Técnicas multiescala en el seguimiento de la vegetación con imágenes de satélite

Abstract

This Thesis focuses on the study of vegetation as one of the elements that structure the complexity of terrestrial agro-ecosystems. It is a study that has been developed on two fundamental pillars. On the one hand, the data on the vegetation is obtained from the images acquired by the Earth observation satellites, thanks to the vegetation indices. On the other hand, the tools used to study these data are those provided by fractal and multifractal analysis. Satellite images are today one of the most important sources of information on the Earth's surface. Thanks to them, continuous monitoring of the state and health of the environment is possible. Its importance has grown throughout its half century of life, mainly due to the impact of global climate change on ecosystems and people's lives. The ability of imagery to immediately have historical data on vegetal cover is one of the cornerstones of understanding climate hazards and mitigating related effects. In these fifty years, the images of the earth's surface obtained by satellites have been improving in detail (spatial resolution), in acquisition frequency (temporal resolution), in regions of the electromagnetic spectrum (spectral resolution) and in the dynamic range of digital values (radiometric resolution). Thanks to this, today a large volume of information is available that makes it necessary to have new techniques for its characterization and analysis. Because one of the inherent characteristics of this information is the concept of scale, understood as the different resolutions listed above, we find the second pillar that supports this Thesis: the multiscale characterization of the information. Thus, for the multiscale characterization of the information provided by satellite images, the powerful tool provided by fractal and multifractal analysis is used. With these tools, an attempt is made to find answers to questions such as: How comparable is the information on vegetation provided by images of different pixel size, taken by different satellites? Can this information be characterized by a single parameter, such as the fractal dimension? How similar are the reflectivity patterns in the spectral bands of different satellites, particularly those used to calculate the vegetation indices? Or, can the correlation between the vegetation index and the soil moisture index, both obtained with satellite images, be characterized by the multifractal spectrum? The fractal theory framework provides a multiscale analysis algorithm that is easily implemented to satellite images, the box-counting algorithm, which allows calculating the fractal dimension of an object, in our case of an image. We have used this algorithm to characterize the vegetation index obtained with images of 250 m and 30 m pixel size. Thus, it has been possible to study the behaviour of the vegetation index between scales ranging from 960 m to 16 km, calculating the fractal dimension of the segmented NDVI in different ranges of values. Large differences are observed in the fractal dimensions of the indices at 250 m and at 30 m when the vegetation is scarce or absent, the differences being minimal when the vegetation covers more than 40% of the pixel. The multifractal analysis framework provides another multiscale algorithm, equivalent to box-counting, which is also easily implemented to images, the Chhabra-Jensen algorithm. With this algorithm, the two parameters that determine the so-called "singularity spectrum" or "multifractal spectrum" can be directly calculated. We have used this algorithm to study the reflectivity patterns in the common spectral bands of two satellites, with images of 30 and 4 m pixel size. It has also been used to study two vegetation indices obtained with these bands. From the geometric analysis of the multifractal spectrum, the reflectivity patterns of each of the visible and near-infrared bands have been characterized, as well as two vegetation indices obtained with them (NDVI and EVI), depending on the spatial resolution and radiometric of the images. It is the first time that the behavior of these patterns has been reported. It has also been applied in the study of the correlation between the vegetation index and the soil moisture index obtained with images of 500 m pixel size. This study also used a variant of multifractal analysis, very new in the field of remote sensing, called “joint multifractal analysis”. This analysis is used for the simultaneous study of several correlated multifractal measurements, such as vegetation and soil moisture indices. Finally, we have approached the study of the dynamics between the vegetation and soil moisture indices throughout an annual cycle and their correlation, focusing on a Mediterranean pasture landscape.