Development of an earth observation processing chain for crop biophysical parameters at local and global scale

Resumen

This thesis’ topics embrace remote sensing for Earth observation, specifically in Earth vegetation monitoring. The Thesis’ main objective is to develop and implement an operational processing chain for crop biophysical parameters estimation at both local and global scales from remote sensing data. Conceptually, the components of the chain are the same at both scales: First, a radiative transfer model is run in forward mode to build a database composed by simulations of vegetation surface reflectance and concomitant biophysical parameters associated to those spectrum. Secondly, the simulated database is used for training and testing non-linear and non-parametric machine learning regression algorithms. The best model in terms of accuracy, bias and goodness-of-fit is then selected to be used in the operational retrieval chain. Once the model is trained, remote sensing surface reflectance data is fed into the trained model as input in the inversion process to retrieve the biophysical parameters of interest at both local and global scales depending on the inputs spatial resolution and coverage. Eventually, the validation of the leaf area index estimates is performed at local scale by a set of ground measurements conducted during coordinated field campaigns in three countries during 2015 and 2016 European rice seasons. At global scale, the validation is performed through intercomparison with the most relevant and widely validated reference biophysical products. The work elaborated in this Thesis is structured in six chapters including an introduction of remote sensing for Earth observation, the developed processing chain at local scale, the ground LAI measurements acquired with smartphones, the developed chain at global scale, a chapter discussing the conclusions of the work, and a chapter which includes an extended abstract in Valencian. The Thesis is completed by an annex which include a compendium of peer-reviewed publications in remote sensing international journals. The outline of each chapter is summarized as follows: - Chapter 1 introduces the reader into the framework of remote sensing for Earth observation and reviews the main definitions, used methodologies and approaches for biophysical parameter estimation from remote sensing data. - Chapter 2 reviews the fundamentals of radiative transfer model inversion detailing the PROSAIL formulation basis and the main features of the machine learning regression techniques used for the inversion. The last part of the chapter describes the developed processing chain at local scale in the framework of ERMES project. - Chapter 3 reviews the in situ measurements acquisition procedure using classical instrumentation such as LAI-2000 and DHP techniques. This chapter introduces and describes the in situ LAI measurements acquired with new technologies such the use of smartphones through a dedicated application called PocketLAI. The chapter addresses the validation of the LAI retrieval chain at local scale with ground data. - Chapter 4 describes the developed processing chain at global scale in the framework of Land-SAF project (exploitation of EUMETSAT satellites) and the indirect validation of the estimates by intercomparison with reference biophysical products such as MOD15A2, GEOV1 and VEGA products. - Chapter 5 concludes up the Thesis’ achievements and discusses the main conclusions. - Chapter 6 provides a complete overview of the Thesis in Valencian. - Annex attaches the peer-reviewed scientific publications directly related with the work conducted in this Dissertation.