2020/10/13 by J. Emmanuel Johnson, Valero Laparra, Johnson, J. Emmanuel +5
Environmental Science · #Remote Sensing in Agriculture #Soil Geostatistics and Mapping #Remote Sensing and LiDAR Applications
paper · pdf · doi:10.48550/arxiv.2010.06476
Information theory is an excellent framework for analyzing Earth system data\nbecause it allows us to characterize uncertainty and redundancy, and is\nuniversally interpretable. However, accurately estimating information content\nis challenging because spatio-temporal data is high-dimensional, heterogeneous\nand has non-linear characteristics. In this paper, we apply multivariate\nGaussianization for probability density estimation which is robust to\ndimensionality, comes with statistical guarantees, and is easy to apply. In\naddition, this methodology allows us to estimate information-theoretic measures\nto characterize multivariate densities: information, entropy, total\ncorrelation, and mutual information. We demonstrate how information theory\nmeasures can be applied in various Earth system data analysis problems. First\nwe show how the method can be used to jointly Gaussianize radar backscattering\nintensities, synthesize hyperspectral data, and quantify of information content\nin aerial optical images. We also quantify the information content of several\nvariables describing the soil-vegetation status in agro-ecosystems, and\ninvestigate the temporal scales that maximize their shared information under\nextreme events such as droughts. Finally, we measure the relative information\ncontent of space and time dimensions in remote sensing products and model\nsimulations involving long records of key variables such as precipitation,\nsensible heat and evaporation. Results confirm the validity of the method, for\nwhich we anticipate a wide use and adoption. Code and demos of the implemented\nalgorithms and information-theory measures are provided.\n