Spatio-temporal Continuous Vegetation Parameter Inversion Methodology Over Complex Surface Based On Remote Sensing

Vegetation is the general term of the plant community on the earth surface,which is an important part of the ecological environment.Accurate acquisition of vegetation parameters and generation of vegetation products are very important in global change studies.Spatial heterogeneity and topography are the significant characteristics of complex land surface,which have significant influences on the surface radiation transfer and parameter inversion process,and are the bottleneck limiting the improvement of inversion accuracy of vegetation parameter products.Existing vegetation parameter inversion algorithms are usually established for scenes composed of a single uniform surface.However,due to the lack of consideration of surface heterogeneity and topographic features in the inversion process,the accuracy of existing products is rather low in complex surface areas.With the improvement of computing power,it is possible to simulate the radiative transfer process of complex surface with computer simulation models.At the same time,the emergence of high-resolution landcover data provides data support for describing complex surface scene structure,and also provides new ideas for vegetation parameter inversion in heterogeneous surface.Focusing on the objective of continuous vegetation parameter inversion over complex land surface,this paper firstly parameterized and quantitatively analyzed the features of complex land surface,then analyzed the topographic effects in LAI inversion and proposed a LAI inversion method for mountainous areas,finally developed a climate integrated method to generate spatiotemporal continuous vegetation parameter products.The main research contents and conclusions are as follows:(1)The landcover heterogeneity of global surface and the coupling relationship between vegetation and topography were analyzed based on high resolution data.Based on a 30 m land cover data-Globe Land30,a parameterization scheme is proposed to quantitatively describe the heterogeneity characteristics of mixed pixels.The number of endmember types and the area ratio of each endmember within a pixel scene are used to describe the composition feature of the pixel.The canopy height levels of different vegetation types were defined to describe the boundary characteristics of patch mosaicking within the pixel and to describe the fragmentation degree of mixed pixels.Global distribution of endmember numbers and boundary length on the 1km scale were produced.Based on 30 m resolution DEM data and landcover data,vegetation distribution characteristics under different topographic features were analyzed.The following conclusions are drawn from the analysis of the global complex surface features:On the 1km scale,the heterogeneity caused by land cover mixture is widespread globally.Only 35%/25.8% of the pixels in the land/vegetation area are covered by a single surface type,i.e.,pure pixels.Most mixed pixels are composed of multiple different vegetation types,accounting for 64.0% of all pixels in the global vegetation areas.Mixed pixels are more common in the ecological transition zone.The surface of the mixed pixels with a larger number of endmembers is usually more fragmented,but the mixed pixels composed of two endmembers may also be very fragmented,e.g.,boreal forests.The intra-heterogeneity degree of biomes from high to low was: savanna,deciduous needleleaf forest,evergreen needleleaf forest,deciduous broadleaf forest,shrub,broadleaf crops,grassland/cereal crops,and evergreen broadleaf forest.More than 16% of the world's surface area has a slope greater than 15°.In mountain areas,the proportion of forest and grassland is much higher than that of other vegetation types,and forests are even more dominant.There is no obvious difference in the distribution of different vegetation types on the aspects on a global scale.(2)The topographic effect of LAI inversion was analyzed based on the computer simulation model-DART model.In this paper,the DART model was used to simulate the red and near-infrared reflectance data of sloped continuous vegetation with different parameter settings(including leaf parameters,canopy structure parameters,soil background parameters and topographic parameters,etc.),and to quantitatively analyze the influence of topography on slope LAI inversion.Firstly,an artificial neural network(ANN)model was used to construct LAI inversion algorithm without considering the influence of terrain.Then,the simulated slope reflectance was input into the ANN model to obtain biased LAI inversion values for analyzing LAI topographic errors.The results show that the terrain error of LAI inversion increases linearly with the increase of the slope,and the average relative deviation can reach 51% when the slope is 60°.The topographic error of LAI inversion is related to canopy density.Except for sparse canopy,the existence of terrain usually leads to lower LAI inversion results.LAI inversion error is closely related to local incident angle as aspects changes.In dense canopies,the inversion value of LAI gradually decreases from the sunny to the shady slope,while the law is opposite in the sparse canopy.This is because in the sparse canopy scene,the terrain effect of red band reflectance is stronger,while in the dense canopy,the red band is close to saturation,and the near-infrared band is more affected by the terrain.The results of this study contribute to a better understanding of the topographic effect of LAI inversion and provide a better strategy for mountain LAI inversion.(3)Based on random forest model and DART model,a mountain LAI inversion method was proposed.In this study,the DART model was used as the benchmark to simulate the forest canopy reflectance using different leaf parameters,canopy structure parameters,soil background parameters and topographic parameters.Spectral response functions of different sensors are used to obtain the reflectance of specific bands of different sensors.Based on the generated reflectance dataset,the random forest model is used to train the mountain LAI inversion model.Then,the reflectance data,geometry parameters and topographic parameters of remote sensing images were input to obtain LAI inversion results over mountainous areas.Simulation data and LAI reference map of Wanglang area in Sichuan were used to verify this algorithm and compared with existing LAI inversion methods(3DRT lookup table)and mountain LAI inversion strategies.The results show that the proposed algorithm can correct LAI inversion errors under different terrain features.The inversion accuracy of this algorithm is(RMSE: 0.973;R2:0.508)was superior to existing mountain LAI inversion strategies(geometric parameter correction strategy(RMSE: 1.465;R2:0.193)and reflectance terrain correction strategy(RMSE: 1.720;R2:0.278)),which has great potential in mountain forest LAI inversion.(4)The climate integrated gap filling method is proposed for the spatial temporal reconstruction of vegetation parameters.This research proposed a method to reconstruct the NDVI time series coupled with climatic factors(sun downward radiation,precipitation and temperature).Based on the over 40 years of Landsat-5,Landsat-7 and Landsat-8 observation in the Qilian Mountains region,the spatiotemporal continuous 8-day NDVI products were obtained.Taking Sentinel-2 data as reference,the performance of this method and traditional time series reconstruction methods(such as HANTS and SG filtering)on reconstructing Landsat data are compared and analyzed.Results show that the method has similar accuracy with the mainstream approaches(RMSE close to 0.066)when gaps is not significant,and when the gap ratio is more than 50%,the reconstruction ratio and accuracy of the proposed(RI: 99.9%;RMSE: 0.089)were significantly higher than that of traditional time series reconstruction method(RI: an average of 45.3%;RMSE: 0.178).The reconstruction results of this method can provide effective data support for long time series vegetation analysis in the study area.