Show simple item record

dc.contributor.authorGiraldo Osorio, Juan Diego 
dc.date.accessioned2012-09-25T06:49:02Z
dc.date.available2012-09-25T06:49:02Z
dc.date.issued2011-12
dc.description.abstract[SPA] La región Sudano Saheliana en África Occidental, una de las más pobres del planeta, se caracteriza por una gran variabilidad de las precipitaciones y un rápido crecimiento demográfico. Dada su vulnerabilidad climática, en la región los eventos extremos (sequías e inundaciones), causan cuantiosos daños que con frecuencia involucran perjuicios a la propiedad y pérdida de vidas humanas. El conocimiento de las tendencias plausibles de los eventos extremos de precipitación es crucial desde una perspectiva política, para la administración de los recursos hídricos a escala de cuenca. Los análisis basados en datos observados y obtenidos desde Modelos Climáticos Regionales (Regional Climate Models -RCM-), han revelado un comportamiento no estacionario de las variables hidrometeorológicas. Los análisis de frecuencia tradicionales son estacionarios, pero se han detectado tendencias en las series históricas que obligan a utilizar modelos estadísticos no estacionarios para ajustar los parámetros de la Función de Densidad Probabilidad (Probability Density Function, PDF). Además, las proyecciones de cambio en los valores de los eventos extremos han mostrado gran divergencia entre los diferentes RCMs, incrementando la incertidumbre en las predicciones. En este contexto, metodologías que consideren tanto la variabilidad proporcionada por cada RCM, como la naturaleza no estacionaria de las series temporales, deberían ser aplicadas para la construcción de mapas de riesgo de eventos hidrometeorológicos extremos. En la presente Tesis, se ha modelado la naturaleza no estacionaria de las series mediante las herramientas GAMLSS (Generalized Additive Models for Location, Scale and Shape), que permiten ajustar con mucha flexibilidad los parámetros de las PDF seleccionadas. Para enfrentar el problema de la incertidumbre generada por la divergencia en las predicciones, se ha optado por construir PDF ensemble utilizando técnicas de bootstrapping. Si bien, en los primeros trabajos presentados no se construyeron ensembles puesto que se disponía de muy pocos RCMs. Cuando el Proyecto Europeo ENSEMBLES hizo disponible a la comunidad científica todos los RCMs considerados para el dominio espacial África Occidental, se justificó la construcción de los ensembles multimodelo. Desde entonces, el trabajo de Tesis se ha centrado en la identificación y desarrollo de metodologías robustas y reproducibles de construcción de PDF ensemble de aplicación regional. El aporte de cada RCM para la construcción de las PDF ensemble se ha estimado de acuerdo con el método REA (Reliability Ensemble Average). Los factores REA se calcularon de acuerdo con una medida cuantitativa simple que mide el acuerdo entre la distribución de referencia (observada desde los datos históricos, o estimada para el futuro como una combinación de los RCMs), y las distribuciones provistas desde los datos de los RCMs. Se ha trabajado en la Cuenca del Río Senegal, considerando como variables la Longitud Máxima Anual de las Rachas Secas (Annual Maximum Dry Spell Length, AMDSL), y la Precipitación Diaria Máxima Anual (Annual Maximum Daily Rainfall, AMDR), con el objetivo de generar mapas de riesgo y evaluar impactos futuros. El análisis regional de las AMDSL, considerando las ecorregiones al interior de la cuenca, ha predicho aumentos en su media y desviación estándar. Para las AMDR se construyeron mapas interpolados, desde estadísticos derivados de la PDF ensemble para cada sitio de análisis, junto con los intervalos de confianza. La diferencia calculada para mapas construidos en dos años de referencia (1990 como año de referencia del cambio, y 2050 como horizonte de predicción), reveló las tendencias espacio-temporales de las diferentes variables estudiadas. La distribución espacial de las tendencias de las AMDR demuestra que las precipitaciones extremas aumentarán significativamente en el Valle del Río Senegal debido al incremento en la media y la desviación estándar en esta región. Las conclusiones del trabajo, especialmente las distribuciones espaciales, constituyen verdaderos aportes a los procesos de toma de decisión, y al desarrollo de estrategias de mitigación y adaptación, para enfrentar el aumento de recurrencia de los eventos extremos debido al cambio climático. Concretamente en el Valle del Río Senegal se ha observado una clara tendencia a aumentar de las precipitaciones máximas diarias para el horizonte 2050, en una zona especialmente sensible a las inundaciones debido a la alta densidad demográfica y su dependencia de la agricultura de recesión. Por ello, como medida de adaptación al cambio climático, se considera el seguimiento y alerta de inundaciones basadas en teledetección. Esta alternativa es aún más atractiva cuando los medios para realizar mediciones en tierra son escasos y contienen información poco confiable, como en el caso del Valle del Río Senegal. En este sentido, se desarrolló una herramienta computacional para el seguimiento dinámico de inundaciones utilizando imágenes de satélite. Ello requiere la utilización de imágenes con alta resolución temporal, con el inconveniente de que tales imágenes usualmente presentan resoluciones espaciales bajas. Con el fin de aumentar la resolución espacial, se ha desarrollado una metodología innovadora de Análisis de Sub-píxeles (Sub-pixels Analysis, SA) que, considera la topografía digital junto con algunos de sus atributos derivados. Aplicando esta herramienta para el seguimiento de un evento de inundación en el Valle del Río Senegal, se ha obtenido una mejora sustancial en la delimitación de las áreas inundadas con el SA, comparándolo con un método de clasificación supervisado tradicional que ha utilizado las mismas imágenes de satélite. [ENG] The Sudano Sahelian region in West Africa, one of the poorest in the World, is characterized by the great variability of rainfall and the rapid population growth. Given its climate vulnerability, extreme events of drought and rainfall cause extensive damage, which often involve property damage and human lives losses. The knowledge about plausible trends of extreme events is crucial from a policy perspective, for water resources management at basin level. Analyses based on observed data and obtained from Regional Climate Model (RCM) have revealed a nonstationary behaviour of hydrometeorological variables. The traditional frequency analysis suppose stationarity. Nevertheless, non-stationarity probabilistic models should be used to adjust the Probability Density Function (PDF) parameters, because have been detected significant trends in the historical data. Also, change projections of extreme events have shown great divergence between RCMs, increasing the forecast uncertainty. In this context, methodologies should be applied to take into account the variability provided by RCMs, as well as non-stationary nature of time series in order to draw risk maps of extreme events (drought and extreme rainfall). The GAMLSS tool (Generalized Additive Models for Location, Scale and Shape) has been chosen to model non-stationary nature of time series obtained from RCMs. GAMLSS is flexible to adjust the parameters of the selected PDF. It was decided to build PDF ensemble using bootstrapping techniques, to face the uncertainty problem generated by the models forecast divergence. The early papers presented in the compendium, ensembles were not built because a small number of models were available at that time. However, releasing to the scientific community all of the RCMs which were considered for Africa domain, by the ENSEMBLE European Project, the construction of ensembles was justified. Since then, the Thesis has been focused on identifying and developing robust and reproducible methods to build PDF ensemble of regional application. The weight of each RCM in the PDF ensemble was estimated via REA method (Reliability Ensemble Average). The REA factors were calculated according to a simple quantitative metric which measures the “distributional agreement” between the reference distribution (from historical data, or estimated for the future as a combination of RCMs), and distributions from data provided by RCMs. The Senegal River Basin has been the study zone. The Annual Maximum Dry Spell Length (AMDSL) and the Annual Maximum Daily Rainfall (AMDR) have been considered as work variables, with the objective of drawing risk maps and evaluating future impacts. The AMDSL were regionally analyzed, taking into account the identified ecoregions within the basin. The analysis has shown that, in general, the models predict increases in mean and standard deviation of the AMDSL. Interpolated maps were built from statistics derived from the PDF ensemble at each grid-site, together with confidence intervals for the AMDR. The difference between the maps for two reference years (1990 as reference year of change, and 2050 as a plausible prediction horizon) was assessed, and spatiotemporal trends of the different variables studied were obtained. The spatial distribution of trends shows that extreme precipitation will increase significantly in the Senegal River Valley, due to the increase of the AMDR mean and standard deviation at the region. The outcomes of the work, specially the maps with the spatial distributions, constitute real contributions to decision making processes, and to develop mitigation and adaptation strategies to cope with extreme events associated with climate change. In the Senegal River Valley, the results have shown a clear rising trend of maximum daily rainfall for the forecast horizon 2050. The Senegal River Valley is particularly sensitive to floods, due to the large number of people who are settled on the riverside area and depend on recession agriculture. Therefore, as an adaptation measure to climate change, the flood monitoring and warning based on remote sensing has been considered. The monitoring of natural phenomena with remote sensing in extensive areas is a reliable and economical alternative. It is even more attractive when the means to conduct ground-based measurements are sparse and contain unreliable information, such as the Senegal River Valley. In this sense, a computational tool for floods detection, using satellite imagery was developed. However, the flood monitoring requires using satellite images with high temporal resolution due to its highly dynamic nature, with the drawback that such images usually have low spatial resolutions. In order to improve the spatial resolution, a Sub-pixels Analysis (SA) tool has been developed, taking into account the underlying digital topography, together with some derived attributes. The tool was applied to monitoring a flood event in the Senegal River Valley. A significant improvement in the flooded areas delineation with SA was achieved, compared with a supervised classification method that used the same satellite images.es_ES
dc.description.abstract[ENG] The Sudano Sahelian region in West Africa, one of the poorest in the World, is characterized by the great variability of rainfall and the rapid population growth. Given its climate vulnerability, extreme events of drought and rainfall cause extensive damage, which often involve property damage and human lives losses. The knowledge about plausible trends of extreme events is crucial from a policy perspective, for water resources management at basin level. Analyses based on observed data and obtained from Regional Climate Model (RCM) have revealed a nonstationary behaviour of hydrometeorological variables. The traditional frequency analysis suppose stationarity. Nevertheless, non-stationarity probabilistic models should be used to adjust the Probability Density Function (PDF) parameters, because have been detected significant trends in the historical data. Also, change projections of extreme events have shown great divergence between RCMs, increasing the forecast uncertainty. In this context, methodologies should be applied to take into account the variability provided by RCMs, as well as non-stationary nature of time series in order to draw risk maps of extreme events (drought and extreme rainfall). The GAMLSS tool (Generalized Additive Models for Location, Scale and Shape) has been chosen to model non-stationary nature of time series obtained from RCMs. GAMLSS is flexible to adjust the parameters of the selected PDF. It was decided to build PDF ensemble using bootstrapping techniques, to face the uncertainty problem generated by the models forecast divergence. The early papers presented in the compendium, ensembles were not built because a small number of models were available at that time. However, releasing to the scientific community all of the RCMs which were considered for Africa domain, by the ENSEMBLE European Project, the construction of ensembles was justified. Since then, the Thesis has been focused on identifying and developing robust and reproducible methods to build PDF ensemble of regional application. The weight of each RCM in the PDF ensemble was estimated via REA method (Reliability Ensemble Average). The REA factors were calculated according to a simple quantitative metric which measures the “distributional agreement” between the reference distribution (from historical data, or estimated for the future as a combination of RCMs), and distributions from data provided by RCMs. The Senegal River Basin has been the study zone. The Annual Maximum Dry Spell Length (AMDSL) and the Annual Maximum Daily Rainfall (AMDR) have been considered as work variables, with the objective of drawing risk maps and evaluating future impacts. The AMDSL were regionally analyzed, taking into account the identified ecoregions within the basin. The analysis has shown that, in general, the models predict increases in mean and standard deviation of the AMDSL. Interpolated maps were built from statistics derived from the PDF ensemble at each grid-site, together with confidence intervals for the AMDR. The difference between the maps for two reference years (1990 as reference year of change, and 2050 as a plausible prediction horizon) was assessed, and spatiotemporal trends of the different variables studied were obtained. The spatial distribution of trends shows that extreme precipitation will increase significantly in the Senegal River Valley, due to the increase of the AMDR mean and standard deviation at the region. The outcomes of the work, specially the maps with the spatial distributions, constitute real contributions to decision making processes, and to develop mitigation and adaptation strategies to cope with extreme events associated with climate change. In the Senegal River Valley, the results have shown a clear rising trend of maximum daily rainfall for the forecast horizon 2050. The Senegal River Valley is particularly sensitive to floods, due to the large number of people who are settled on the riverside area and depend on recession agriculture. Therefore, as an adaptation measure to climate change, the flood monitoring and warning based on remote sensing has been considered. The monitoring of natural phenomena with remote sensing in extensive areas is a reliable and economical alternative. It is even more attractive when the means to conduct ground-based measurements are sparse and contain unreliable information, such as the Senegal River Valley. In this sense, a computational tool for floods detection, using satellite imagery was developed. However, the flood monitoring requires using satellite images with high temporal resolution due to its highly dynamic nature, with the drawback that such images usually have low spatial resolutions. In order to improve the spatial resolution, a Sub-pixels Analysis (SA) tool has been developed, taking into account the underlying digital topography, together with some derived attributes. The tool was applied to monitoring a flood event in the Senegal River Valley. A significant improvement in the flooded areas delineation with SA was achieved, compared with a supervised classification method that used the same satellite images.es_ES
dc.formatapplication/pdfes_ES
dc.language.isospaes_ES
dc.publisherJuan Diego Giraldo Osorioes_ES
dc.rightsAtribución-NoComercial-SinDerivadas 3.0 España*
dc.rights.urihttp://creativecommons.org/licenses/by-nc-nd/3.0/es/*
dc.titleAnálisis de la variabilidad y tendencias de eventos extremos de precipitación en el contexto del cambio climático: desarrollo de una herramienta de seguimiento dinámico de inundacioneses_ES
dc.typeinfo:eu-repo/semantics/doctoralThesises_ES
dc.contributor.advisorGarcía Galiano, Sandra Gabriela 
dc.date.submitted2012-06-15
dc.subjectModelos Climáticos Regionaleses_ES
dc.subjectVariables hidrometeorológicases_ES
dc.subjectFunción de Densidad Probabilidades_ES
dc.subjectProbability Density Function (PDF)es_ES
dc.subjectCuenca del Río Senegales_ES
dc.subjectHidrologíaes_ES
dc.subjectSenegal River Valleyes_ES
dc.subjectReliability Ensemble Averagees_ES
dc.subjectENSEMBLE European Projectes_ES
dc.subjectRegional Climate Model (RCM)es_ES
dc.subjectHydrologyes_ES
dc.identifier.urihttp://hdl.handle.net/10317/2773
dc.contributor.departmentUnidad predepartamental de Ingeniería Civiles_ES
dc.identifier.doi10.31428/10317/2773
dc.rights.accessRightsinfo:eu-repo/semantics/openAccesses
dc.description.universityUniversidad Politécnica de Cartagenaes_ES


Files in this item

This item appears in the following Collection(s)

Show simple item record

Atribución-NoComercial-SinDerivadas 3.0 España
Except where otherwise noted, this item's license is described as Atribución-NoComercial-SinDerivadas 3.0 España