Inversion Of Leaf Area Index Based On Multi-anglular Information And Vegetation Index From Remote Sensing Data

Leaf Area Index(LAI)can effectively describe the characteristics of vegetation canopy structure,and is closely related to the energy and material exchange between ground surface and atmosphere.It is an important physical quantity for vegetation growth dynamic monitoring and regional/global environmental researches.Based on different remote sensing data,plenty of LAI products have been released.However,in practical application,the current LAI products have some common problems,which are embodied in:(1)insufficient spatial resolution.The spatial resolution of the current LAI products is between 0.5km to 5km,which are not suitable for researches in regions;(2)The accuracy of the current LAI product is insufficient.The accuracy of the current LAI products cannot meet the demand of "maximum error not exceeding max(0.5,20%)" proposed by global climate observing system(GCOS)in practical application;(3)Poor temporal and spatial continuity.Because remote sensing observations are affected by cloud,rain,atmosphere and satellite revisit cycle,most of the existing LAI time series products perform poorly in temporal and spatial continuity with lots of missing data.At present,there is still a lack of high-quality LAI products.With the development of satellite payload,remote sensing data are continuously optimized providing us higher spatial-temporal resolution data and richer surface spectral information.Research for effective use of the rich surface information provided by the new data,relieving the ill-posed problems in inversion,improving the performance of existing inversion algorithms,and increasing the accuracy and spatio-temporal continuity of LAI products is of great importance.This study focuses on solving the shortcomings of the current LAI inversion algorithms and products,the main research contents and conclusions are as follows:(1)LAI inversion algorithm synery use of multi-angle remote sensing information.The multi-angle features of remote sensing data,such as the angle information of multi-day observations,have not been effectively used in the current LAI inversion algorithms.The algorithm based on the strategy of "independent inversion and periodic synthesis" separates the multi-angle information from each other and cannot be used to constrain the inversion of LAI,which limits the accuracy of LAI products.In this study,the MODIS data within 8 days are used to construct the multi-angle observation set,and we optimize the cost equation in the current LAI inversion algorithm,designing multiple cost equations with different constraint levels to maximize the constraint of multi-angle observations on LAI inversion.In addition,based on the simulation data,the view angle combinations with different characteristics are analyzed from the aspects of inversion accuracy,uncertainty,distribution of possible solutions,saturation problem.The priori knowledge about the effectiveness of different angle combinations on LAI inversion is obtained,and the optimal view angle pair(OVAP)is determined.Based on the improved cost equation and the optimal angle combination,this paper develops a LAI inversion algorithm synery use of multi-angle remote sensing information.The results showed that this algorithm can effectively contrain LAI inversion by multi-angle data,and significantly reduce the amount of acceptable solutions and uncertainty of inversion;Compared with inversions based on single angle data,this algorithm can improve the inversion accuracy by nearly 50%,and can effectively improve the inversion saturation problem,and the saturation position is increased from 2.98 to 4.16;This algorithm effectively improves the overestimation of 8-day synthetic MODIS LAI products,and the inversion error is reduced from 0.79 to 0.71.(2)Research on LAI and LCC synergy inversion method based on vegetation index combinations.The existing algorithms input reflectance observations on "red-near infrared" bands ignoring the effective information in other bands that wastes abundant information of multi-bands remote sensing data and fails in well constraining LAI inversion.LAI and LCC have strong correlation,but they have great unconsistency when inverted independently.In this study,a collaborative inversion method of LAI and LCC is designed.The vegetation index(VI)is used to replace the "dual band" input of the existing algorithm.While effectively coordinating the multi-bands information,the clearer directivity of VI to canopy parameters can strengthen its constraint on parameter inversion,and has the potential to decouple the interaction of relevant parameters and improve the inversion accuracy.Based on the advantages of sentinel-2 multi-band data,a large number of vegetation indexes with different characteristics are calculated,and two-dimensional VI combination 2D-VIs and three-dimensional VI combination 3D-VIs...are constructed.Based on the simulated and measured spectra,the difference of VI combination with different characteristics on LAI/LCC collaborative inversion constraint ability and the correlation between VI combination dimension and inversion constraint are analyzed,and the a priori knowledge of the inversion effectiveness of different index combinations is obtained,The optimal VI index combination of Lai and LCC is determined.It is found that compared with red NIR band inversion,the optimal VI combination represented by MSAVI-Macc01,MSAVI-RERI705 and MTVI2-RERI705 can significantly enhance its constraint ability on LAI / LCC inversion and reduce the inversion uncertainty.The inversion accuracy of LAI and LCC is improved by about 19% and 53% respectively;The 3D-VI combination MSAVI-Macc01-MSR can greatly improve the LAI inversion accuracy on the simulated data set,and the inversion error is reduced by 58% compared with the2D-VI combination.(3)LAI spatiotemporal reconstruction algorithm coupling meteorological data and neural network model.The existing time series reconstruction algorithms based on filtering or function fitting can not reasonably reconstruct LAI time series with continuous missing values or missing values in key phenological periods.Due to the longer revisit period of high-resolution satellites and the serious impact of cloud,rain and atmosphere on remote sensing data,the reconstruction ability of existing algorithms for high-resolution LAI products is insufficient.In this study,a new LAI reconstruction algorithm coupled with meteorological data is developed,and a meteorological data driven back propagation neural network(MBPNN)is designed to reconstruct discontinuous LAI time series through a two-step process by using vegetation phenological information.Due to the strong dependence of vegetation growth on meteorological conditions,even if there are many missing observations,the algorithm can still ensure the reconstruction of the reasonable growth trajectory of LAI.Based on the measured ground data,the reconstructed LAI is verified in temporal and space.The results show that the predicted MBPNN LAI(RMSE =0.4083)has similar accuracy with the observed Landsat LAI(RMSE = 0.4076);Compared with the ground measured LAI time series,MBPNN LAI can maintain a growth trajectory similar to the measured data even when the data is missing for more than 100 days(RMSE = 0.1620);The comparison with time series harmonic analysis(HATS)algorithm shows that the reconstruction accuracy of MBPNN algorithm is more stable regardless of the length / location of missing data,and the reconstruction performance of this algorithm is better when the data is missing for 50 days or more.The methods proposed in this paper have certain theoretical significance and practical application value for further improving the current LAI product quality.