Evaluación de la evapotranspiración real en ambientes semiáridos mediante teledetecciónevolución previsible de la evapotranspiración de referencia a lo largo del siglo XXI en la Demarcación Hidrográfica del Segura

Abstract

This thesis is framed within the Segura Hydrographic Demarcation, a typical semi-arid Mediterranean region located in the southeast of the Iberian Peninsula, in which available water resources are scarce. This scarcity makes DHS especially vulnerable to climate change, which consequences are already a reality at a global level, and which could lead to a decrease in the contributions in natural regime and of the recharge of aquifers in the coming decades. This context makes it essential to carry out studies aimed at obtaining the best possible information that allows to continue improving the planning, sustainable management and modeling of water resources. In this sense, the main objective of this thesis is to contribute to the improvement of the available information regarding different aspects related to the evapotranspiration process in semi-arid environments. The development of this thesis has been structured in 3 large chapters. In the first of these 3 chapters has carried out a comparison of different methods for the estimation in the DHS and on clear days of the instantaneous air temperature at the time of the daytime passage of the TERRA and AQUA satellites (Tinst) and the maximum and minimum air temperatures (Tmax and Tmin). This chapter derives of the need of a large part of the models that are currently used for modeling ETR, of instantaneous air temperature layers at the time of passage of the TERRA and AQUA satellites. In this modeling it is centered the second of the development chapters of the thesis. Specifically, in this first chapter have been compared 9 methods for the estimation of Tinst and 7 methods for the estimation of Tmax and Tmin. Among these methods, 2 machine learning methods have been used, Random Forest (RF) and Support Vector Machine (SVM), until now little used to estimate air temperature. In addition, has been used a multiple linear regression model (MRLM), ordinary kriging (OK), the TVX model, the Parton-Logan model and 3 regression models from MRLM, RF and SVM. As predictor variables of the different models of regression, have been used a combination of geographic and environmental variables, which can be easily obtained from digital elevation models or satellite images. These variables have been used in other studies to estimate air temperature. The best results for the estimation of Tinst have been obtained for RF, the 3 regression-kriging models and for Parton-Logan, which, unlike the rest of the models, does not require observed data of Tinst. For its part, for the maximum temperature, the best results have been obtained for RF and the regression-kriging model based on RF, while for the minimum temperature have not been obtained significant differences between the different methods. In the second chapter different variations have been proposed to the 3T and TDTM models, recently developed to obtain ETR values in semi-arid environments from of data obtained by remote sensing. The proposed variations have been validated with the data obtained for 2 flow towers (ES-LJU and ES-AMO) located in the province of Almería in scrub areas and in a semi-arid environment. In addition, in this chapter has been analyzed the spatial distribution of the ETR values obtained for said variations in the DHS. For the different variations proposed for the 3T model, the results obtained have been considerably worse than in other studies, with a RMSE ranging between 0,97 and 20 mm for ES-LJU and between 0,83 and 2,38 mm for ES-AMO. These high values are motivated mainly due to the high overestimation of transpiration obtained in this chapter. Howewer, for the TDTM-NDVI-corr model, one of the variations proposed in this chapter of the TDTM model, very promising have been obtained (RMSE of 0,32 mm for ES-AMO and of 0,33 mm for ES-LJU). Furthermore, the spatial distribution obtained for TDTM-NDVI-corr model is the most coherent with the characteristics of the study area of all the proposed variations. In the third chapter, a methodology has been proposed to obtain, evaluate and compare multimodel ensembles (MMEs) of ET0 based on some of the methods that traditionally have been used for this and in two machine learning methods (Random Forest and Support Vector Machine), in order to obtain daily series of ET0 for the 21st century and for the climate change scenarios RCP4.5 and RCP8.5, which have the least possible bias, within of the uncertainty associated with the entire process of obtaining regionalized projections temperature. ET0 values have been obtained from the method of Hargreaves calibrated. Among the 11 MMEs analyzed, the best performance has been obtained for the MME based on Random Forest, which is significantly better than for the rest of MMEs and individual ET0 series, except for the MME based on the LAD method. Also, in this chapter from the MME for which the best results have been obtained (MME based on Random Forest) has been analyzed the impact of climate change on ET0 in the DHS. In relation to this, on an annual scale a significant increase of the ET0 has been obtained for the two scenarios emissions contemplated and that the greatest increase would be recorded in the headwater areas of the Segura and Mundo rivers, the area where most of the DHS water resources are produced. In this thesis, the information acquisition and purification process, as well as the automation of the work process and the analysis and obtaining of results, will be based on the use of open Software, specifically in the use of GRASS as a Geographic Information System and R-CRAN for the analysis and download of the data. This will facilitate the reproducibility of the different chapters of this thesis in other studies.