基于ADC和MODIS遥感数据的高寒草地地上生物量监测研究——以黄河源区为例

摘要

We collected grassland biomass and MODIS satellite remote sensing data,and calculated vegetation indices (VIs) from data obtained by an agricultural digital camera (ADC) in the Yellow River Headwater Region (YRHR) in August of 2015-2016.We explored the correlations between each of three ADC vegetation indices (NDVIADC,SAVIADC,and GNDVIADC) and field-measured grassland above-ground biomass (AGB),and selected the optimal ADC vegetation index to construct an AGB inversion model.Grassland AGB inversion models based on ADC vegetation indices and MODIS NDVI (denoted as NDVIMOD) were constructed,and the accuracy of each model was evaluated by leave-one-out cross validation (LOOCV) to identify the optimal grassland AGB monitoring model.The NDVIADC was used to correct the NDVIMOD to obtain the optimized grassland AGB model with high resolution and accuracy.The results showed that:1) among the three VIs-based ADC indices,the NDVIADC was most closely related to the AGB of alpine grassland,followed by SAVIADC and GND-VIADC.2) The NDVIADC-based AGB monitoring model (RMSEP:383.55-393.18 kg DW/ha;r:0.65-0.66)was more accurate than the NDVIMOD model (RMSEP:421.08-427.00 kg DW/ha;r:0.55-0.58).Therefore,the grassland AGB inversion value from the NDVIADC-based model was much closer to the actual grassland AGB in YRHR,and the sampling values of NDVIMoD were higher than those of NDVIADC as a whole.3) Among the four models based on NDVIADC,the linear and power models showed better performance in grassland AGB simulations.The linear model (y=3248.93XNDVIADC-305.59,RMSEP=383.55 kg DW/ha,r=0.66) was more accurate than the power model,and the linear model provided a novel and simple method to estimate grassland biomass in the study area.4) There was a strong correlation between NDVIADC and NDVIMOD;therefore,we could obtain an optimized grassland AGB model by using NDVIADC to correct NDVIMOD.The optimized model was y=2455.54XNDVIMoD-301.69.This model could be used to estimate the grassland biomass in YRHR on a large scale,and its precision was close to that of the field measurements.%利用2015-2016年8月采集的黄河源区草地生物量数据和MODIS卫星遥感资料,结合农业多光谱相机(agricultural digital camera,ADC)获取的植被指数数据,比较分析3种ADC植被指数(NDVIADC、SAVIAm和GND-VIADC)与野外实测草地地上生物量(above-ground biomass,AGB)数据的相关性,筛选出适合构建草地AGB反演模型的ADC植被指数;结合MODIS NDVI(记作NDVIMoD)构建草地地上生物量反演模型,采用留一法交叉验证方法评价各模型精度,确立适宜模拟研究区草地AGB的最优模型;并利用NDVIADC校正NDVIMOD,获得高分辨率、高精度的草地AGB遥感监测改进模型.结果表明,1)基于ADC获取的3种植被指数中,NDVIADC与高寒草地地上生物量关系最为密切,其次为SAVIADC,拟合效果最差的是GNDVIADC;2)基于NDVIADC建立的草地AGB监测模型的精度(RMSEP介于383.55～393.18 kg DW/hm2;r范围为0.65～0.66)远高于NDVIMOD的模型精度(RMSEP介于421.08～427.00 kg DW/hm2;r范围为0.55～0.58),NDVIADC反演得到的草地AGB更接近于黄河源区草地实际生物量,且相较于NDVIADC,NDVIMoD的样本值整体偏高;3)在NDVIADC构建的4类模型中,线性和乘幂模型模拟研究区草地AGB的能力较好,但线性模型精度更高(y=3248.93×NDVIADC-305.59,RMSEP=383.55kg DW/hm2,r=0.66),该模型为黄河源区草地生物量的估测提供了一个新型且易操作的方法;4)NDVIADC与NDVIMOD相关性较高,利用NDVIADC校正NDVIMOD可以改进草地AGB遥感反演模型,优化模型为y=2455.54×NDVIMoD一301.69.该模型可在大尺度范围内估测黄河源区的草地生物量,且模型精度接近于地表测量法的监测精度.