Research On Spatiotemporal Fusion Of Multi-source Remote Sensing LAI Data Based On Local Kernel Regression And ESTARFM Mode

Leaf Area Index(LAI)is an important parameter for describing the structure of vegetation canopy,However,the quality of LAI remote sensing products is often compromised by various factors,such as clouds,aerosols,snow,algorithms and instrumentation problems,resulting in poor quality or missing data,which seriously affects the application of LAI datasets.Therefore,higher spatial and temporal resolution of LAI distribution information is required for modelling surface and ecosystem processes at regional scales and smaller scales,especially in areas with complex and heterogeneous topography.This paper proposed a new fusion method for LAI time-series data based on improved S-G filtering and unsupervised classification local kernel regression,and conducted a normalized fusion study in the Chinese region using MODIS LAI,PROBA-V LAI,and VIIRS LAI product data from 2014—2018 to improve the consistency,continuity and accuracy.Based on this,the Heihe River Basin was used as the study area,and the 2014 normalized fused LAI time-series data(8 days,500m)and the Heihe River Basin monthly synthetic dataset(1 month,30m)were used,based on the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model(ESTARFM)to generate higher spatial and temporal resolution LAI data(8 days,30m).The main findings are as follows:(1)The LAI fusion data obtained by using the improved S-G filtering and unsupervised classification based local kernel regression method proposed in this paper had good continuity and high accuracy.The fusion result showed good agreement with the frequency distribution and temporal variation of the LAI values of the source product data and other LAI product data(MCD15A2H,MOD15A2H,VNP15A2H,PROBA-V),and had good correlation with R~2of0.83,0.74,0.82 and 0.89 respectively.The frequency of missing data generally decreased and Temporal continuity improved,with a reduced average missing frequency of 4.66%for the normalised fusion LAI compared to the MCD15A2H LAI(19.50%),MOD15A2H LAI(25.22%),VNP15A2H LAI(23.37%)and PROBA-V LAI(8.77%).Compared with other product data,the normalised fusion LAI had the best correlation with ground truth values,with a coefficient of determination(R~2)of 0.76,0.03-0.2 higher than other product data,and a root mean square error(RMSE)of 1.16m~2/m~2,0.1-0.66m~2/m~2 lower than other product data,providing a high degree of accuracy.(2)The higher spatio-temporal resolution LAI data(8 days,30m)generated by fusion in areas of high heterogeneity,based on the ESTARFM model,had good accuracy.The spatial distribution of the ESTARFM-spatio-temporal fusion LAI(8 days,30m)and the input low spatial resolution data-normalised fusion LAI,the details were basically the same,with improved spatial resolution,clearer texture structure,and richer spatial information.The accuracy validation using the 2014 Heihe River Basin measured LAI dataset showed that the coefficient of determination of ESTARFM-temporal fusion LAI with the measured values was R~2=0.74 and the root mean square error was RMSE=0.85m~2/m~2.(3)The time series curves of regional normalised fusion LAI in China from 2014 to 2018are relatively smooth,with generally similar trends,intra-annual variation with a single-peaked distribution and small inter-annual variation,with a multi-year average value of 1.22 m~2/m~2.There were differences in the trends of annual-scale LAI values for different land cover types,with deciduous coniferous forest LAI values showing a fluctuating downward trend and other land cover types showing an upward trend.All other land cover types showed an increasing trend in LAI values.There was a clear seasonal variation and a more consistent trend in LAI for different land cover types.Spatially,the spatial distribution of the normalised fused LAI values in China showed an increasing pattern from north-west to south-east,with a single-peaked distribution in the growing season(April-September)in the Heihe River Basin in 2014,with a maximum value of 4.38 m~2/m~2.The low value areas were mainly located in the lower reaches of the Black River,while the high value areas are mainly located in the upper and middle reaches of the Black River.