Accuracy Evaluation And Model Improvement Of The MODIS Bi-directional Reflectance Model Product For Snow-covered Forest Area

The northern forest is the largest terrestrial ecosystem on the Earth,with annual snow cover covering 6 to 9 months.Forest snow plays a vital role in the global water cycle,surface radiation energy budget and climate change,and is a valuable source of water for more than one billion people worldwide.Multi-angle remote sensing technology based on the theory of electromagnetic wave anisotropic reflectance plays an increasingly important role in global forest remote sensing.Researchers have established a series of bi-directional reflectance models for forest areas,which are widely used in a range of applications and researches during the vegetation growing season.However,in the dormant season of vegetation,the phenological changes of forest vegetation and the occurrence of snow cover make the reflectance characteristics of forest ecosystems different from the growing season.The existing remote sensing reflectance anisotropy model and data products in the vegetation dormant season,especially in the snow-covered forest area,and the research and modeling of the anisotropic reflectance characteristics of the forest in this season need to be strengthened and improved.At present,the surface reflectance anisotropy models used to retrieve the land surface Bi-directional Reflectance Distribution Function(BRDF)/albedo based on Moderate Resolution Imaging Spectro-radiometer(MODIS)data are not directly considering the influence of winter snow in the northern forest area.This makes the BRDF products in the winter of northern forest areas have greater uncertainty and less amount of full-retrieval pixels than those in vegetation growing season.Focusing on this scientific problem,the paper takes the MODIS BRDF model product as the research object,and the Multi-angle Imaging Spectro-Radiometer(MISR)land-surface(LS)Bi-directional Reflectance(BRF)product(MILS_BRF)as the reference data,and carries out the research of preprocessing methods on two types of multi-angle remote sensing data.On this basis,the effects of phenological changes and snow on the characterizing ability and accuracy of the surface anisotropic reflection characteristics of MODIS BRDF model products were evaluated.Finally,the improved method of the MODIS BRDF model and the inversion strategy in the snow-covered forest area are studied.The preprocessing of multi-angle remote sensing data is the basis for the accuracy evaluation and improvement of BRDF model products.Specific to the research objectives of this thesis,the MODIS BRDF products(MCD43A1)is the object of accuracy evaluation and improvement,and the MILS_BRF is the reference data for evaluation,that is,the MILS_BRF is assumed to represent the true BRF value of the land surface.The projections of both MCD43A1 and MILS_BRF are two different and less common projection,and the nominal resolutions of the two are also different.In order to properly use the MILS_BRF to evaluate the accuracy of the MCD43AI and improve the MCD43A1 data projection time and efficiency,it is necessary to preprocess the two data.The pre-processing work includes spatially accurate co_registration of MILS_BRF and MCD43A1 data and MCD43A1 fast projection transformation.Because of the insufficiency for existing software to deal with the data space registration problem of MILS_BRF and MCD43A1,an accurate spatial co-registration method for multi-angle multi-source remote sensing data based on self-defined projection grids is proposed.The proposed method solves the problem that MISR L2 product data can not be directly and accurately registered with other remote sensing data by using common remote sensing data processing softwares,which only resamples once and can avoid the possible information loss caused by repeated resamples.The basic ideas and objectives of the proposed method are also applicable to the spatial accurate co-registration of other similar multi-angle remote sensing data.This study also proposes a practical fast reprojection method for sinusoidal(SIN)projection data that includes 3 major features.(1)Only pixels in target projection space that have corresponding points in MODIS products("effective pixels")are further processed.(2)Tags of effective pixels and their corresponding points in MODIS products are retrieved and stored by 1-dimensional(1-D)index of image matrix rather than 2-dimensional subscripts.(3)Pixel values in target images are assigned from MODIS products in bulk rather than in a loop procedure that assigns pixels one by one.In a comparative analysis,the proposed method performs better than the other three tested methods.Based on these improvements,use of prestored 1-D indices of corresponding pixel pairs significantly speeds reprojection.It therefore shows the potential for batch processing of large volume of MODIS products with a SIN projection in reasonable time.The study evaluates the performance of the MCD43A1 C6 on different observation geometries based on MILS BRF,and to preliminarily analyze phenological phase and snowfall event impacts on the accuracy of the product.These works show that the RTLSR model has a wide range of adaptability,MCD43A1 product has good consistency for a long period,and accuracy of the product can meet the requirements of many applications and researches.Most of these efforts focus on the growing season of vegetation.In contrast,validation work on the dormant season and snow cover still need more attention.The performance of the MCD43 product was specifically studied during dormant season and snow cover.However,these researches mainly presented the results about albedo,instead of accuracy of BRF on different observation geometries.In this study,a typical region in the central part of Northeast Asia is selected as the study area and the performance of MCD43A1 C6 BRDF model is analysed in various observation geometries and phenological phases,using the MILS_BRF as the reference data.In addition,the impacts of land cover types and snow covers on the model accuracy are evaluated using MODIS land cover type product and snow cover products.The spatio-temporal characteristics of the observation geometry and product area coverage of MILS_BRF have significant impacts on the conclusions of this assessment.Therefore,this paper analyzes the spatial and temporal distribution characteristics of MILS_BRF product coverage and observation geometry from 2011 to 2015.The results of the analysis give explicit and comprehensive information on the temporal and spatial characteristics of the MILS_BRF and the geometric limitations of the observations.The results and analysis show that when using MILLS_BRP to evaluate the accuracy of the BRDF model,it is necessary to consider the spatiotemporal characteristics of its product coverage and observation geometry.First,the intra-annual variation of the MILS_BRF coverage/coverage location needs to be considered to avoid misleading interpretations due to differences in data availability for different months or surface types.In addition,when analyzing the performance of the BRDF model at different azimuth or different observation zenith angles,the annual variation of the MILS BRF observation geometry must be considered to exclude the interference of the annual variation pattern of the MILS_BRF observation geometry,and the performance characteristics of the BRDF model itself are identified.In addition,the discovery of large-area product vacancies consisting mainly of crop and crop/natural vegetation mixed grounds for four consecutive months in the study area provides a potential direction for improving the MILS_BRF inversion algorithm.Based on the reference data of MILS_BRF,the results of accuracy analysis of MODIS BRDF product show the overall excellent performance of MCD43A1 C6 product to represent the anisotropic reflectance of land surface with root mean square error(RMSE)of 0.0262 and correlation coefficient(R)of 0.9537 for all available comparable samples of MILS_BRF and BRF predicted by MCD43A1 model.The model accuracy varies in different months,which is related to phenological phases of the study area.The accuracy of MCD43A1 model of pixels labelled as 'snow' by MCD43 is obviously low with RMSE/R of 0.0903/0.8401.This paper defines the ephemeral snow event as the date of the nominal comparison,MODIS snow coverage(MOD10A1 and MYD10A1)synthetic product snow coverage ratio is not less than 30%,but the MCD43A1 model parameters are not inferred as snow pixels.Ephemeral snow fall events further decrease the accuracy of MCD43A1 model with RMSE/R of 0.1001/0.7715,though the MCD43A1 model parameters are labelled as 'best quality,full inversion' and 'snow free'.This is inconsistent with MODIS snow cover product.These results provide meaningful information to MCD43 users,especially those,whose study regions are subject to phenological cycles as well as snow cover and change.The RossThick-LiSparseR(RTLSR)model of the current operational algorithm for the production of MODIS BRDF products does not contain the high reflectivity of snow cover and the forest ground with obvious forward scattering characteristics.The accuracy analysis of MODIS BRDF products which implemented the RTLSR model shows that the precision of the snow forest region is significantly lower than that of the forest growth season.At the same time,the full inversion data of winter in this region is significantly less than the growth season.Therefore,this paper studies the method of improving the BRDF model and its inversion strategy in snow forest area considering snow reflection characteristics.The realization form of snow reflectance kernel and the multi-kernel model formula of snow-bearing kernel are given.In this paper,the strategy of combined inversion of nuclear combination and historical data is analyzed in order to solve the problem of pixel reduction after the increase of the kernel.The accuracy of the combined inversion results of 9 models and data is evaluated.The difference between the model inversion accuracy and the number of full inversion pixels under different kernel combination modes and different historical years of input data is studied after adding the snow reflectance model.The 5-year data for the 2011-2015 of the study area shows that the 4-kernel 2-year(4K2Y)model and the inversion strategy combination performed best for the deep winter season(December and January).In other words,in this combination,the inversion accuracy of the modified model can be higher than that of MODIS operational model with RMSE 0.0279<0.0319 and R 0.9403>0.9142,and the number of full inversion pixels is increased significantly(more than 2 times of current operational algorithm).