中高分辨率遥感协同反演冬小麦覆盖度

为了开展高精度、高时空分辨率的植被覆盖度(fraction vegetation cover,FVC)监测,该文以华北地区冬小麦地为研究对象,采用4期高分一号卫星多光谱(GF1-PMS)、多光谱宽幅(GF1-WFV)与环境一号卫星多光谱(HJ1-CCD)3种传感器同期影像数据集,基于像元二分法模型,研究多源中高分辨率遥感影像协同估算FVC方法.以基于高空间分辨率GF1-PMS影像反演的FVC作为检验数据,对单源直接获取法、多源全生育期法、多源分期法3种反演模型进行了分析比较.研究结果表明:HJ1-CCD、GF1-WFV数据与GF1-PMS数据的FVC直接反演结果具有较高的一致性,但在冬小麦的初期生长阶段,受卫星观测角度效应的影响,GF1-WFV与HJ1-CCD的FVC结果偏高,偏差随冬小麦的成熟封垄而逐渐减弱;多源分期法的时空反演得到的FVC精度最高,GF1-WFV的决定系数为0.984,均方根误差为0.030;HJ1-CCD的决定系数为0.978,均方根误差为0.034;而在缺少GF1-PMS匹配数据时,可通过多源全生育期法提高GF1-WFV与HJ1-CCD数据的反演精度,GF1-WFV的决定系数为0.964,均方根误差为0.044;HJ1-CCD的决定系数为0.950,均方根误差为0.052.通过多传感器的联合反演获取时间序列的高精度的FVC数据,可为研究植被生长状况及生态环境动态变化提供数据基础.

Coordination inversion methods for vegetation cover of winter wheat by multi-source satellite images

Abstract: Fraction vegetation cover(FVC)can be used to indicate the growing status of vegetation, which is an important input for some ecological models, hydrological models, meteorological models,and so on. And FVC data set with high precision, hightemporalresolution, and highspatial resolution is critical to global change monitoring. Unfortunately, current FVC products are produced using only one kind of remote sensing image, and thus their spatial coverage and temporal coverage are limited. Aiming at acquiringcontinuous FVC data in space and time, we explored the estimation methods of FVC of winter wheat in North China Plainusing high and medium resolution images jointly. This study focused on dimidiate pixel model by combining multi-source images includingGF1-PMSimages with spatial resolution of 8m, GF1-WFVwithspatial resolution of 16m, and HJ1-CCD withspatial resolution of 30m. Fourphases ofremote sensing images of those 3sensors were selected as data source to conduct the experiments, which covered 4growth periods of the winter wheat, includingturning green &rising stage(March 23, 2015 and March 29,2015) and jointing&flowering stage(April 28, 2014 and May 5, 2014).Within the coincidence regions of those 3 kinds of images, we selected randomly 160 winter wheat sample areas (240 m×240 m) as the regression samples, and chose randomly another 80 winter wheat sampleareas(240 m×240 m) as the checking samples to verify the performanceof the methods.Usingthese regression samples, we developed multi-source whole-growth-period method (MWM) and multi-source single-growth-period method (MSM) based on the bottom-up method.We compared and analyzedthe single-source inversion method (SIM), MWM and MSM based on the estimated FVC result using high spatial resolution GF1-PMS images. The results indicated that the FVC estimationsof HJ1-CCD, and GF1-WFV imagesusing SIM method werehighly consistentwiththose of GF1-PMS images, and their R2values wereboth higher than 0.9. However, due to theobservationangle effectof GF1-WFV and HJ1-CCD sensors, the estimated FVCswerea little higher in the early growing stages of winter wheat, and the bias decreased gradually with the closing of winter wheat canopy. Compared with SIM method, MWM method and MSM method bothworked more effectively and generated higher accuracy. Among those twomulti-source methods, MSM method showed the relatively higher accuracy, and itsdeterminant coefficients R2was 0.984 and the root mean square error(RMSE)was 0.030 using GF1-WFV images, while the R2was 0.978 and the RMSE was 0.034 using HJ1-CCD images. The R2 of MWM method was0.964 and the RMSE was 0.044 using GF1-WFV images, and the R2 was 0.950 and the RMSE was0.052 using HJ1-CCD images. Comparison indicatedthat MWM can be utilized to improve the FVC estimation accuracy using GF1-WFV and HJ1-CCD images when there are no matching GF1-PMS images over the same period. Thisresearch shows that the synergeticinversion method of winter wheat FVC with multi-source satellite images can generatelong time series and high precision FVC products, which can provide the critical data setfor vegetation growth monitoring,monitoring of ecological environmentand global change detection.