Generating High Spatio-temporal Resolution LAI Data Based On A Modified FSDAF Model

With the development of remote sensing technology,multi-platform and multi-type remote sensing images have become the key data for monitoring and analyzing land surface conditions and dynamic changes from regional to global scales.Leaf area index(LAI)is one of the important parameters characterizing the structural characteristics and biophysical processes of vegetation canopy,and is also considered as an important variable in carbon and water cycles and energy exchange in terrestrial ecosystems.Accurate information on LAI is of great significance for the study of global change.Due to the constraints of cloud pollution,seasonal snowfall and instrument conditions,no existing sensors can both take into account high spatial and temporal resolution,thereby there is a problem of insufficientacquisition capacity of remote sensing data.At present,most of satellite-based leaf area index products are derived from high temporal-low spatial resolution data,with serious problem of mixed pixels,which restricts the study of vegetation change in highly heterogeneous areas.On the other hand,there is no LAI product with medium and high spatial resolution,and the temporal resolution is relatively low,which makes it difficult to meet the needs of dynamic tracking of vegetation growth process.Therefore,it is of great significance to develop the spatial-temporal data fusion algorithm to simulate and generate high spatial-temporal resolution data,combining the advantages of different sensors.It will meet the demands for remote sensing data due to the deepening of remote sensing application,and provide an effective solution to alleviate the spatial-temporal contradiction of remote sensing data.In this paper,an improved spatio-temporal data fusion model is proposed,which combines the pixel decomposition downscaling method with the flexible spatio-temporal data fusion model(FSDAF).Based on MODIS product data and Landsat 8 OLI data,high spatial and temporal resolution leaf area index data are generated.Firstly,based on high-quality pure pixel of MODIS leaf area index and reflectance data,and combined with support vector regression algorithm,the LAI inversion model is constructed and applied to Landsat 8 OLI reflectance data estimation to obtain Landsat scale LAI.Secondly,MODIS LAI data are downscaled to30 m resolution by means of the mixed pixel decomposition,and the accuracy of input data of the data fusion model has been improved.Finally,using the modified FSDAF model,8-day interval LAI data with a resolution of 30 m are generated by blending theMODIS LAI data and the Landsat retrieval LAI data,and the change and distribution of the cultivated land leaf area index with the growth season are analyzed.The accuracy of the fusion results are validated using the Sentinel-2A LAI data,and compared with the simulated LAI by FSDAF and STARFM.In this paper,two sample plots in Songyuan city,Jilin province,China are selected as the research area for experimental analysis and validation.The results show that :(1)downscaling the MODIS input data based on the linear spectral mixing model can improve the accuracy of data fusion by replacing the low-resolution resampled data in FSDAF;(2)the correlation coefficients between predicted LAI and Sentinel-2A LAI are 0.83 and 0.78,with the Root-Mean-Square Error of 0.57 and0.44,respectively,which are higher than other two methods,namely FSDAF(0.79 and 0.67)and of STARFM(0.54 and 0.52),indicating that the modified FSDAF model has higher accuracy;(3)the predicted LAI images using modified FSDAF model show better spatial details,and the blurring of the boundaries among different ground objects on remote sensing images can be improved to a certain extent.The modified FSDAF model can be applied to the highly heterogeneous regions to generate leaf area index data,which has the potential to be extended to predict the time series data of other biophysical variables.