Retrieval And Validation Of Grassland Key Photosynthetic Parameters Using Remote Sensing Data In Northern China

Grassland productivity measures production capacity of the grassland in its natural environment,and it is related to food,fiber,and fuel for both human and livestock.Therefore,the study of grassland productivity is critical to the management and planning of the grassland resources.Large-scale grassland productivity monitoring and a variety of productivity models have been made possible using a variety of remote sensing methods.A major concept for modeling vegetation productivity is the light use efficiency(LUE)approach,where leaf area index(LAI),fraction of absorbed photosynthetically active radiation(FPAR)and LUE are three important parameters of the LUE models and their accuracy is crucial for the application of the models for grassland productivity monitoring.In this study,to better monitor grassland productivity,the retrieving and validating methods of key grassland photosynthetic parameters were studied using flux tower data,ground experimental data and satellite remote sensing data.First,relevant remote sensing products were validated to assess their accuracy in grassland;then the algorithms to retrieve photosynthetic parameters(LAI,FPAR and LUE)were developed;and finally grassland LUE and gross primary productivity(GPP)were retrieved using photochemical reflectance index(PRI)and solar-induced chlorophyll fluorescence(SIF)data.The main research contents and conclusions of the paper are as follows:1.Three GPP products were validated using flux observations from Chinaflux and Fluxnet network.The results showed that the three GPP products had different performance in different grassland types.In general,the breathing earth system simulator(BESS)GPP product is more closer to the ground measured GPP in the grassland area than the other two LUE model-based GPP products(the vegetation photosynthesis model(VPM)GPP product and the moderate resolution imaging spectroradiometer(MODIS)GPP product).In the alpine meadow,VPM GPP and MODIS GPP products all slightly underestimated ground GPP,but the validated three GPP products all overestimated the ground GPP in the typical steppe.The MODIS GPP product also showed higher seasonal variability than the other two products,which was resulted from the obvious seasonal variability of the two product algorithm inputs:FPAR and VPD.The results also found an obvious bias of VPM GPP product with the ground measured GPP during the growing season,the MODIS EVI product,which is an important product algorithm input of the VPM GPP product,also showed a similar bias with the ground measured GPP,which is considered as the error source.Moreover,the research also studied the performance of GPP products and their input parameters under drought conditions,results showed overestimations of the GPP products as compared to flux observations,where the overestimation was caused by a lack of sensitivity of GPP model parameters.2.The MODIS and GEOLAND2 Version 1(GEOV1)LAI and FPAR products were validated using ground observations collected from validation network of remote sensing products in China(VRPC).For the LAI products,results showed that GEOV1 LAI is best correlated with the reference maps.The MODIS products performed well for biomes with low LAI values,but considerable uncertainty existed when the LAI was larger than 3.0 m~2/m~2.The MODIS anomalies were mostly caused by the surface reflectance uncertainty,shorter temporal resolutions and inconsistency between simulated and MODIS surface reflectances.For the FPAR products,MODIS products were more accurate as compared to ground measurement than GEOV1 FPAR product,and the three products all had certain overestimation when FPAR value is small.MODIS FPAR products show a large fluctuation in the annual change curve.3.Compared the predictive power of regression approaches and hybrid geostatistical methods to retrieve grassland LAI in the meadow steppe of Hulunber,China.The regression methods evaluated include partial least squares regression(PLSR),artificial neural networks(ANNs)and random forests(RFs).The two hybrid geostatistical methods were regression kriging(RK)and random forests residuals kriging(RFRK).Based on comparison with ground measurements,the accuracy and stability of the methods were ranked with the order:RK>RFRK>RF>ANN>PLSR.Then,a daily 30 m LAI time series was generated by using a spatial and temporal adaptive reflectance fusion model(STARFM)combined with an LAI retrieval radiative transfer model(PROSAIL),the fused data included Landsat 7&8、Sentinel-2 and MODIS.Results showed that the PROSAIL-generated LAI maps from the four data source with different spatial resolutions all exhibited a high accuracy,the STARFM model also demonstrated a very good performance in generating LAI dataset under homogeneous conditions,but for heterogeneous land surfaces with small patches,the fuse performance was not as satisfactory.4.Retrieved vegetation LUE and GPP in north China using MODIS-PRI and SIF data.Results showed that the PRI-LUE relationship can be affected by many factors,such as the satellite viewing angle and the canopy structure.PRI and LUE exhibited some correlations,and the relationship can be improved by adding meteorological factors,but LUE retrieving using PRI method in a large area is still difficult.Good relationships were observed between SIF and LUE and between SIF and GPP both on the 8-day and monthly scale.However,the relationship would not hold for sparsely vegetated areas as the result of weak SIF signal from vegetation canopy leading to the SIF-LUE and SIF-GPP relationships less satisfactory than that of EVI and NDVI.The study provides evidence for improving remote sensing simulation accuracy of grassland productivity and application of remote sensing products in grassland areas in China.It also provides a new means for grassland productivity and carbon cycle simulation monitoring in northern China,and could be of great significance for global carbon cycle research.And the PRI-and SIF-based LUE and GPP modeling add new information to the study area and can be further used in a wide variety of applications,especially in evaluating the ecological service and carbon cycle under global climate change.