Remote Sensing Estimation And Authenticity Test Of Forest LAI At Different Spatial Scales

The "scale effect" of remote sensing data is objective.Remote sensing data at different spatial scales,models for remote sensing inversion and remote sensing products have such scale effects.Whether the spatial information of data points can match the spatial information of pixels is still a problem to be solved.How well do the ground measurement points represent the remote sensing pixels,how to obtain the relative true value of the pixels and how large the spatial resolution can truly reflect the leaf area index(LAI)of forest areas are all important issues in quantitative remote sensing.Based on the concepts of feature accuracy and relative uniformity,the measured space scope of two plant analyzers[LAI-2200 and tracing radiation and architecture of canopies(TRAC)]were calculated under the condition of keeping the real observed area of ground measured data consistent with that obtained by remote sensing pixels.The relative pixels at different scales were found by combining the three different spatial resolution remote sensing images of GF-2 with 4.1 m spatial resolution,Sentinel-2 with 10 m spatial resolution and Landsat-8 with 30 m spatial resolution.True value and vegetation index were used to construct the estimation model,and the optimal estimation model was obtained by comparing them.The differences of estimation of forest leaf area index at different spatial scales were analyzed.Meanwhile,the spatial representativeness of data sets at three scales was evaluated by introducing indicators,and the optimal scale for estimating leaf area index was obtained.According to the law of conservation of matter,the relative truth values of large scale were obtained,the accuracy of different scale conversion methods were evaluated and the authenticity of MODIS LAI products were tested.The results showed that the point of LAI-2200 data can be represented by the maximum spatial range of 12m*12 m pixel scale,so GF-2 and Sentinel-2 use single-point model,and Landsat-8 use traditional simple average multi-point model.The unary exponential model was constructed using SR,NDVI,SAVI and ARVI vegetation indices.The results showed that the correlation between SR and LAI is higher than that of NDVI,RVI and SAVI,and the estimation of the research area LAI can be accurately achieved by the unary exponential model using SR and LAI.Remote sensing images of with different spatial scales have some certain influence on LAI modeling.The estimation results were validated on the scales of 30 m and 100 m,indicating that Sentinel-2 based on the scale of 10 m is more suitable for estimating forest area LAI.The spatial representativeness of the three scales obtained by the optimal model was evaluated,and the results were consistent with the space scope measured by the instrument,the single-point datasets of GF-2 and Sentinel-2 for the two-scale pixels have better spatial representativeness.The spatial resolution of GF-2 is the highest,the coverage type of the pixels is more single,the pixels are more uniform,and the coefficient of sill(CS)is the smallest,but the modeling precision and fitting are not the best;Landsat-8 has a spatial resolution of 30 m,and the vegetation cover type in the pixel is slightly more complex than the other two scales,so the uniformity of the pixel is relatively low and the spatial representation is relatively poor.For forests,the higher the resolution does not necessarily fully estimate the forest LAI distribution.The Sentinel-2 remote sensing image with 10 m spatial resolution can better predict forest LAI and is the optimal scale for estimating forest LAI.According to the law of conservation of matter,the accuracy of five scale conversion methods is verified based on the relative truth value of the optimal scale.It is concluded that the frequency weight method has the highest accuracy when the relative truth value is obtained.Finally,the authenticity of MODIS LAI products is was tested.The results of authenticity test showed that the estimation results of MODIS LAI products are were quite different.Therefore,the estimation accuracy of MODIS LAI remote sensing products needs to be improved.It can not be directly used as input data of mechanism model,so it needs to be corrected before it can to be used as initial input data for model simulation.