Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Chili → Actual Evapotranspiration Estimates in Arid Regions Using Machine Learning Algorithms with In-Situ and Remote Sensing Data

Pontificia Universidad Catolica de Chile (2020)

Actual Evapotranspiration Estimates in Arid Regions Using Machine Learning Algorithms with In-Situ and Remote Sensing Data

Mosre Poller, Josefina Amanda

Titre : Actual Evapotranspiration Estimates in Arid Regions Using Machine Learning Algorithms with In-Situ and Remote Sensing Data

Estimaciones reales de la evapotranspiración en regiones áridas mediante algoritmos de aprendizaje automático con datos de detección remota e in situ

Auteur : Mosre Poller, Josefina Amanda.

Université de soutenance  : Pontificia Universidad Catolica de Chile

Grade : Master of Science in Engineering 2020

Résumé
Evapotranspiration (ET) is a relevant hydrological process in arid regions where water is vital for the development of local communities and ecosystems. ET estimations in arid regions have been historically a great challenge, because these landscapes mainly consist of sparse vegetation adapted to drought conditions, which do not comply with many of the assumptions used in traditional ET estimation methods. Nevertheless, in arid areas several studies have shown good results when implementing empirical regression formulas that, despite their simplicity, are comparable in accuracy to more complex models. Although many types of regression formulas to estimate ET exist, there is no consensus on what variables must be included in the analysis. In this research, I used machine learning algorithms to find the main variables that predicts daily and monthly ET in arid regions using linear regression equations. Meteorological data alone and then combined ZiWh UemoWe VenVing YegeWaWion indiceV (VI¶V) ZeUe XVed aV inSXW in monWhl\ estimations. In-situ ET fluxes and meteorological data were obtained from ten sites in Chile, Australia and United States. Daily and monthly ET estimations were evaluated in three validation sites, one from each country, obtaining different performance. My results indicate that the available energy is the main meteorological variable that predicts ET fluxes in the assessed sites, even when arid regions are typically described as water-limited environments. The VI that represents better the in-situ ET fluxes is the Normalized Difference Water Index (NDWI), which unlike otheU VI¶V, UeSUeVenWV ZaWeU aYailabiliW\ in plants and soil instead of vegetation activity. The best performance of the linear regression equations was obtained for monthly estimates ZiWh Whe incoUSoUaWion of VI¶V aW Whe U.S. validation site (R2 = 0.82), whereas the worst performance of these equations was obtained for monthly ET estimates at the Australia validation site when only meteorological data are considered. The incorporation of remote-sensing information results in better ET estimations in contrast with estimations obtained when only meteorological data are included in the analysis.

Version intégrale)

Page publiée le 12 juin 2021