Global Retrieval Of Leaf Chlorophyll Content From Optical Remote Sensing Images

Photosynthesis provides the basic material and energy source of all life on earth,and is of great significance for achieving energy conversion in nature and maintaining the carbon-oxygen balance in the atmosphere.As the main pigment in plant photosynthesis,chlorophyll plays a pivotal role in the light-harvesting reactions,and its content(leaf chlorophyll content,Chl _Leaf)is thus directly related to the absorbed photosynthetically active radiation and is a key parameter controlling vegetation productivity.The spatial and temporal dynamic information of Chl _Leaf is the basis of accurate simulation of the carbon flux in terrestrial ecosystems.The accumulation of a large amount of satellite remote sensing observation data in recent decades has provided the possibility of generating long-term Chl _Leaf products.However,most existing algorithms for retrieving Chl _Leaf from remote sensing data are designed for small regions,and are difficult to apply to different biomes and different stages in the vegetation growing season.The only released Chl _Leaf product（Croft et al.,2020）has problems in algorithm implementation and a limited time span（2002–2011）,thus does not satisfy the needs of terrestrial ecosystem modeling and dynamic analysis.The satellite-derived canopy reflectance is controlled by leaf biochemical constituents as well as canopy structure.Large variations have been observed in canopy structure among different biomes and throughout vegetation growing seasons.The canopy structure-related factors confound the changes in canopy reflectance caused by Chl _Leafand make the retrieval of global long-term Chl _Leaf complicated and challenging.This study aimed to generate global products of Chl _Leaf using optical remote sensing images.Because of the unique characteristics of Sentinel-2,MODIS and MERIS,the algorithms for retrieving Chl _Leaf from these three sensors and the generation of global Chl _Leaf products using MODIS and MERIS were studied.The key problem of generating a Chl _Leaf product from satellite images is to eliminate the influences of leaf area index（LAI）and canopy non-photosynthetic materials,so as to achieve the seasonal dynamics of Chl _Leaf.Three specific research contents and corresponding conclusions are as follows:（1）Retrieving Chl _Leaf from Sentinel-2 data using a matrix-based vegetation index（VI）combination approachAn improved look-up table method—matrix-based VI combination method was proposed to remove the influence of LAI on Chl _Leaf retrieval.Three 2-dimensional matrix-based relationships between Chl _Leaf and three VI pairs(TCARI&OSAVI,R₇₄₀/R₇₀₅&R₈₆₅/R₆₆₅ and RERI^[705]&RERI^[783])were established using the PROSAIL simulated dataset.Sensitivity studies indicated the three matrices were effective in separating Chl _Leafand LAI spectral signals,as well as in decreasing the influences of leaf angle distribution,soil background,and sun-view geometry.The matrix of RERI_[705]-RERI^[783],where RERI^[705]=(R₇₀₅-R₆₆₅)/R₈₆₅ and RERI^[783]=(R₇₈₃-R₇₀₅)/R₈₆₅,was most sensitive to Chl _Leaf compared with the matrices of TCARI-OSAVI and R₇₄₀/R₇₀₅-R₈₆₅/R₆₆₅.The Chl _Leaf retrievals from both PROSAIL simulated reflectance and winter wheat observed reflectance showed that the matrix of RERI_[705]-RERI_[783]had the highest accuracy in Chl _Leaf estimation.The Chl _Leaf retrieval accuracies using the three matrices of VI pairs were all higher than those using the individual VIs that make up them,as well as the corresponding VI ratios.The standard deviation of matrix cells can be used to represent the retrieval uncertainty of the Chl _Leaf when using the matrix-based approach.The matrix-based VI combination approach allows a more accurate Chl _Leafestimation and has the potential for operational retrieval of Chl _Leaf from multispectral satellite data.（2）Retrieving global Chl _Leaf from MODIS using a multi-level matrix systemTo further eliminate the influence of canopy non-photosynthetic materials on Chl _Leaf retrieval,an additional parameter of canopy fraction of brown leaves（f _B）was introduced into the PROSAIL model.A multi-level matrix system,which consisted of the f _B matrices of RGVI-NBR2 and the Chl _Leaf matrices of CI_green-SR,was built to estimate f _B and Chl _Leaf iteratively from MODIS data.Based on the algorithm,the MODIS global long-term Chl _Leaf product was generated.Sensitivity studies indicated the Chl _Leaf matrix of CI_green-SR was mostly affected by f _B and the f _B matrix of RGVI-NBR2 was mostly affected by Chl _Leaf.By using the proposed matrix system,the influence of canopy non-photosynthetic materials on MODIS Chl _Leaf estimation was reduced,and the seasonal variations of Chl _Leaf at sites of different biomes were achieved.The generated MODIS Chl _Leaf global maps showed distinct temporal and spatial variations,with high Chl _Leaf values throughout the year in evergreen broadleaf forests（>30μg/cm²）,and significant seasonal dynamics in deciduous broadleaf forests,mixed forests,shrublands,croplands and grasslands.The divergence in the temporal trends of Chl _Leaf and LAI in deciduous broadleaf forests provided evidence of the algorithm’s effectiveness in separating the spectral signals of Chl _Leaf and LAI.Despite the underestimation of high Chl _Leaf values in croplands,preliminary validation of the MODIS Chl _Leaf product against the ground measurements at evergreen needleleaf forest,deciduous broadleaf forest,cropland and grassland sites achieved a quite high accuracy of R²=0.56,RMSE=5.73μg/cm² and NRMSE=15.3%.The generated long-term global Chl _Leaf product provides new opportunities for analyzing vegetation physiological dynamics and improving the carbon cycle modeling in climate change studies.（3）Retrieving global Chl _Leaf from MERIS using a neural network methodThe spectral signal of canopy non-photosynthetic materials was simulated using four different ways in PROSAIL model.Four neural networks were trained using these different simulations.The seasonal variations of Chl _Leaf retrieved from MERIS using the four neural networks were investigated.The neural network,which was trained using the simulations of the PROSAIL model with f _B introduced as well as the Chl _Leafand LAI distributions at high f _B values constrained was used to generate a high-precision Chl _Leaf product from MERIS data.The estimated time series of MERIS Chl _Leaf using the previous neural network was most consistent with those of MTCI and NDVI,compared with three other neural networks,which were generated based on PROSAIL-simulated datasets without considering canopy non-photosynthetic materials,and considering canopy non-photosynthetic materials by using brown pigment content(C_bp)and by using f _B alone.It also corresponded very well with the in-situ measurements at a sugar maple site.The new MERIS Chl _Leaf product derived in this thesis had seasonal trends consistent with those of LAI,and avoided the problem in the MERIS Chl _Leaf product from Croft et al.（2020）that often show higher values at the beginning and end of vegetation growing seasons.The new product was in similar Chl _Leaf ranges as the time series of MODIS Chl _Leaf in deciduous broadleaf forests,evergreen broadleaf forests and evergreen needleleaf forests,but had obviously higher values at sites of shrubland,grassland and cropland.The new product successfully obtained the seasonal patterns of Chl _Leaf for different biomes and is useful in improving the carbon cycle modeling of terrestrial ecosystems.In this study,different algorithms were developed for retrieving Chl _Leaf for different satellite sensors at global scale and global Chl _Leaf products are generated using these algorithms.By using a matrix-based method or a neural network method,the seasonal variations of Chl _Leaf in different biomes were retrieved.The generated global Chl _Leafproducts from MODIS and MERIS have a wide range of applications such as estimating the carbon flux of terrestrial ecosystems,monitoring ecological environment disturbances,guiding agricultural and forestry production.