Abstract:
Abstract: Leaf area index (LAI) is one of the key parameters to show canopy structure of crops. It is of great importance to obtain accurate LAI for monitoring and estimating the yield. Unfortunately, the LAI estimation results have scale effect resulting from feature space complexity, the difference of different remote sensing data source, and the nonlinear of remote sensing inversion model. So the scaling transformation is necessary when the multi-source remote sensing images are used. Scaling transformation is a process that extending the information and knowledge gained from a scale to other scales, including upscaling and downscaling. For a good upscaling method, the inherent information of high resolution image should be kept in low resolution image although its spatial resolution has been reduced. Statistical relationships between data for upscaling based on remote sensing products, and pixel decomposition method for upscaling the existed spatial continuously data from high-resolution to low resolution are commonly used. But most of agronomy parameters data are from site observation and sampling, these point data are informative and accurate. But the point locations are separated and dispersed, and space representatives are limited. So it is practical to up-scale these point data to spatial continuously data. There are regression analysis method, geo-statistical method, fractal method and others for point data's upscaling, in which counter point sample size requires a high quality, and short of sufficient consideration on effect from various supplementary information related to the object to the sample point. MaxEnt is a general-purpose machine learning method with a simple and precise mathematical formulation. So the MaxEnt model was used in this study for upscaling maize canopy LAI from point data to spatial continuously data. The LAI point data measured in field work, Landsat8 OLI remote sensing image and the meteorological data were the data sources in this study. Firstly, classified the measured LAI data, getting meteorology temperature, surface temperature and relative humidity through spatial interpolation, spectral reflectance and spatial continuously data of vegetation index from remote sensing inversion. The data above could be used as environment variables, and distribution probabilities in different kinds of LAI data could be obtained through the MaxEnt Model. Secondly, we made a superposition for LAI probabilistic forecasting pictures, which converted LAI probability value to quantitative value. Lastly, we classified the research areas, and provided crops planting region, and then made a mask processing for the conversion results using the resulted pictures of classified fields so that we could get the results of LAI upscaling for maize canopy. The result showed that, comparing the upscaling conversion result with measured LAI data through the MaxEnt Model, we could found that R2 equaled 0.601 and RMSE was 0.898, indicating a high correlation between the two. The mutual occlusion between maize canopy leaves, resulting in an overall low result within an acceptable range. Therefore, the MaxEnt model can be used to upscaling crop canopy LAI from point data to spatial continuously data, and this method can be used to other crops as well.