Validation Of Coarse-resolution LAI Products Over Chinese Croplands Using Field Measurements

Leaf Area Index(LAI)is a key parameter to characterize the geometric structure and function of vegetation canopy,simulate ecosystem productivity and energy exchange between the surface and the atmosphere.LAI products are also widely used for surface process simulation,crop growth monitoring and yield estimation and global change research.Therefore,quantitatively evaluating the uncertainty of LAI remote sensing products,that is,evaluating LAI products through ground measurement data,is crucial for the correct use of LAI products.Ground-measured LAI,serving as an important reference for the validation of remote sensing products,are mainly obtained by direct or indirect measurement methods.However,almost all of the current validation data are obtained by indirect optical methods,which is challenging to acquire field LAI of low vegetation,such as wheat and paddy rice in the early growth stages.Existing LAI products are coarse resolution and inversion algorithms were developed on a global scale.Only when the field measurements can represent well the situation within the pixel scale,can the field data be directly used for LAI product validation.Otherwise,it will bring non-negligible errors to the product validation.However,due to the coarse spatial resolution,and the labor-cost of ground observation,validation campaign of LAI products has become the focus and difficulty of LAI products and their application research.This study focuses on the validation of LAI remote sensing products.The field LAI measurements were obtained by the destructive sampling method,combined with the fine-resolution satellite images,to establish semi-empirical statistical models based on the Beer-Lambert law.Then the fine-resolution LAI reference maps are retrieved using these semi-empirical models.The validation of LAI products is carried out based on the fine-resolution LAI reference maps over Chinese croplands,which quantitatively evaluated the contribution of the uncertainty of inversion algorithm and the degree of surface heterogeneity to scale effect.This study provides new ideas and references for improving the applicability of LAI inversion algorithms at the regional scale.The main research contents and conclusions of the paper are as follows:(1)Semi-empirical statistical model of NDVI and LAI were established based on the Beer-Lambert law,combining with the fine-resolution Landsat data to upscale the field measurements.A fine-resolution LAI reference dataset for validation of LAI products over Chinese croplands was developed.This study collected 43 growth stages and 1010 samples based on the destructive sampling method in the four study regions(Beijing,Henan,Heilongjiang and Anhui),which were used to calibrate the parameters of semi-empirical models.The fine-resolution LAI are obtained using Landsat satellite data based on semi-empirial models.Based on GLOBLAND30-2010 land cover product and ground sampling points with GPS corrdinates,eighty reference LAI maps,each with an area of 3 km×3 km and a percentage of cropland larger than 75%,were finally selected as the fine-resolution validation dataset(Val LAI＿Crop).(2)Fine-resolution LAI reference maps serving as a bridge of scale,can compensate for the spatial mismatch between field measurements and coarse-resolution pixels.It is an important step in the validation of coarse-resolution LAI products.Based on the Val LAI＿Crop dataset,the validation of five major global LAI products(MCD15A2H,GLASS,GEOV2,GLOBMAP and GLOCC)was carried out over Chinese croplands.The results showed that GEOV2 LAI product gives the highest accuracy over croplands in China,with R~2of 0.81,RMSE of 0.49,and RRMSE of28.6%;all the products of interests show underestimation in the study region of Beijing,Henan and Anhui,with the relative bias ranges from-82.0%to 8.8%,while they all show overestimation in the study region of Heilongjiang,with relative bias range from11.5%to 56.9%.(3)The study focuses on the physical mechanism of scaling effect.A quantitative calculation method of scaling differences was developed based on two upscaling methods,one called“invert first and then average”and the other called“average first and then invert”.Based on the Taylor series expansion method,the influence of spatial heterogeneity and nonlinearity of model on the LAI scaling differences was quantitatively analyzed,it was found that the second-order derivative terms of red and near-infrared bands could well characterize the scaling differences of LAI remote sensing products,with R~2 of 0.84.The main innovative contributions of this study include:(1)The first fine-resolution validtiaon dataset based on destructive sampling method for LAI products over Chinese croplands(Val LAI＿Crop)was developed.The existing validation datasets are almost all obtained through indirect methods.However,indirect methods using optical intruments could cause uncertainties for crops such as wheat or rice and other low vegetation.Therefore,field LAI measurements based on the destructive sampling method,combined with fine-resolution satellite data establish semi-empirical model based on the Beer-Lambert law.Based on GLOBLAND30-2010land cover product,eighty reference LAI maps,each with an area of 3 km×3 km and a percentage of cropland larger than 75%,were finally selected as the first fine-resolution validation dataset based on destructive samping method(Val LAI＿Crop).(2)Based on the Val LAI＿Crop dataset,this study systematically assessed the uncertainties of five major global LAI products(MCD15A2H,GLASS,GEOV2,GLOBMAP and GLOCC)over Chinese croplands.Currently,the inversion algorithms of LAI products are developed at global scale.However,there are great uncertainties in transferability and applicability of inversion algorithms.Therefore,the uncertainties of global LAI products are evaluated based on Val LAI＿Crop dataset over Chinese croplands,all the products of interests show underestimation in the study region of Beijing,Henan and Anhui,while they all show overestimation in the study region of Heilongjiang with low soil background albedo.The research results provide a reference for the improvement of LAI products in the future.(3)A quantitative calculation method of scaling differences was developed based on two upscaling methods,one called“invert first and then average”and the other called“average first and then invert”.It was found that the coarse-resolution LAI presented underestimation over the area with high soil background albedo and overestimation over the area with low soil background albedo.Based on the Taylor series expansion method,a quantitative correction method for scaling effect based on the second-order derivative and spatial heterogeneity of red and near-infrared bands is developed.It is found that the second-order derivative terms of red and near-infrared bands can well characterize the scaling differences of LAI products.The research is of great theoretical significance for the correction of scaling effect.In summary,this study carried out the validation of global LAI products over Chinese croplands,and provided the world’s first fine-resolution LAI validation dataset based on destructive sampling method.Based on the Val LAI＿Crop dataset and in-situ continuous observations,the uncertainties of global LAI products were investigated over Chinese croplands.The scaling effect of LAI product and its correction method were analyzed as well,which is of great significance for improving the accuracy of the global LAI product inversion algorithm at regional scale.