on the inversion accuracy were analyzed by using different monthly,quarterly and annual data in 2002.The results showed that the linear relationship between biomass density,carbon storage density and remote sensing indexes was weak,with Pearson correlation coefficient less than 0.60,and there were significant differences between biomass density and carbon storage density in different periods.The fitting determination coefficient of biomass density and carbon density by BP artificial neural network models increased by 0.0705 and 0.0762 on average compared with the parameter models.During the period of 2002 to 2017,the biomass density and carbon density of forest vegetation in Sichuan Province showed an increasing trend assisting GPS precise positioning for timing data calibration,and the dynamic change law of the two was consistent.There were obvious periods of overestimation and underestimation of biomass density and carbon density in remote sensing data inversion,and the estimation effect of quarterly data and annual data was not as good as that of monthly data.Different estimation scales had significant influence on the relative error of remote sensing data inversion.