Browsing by Author "Chamorro Chávez, Alejandro"

Now showing 1 - 1 of 1

Open Access
Stochastic and hydrological modelling for climate change prediction in the Lima region, Peru
(2015) Chamorro Chávez, Alejandro; Bárdossy, András (Prof. Dr. rer. nat. Dr.-Ing.)
Climate change has been an important field of research in the past years and certainly is a major concern in the present time. It involves a broad spectrum of subjects and significant different time scales, ranging from decades to thousands or millions of years. Generally speaking, in a climate change scenario a change in the pattern, average or extreme conditions of some variables is observed, and this can be due to many different causes as changing processes in the earth, human activities or extra terrestrial induced factors. This study concentrates on the influences on the climate due to human activities and focuses on the hydrological response to these influences or changes as a primarily goal, for the next few decades. The main motivation is the vulnerability and scarcity of the water availability in the capital of Peru, Lima, and how the area under study will respond to a change in the climate. An important focus of analysis in order to reduce the uncertainty in the predictions is the errors that appears when modeling a given variable or set of variables. This issue is addressed first in regionalization of precipitation and second in the calibration of hydrological models in which a robust parameter estimation is performed. In the first issue concerning to regionalization, External Drift Kriging is applied. In this part of the work the results of regionalization are analyzed with focus on the errors and systematic errors which appear during the modeling. The main goal here is the reduction of these errors through some proposed transformations. Here, three approaches are suggested, namely smoothing of the digital elevation model (DEM) considering a symmetric area, power transformation and smoothing considering a non symmetric area. The second issue concerning the uncertainty in the estimations (discharge) was addressed two-fold, namely by optimizing the objective function by means of a heuristic optimization procedure based on Monte Carlo simulation, and by means of a robust parameter estimation (ROPE) algorithm developed quite recently by Bárdossy and Singh, which in general terms can be used as a general multivariate optimization procedure. The algorithm offers a way of finding a set of “good” parameter vectors, which among other characteristics, are transferable in time. The final result comprises an ensemble of estimations for expected discharge variations accounting for the uncertainty in parameterization and processes description in the models. In this study HVB and HYMOD models are used. The assessment of the impact of climate change in precipitation and temperature is carried out by a statistical downscaling procedure based on a quantil-quantil transformation. Here the information given by the Global Climate Models (GCMs) outputs are transferred to the local scale. Two different GCMs and three scenarios are used in this step. This permitted the definition of a range for the expected future variations for temperature and precipitation. The last chapter of the study addresses the assessment of the discharge in the short term. The goal here is to “infer” the outcome of a random variable (discharge) in the next time step by taking information from past observations (previous steps). As we can regard the observations (time series) as a realization generated from a stochastic process, we can address this issue from a stochastic point of view. The task is addressed first by considering some of the existing autoregressive models (AR process), and second by considering a Copula-based autoregressive model. In order to perform the Copula-based autoregressive model, a given time series (modelled discharge) was transformed into three vectors representing the same original time series but shifted in time. A three dimensional Copula was then fitted to the univariate distributions. For this, a Gaussian model as well as a Beta kernel model expressed in terms of the Beta function was considered.