Back to search

Democritus University of Thrace (DUTH)

Ασαφή υβριδικά μοντέλα για την ανάλυση και αξιολόγηση ακραίων υδρολογικών φαινομένων

Abstract

dc:description

Hydrological extremes events, such as floods and droughts, are natural disasters that are caused by climatic and hydrometeorological disturbances. They have a direct or indirect impact not only over the ecological systems but also on a wide spectrum of operations of anthropogenic systems and they are the cause of loss of human lives.In particular, over the past few decades there is a growing concern in the scientific community about the catastrophic dimensions of these phenomena as increasing trends have been observed both in terms of the frequency of their occurrence and the severity of flood events and drought periods.Fuzzy logic and the theory of fuzzy sets is a relatively new mathematical consideration with which multiple problems related to these phenomena can be approached. Lotfi A. Zadeh (1921-2017), an Iranian−American electrical engineer, was the first to introduce fuzzy logic in 1965. Fuzzy logic is a many−valued logic where variables can take infinite values in a range between 0 and 1 (including the limit values) thus generalizing the classic two−valued logic and classic sets. It has the ability to simulate the human way of thinking either by developing autonomous tools or while used in combination with other methods, and is widely applied in different scientific fields.In this direction, the present doctoral dissertation develops and applies the following innovative hybrid methodologies which are based on the principles of fuzzy logic and the theory of fuzzy sets.A hybrid fuzzy probabilistic methodology was developed based on fuzzy linear regression method and the frequency factor method in order to couple the observed probabilities and the theoretical probability distribution which is based on the historical sample. Initially, the frequency factor KT was calculated with the use of the empirical probabilities in assumptions of a theoretical probability distribution, and then it was carried out a fuzzy linear regression between the considered physical variable (dependent variable) and KT (independent variable). Additionally, it was achieved simultaneously a fuzzy estimation of the mean value and the standard deviation. The proposed methodology was applied to data sets of annual cumulative rainfall and annual cumulative streamflow in order to analyse and classify meteorological and hydrological drought, respectively. It can also be applied to the frequency analysis of flood events by additionally taking into account the probability distribution of the maximum type I (Gumbel distribution). Furthermore, it was achieved a fuzzy classification of meteorological and hydrological drought which was based on the produced fuzzy linear relations.For the analysis and classification of hydrological and meteorological drought, it was used the Tanaka model (1987) as well as a modified fuzzy linear regression model regarding the objective function. Both of the models result to a constraint optimization problem where all the observed data must be included in the produced fuzzy band. Thus, from a mathematical point of view, the problem of fuzzy linear regression leads to a linear programming problem, whilst the constraints express the requirement that all the data must be included within the produced fuzzy band. In other words, the produced fuzzy band contains the entire sample. However, a large band leads to no functional results. Therefore the magnitude of the fuzzy band can be selected as a measure of suitability and furthermore, as objective function.The uncertainty of the system is incorporated into the regression coefficients which are considered to be symmetric triangular fuzzy numbers (STFN). Another two measures of suitability are developed in this Thesis.Annual rainfall series from 1974−’75 to 2009−’10 (in terms of hydrological years), provided by the Directorate for Water of the Decentralized Administration of Crete, were utilized in order for meteorological drought to be analyzed. The precipitation records were obtained from the meteorological station (M.S.) of Gortyna located at the southern part of the Prefecture of Heraklion.Tanaka model was applied so that data with the theoretical probability distributions of normal (N) and Pearson type III (P III) (Equation 1), and lognormal (LN) and log Pearson type III (LP III) (Equations 2). (1) (2)where , are the regression coefficients which are symmetric triangular fuzzy numbers. The model suitability was checked, moreover, through the measures δ1 and δ2 expressed in the following Equations (3) and (4), respectively: (3)where , are the central values of the fuzzy regression coefficients, while are the unbiased estimations of the mean value and the standard deviation of the sample for the log−transformed data. (4)where , and denotes the left boundary, the right boundary and the central value of each fuzzy estimation of lnR, while is the sample mean of the log−transformed data of rainfall. As the value of measure δ1 gets lower and the value of measure δ2 gets higher, a higher model suitability is accomplished The results of the fuzzy linear regression based on Tanaka model, the suitability measures, and the mean value and the standard deviation of the historical sample are presented in the bellow Table 1.Table 1. Summary table with the results of the fuzzy linear regression and the suitability measures in the case of lognormal and log Pearson III distributions. The mean value and the standard deviation of the historical sample are presented as well.In this case study, the best coupling between the observed probabilities which are based on the historical sample with the theoretical ones was achieved in the assumption of log Pearson type III distribution, however normal distribution and Pearson type II distribution also described well the historical sample. Whereas, in the case of lognormal distribution, both the objective function J and the suitability measures δ1 and δ2 got the worst values. Figures 1 illustrates the fuzzy linear regression in P III and LP III distributions. The conventional linear regression is presented as well in Figure 1.Figure 1. Fuzzy linear regression (based on Tanaka model) between the annual cumulative rainfall and the frequency factor KT in the assumption of a) Pearson type III distribution and, b) log Pearson type III distribution.The predefined thresholds of drought categories were fuzzified on the basis of the produced fuzzy relations. More analytically, based on the widely used standardized normal thresholds to drought, the corresponding (fuzzy) annual cumulative discharge (and rainfall) thresholds are determined.Ιn LN and LP III distributions, fuzziness of the drought thresholds increases for low rainfall values because of the increase of KT (in terms of absolute value). This does not happen in N and P III distributions because in the present application, the regression coefficients of KT resulted to crisp numbers. Figure 2 illustrates the fuzzified thresholds in the case of P III and LP III distributions. Figure 2. Fuzzyfied thresholds of drought categories in a) Pearson III distribution and b) log Pearson III distribution. (5)where is each examined (fuzzified) drought threshold k and yj is the observation which is compared to the examined drought threshold k. In Figure 3 below, the nominator of measure Gk,j denotes the hatched area included between the crisp observation and the membership function of the examined drought threshold. The denominator of Gk,j denotes the entire area included by the membership function of the examined drought threshold. Thus, measure Gk,j takes into account all information of the membership function. The examined observation (log−transformed annual cumulative rainfall) is lower than the examined drought threshold since Gk,j <0.5, whereas, when Gk,j>0.5 then the observation is greater than the examined fuzzified drought threshold.Figure 3. Comparison between the fuzzified drought threshold (Z=−1.5) with the (crisp) observation of lnR regarding the hydrological year of 1992−1993. Table 2 below presents the classification of meteorological drought in log Pearson III distribution, where most hydrological years are characterized as mild drought years and mildly wet years, and only one year is characterized by extreme drought.Table 2. Classification of drought in the case of log Pearson III distribution. Regarding hydrological drought, it was utilized the annual cumulative streamflow of the transboundary Evros (Maritsa or Meric) river at Pythio’s bridge. Pythio’s bridge is considered to be a representative position of almost the entire basin of Evros (Maritsa or Meric) river. Annual cumulative discharges were calculated based on the monthly mean values (Angelidis et al., 2010) and refer to the period from 1985−’86 to 2006−’07 (in terms of hydrological years).Two fuzzy linear regression models were performed: a) Tanaka model with the objective function J and b) the modified fuzzy regression model with the objective function F, in the assumption that the log−transformed annual cumulative streamflow (lnV) follows normal distribution (assumption of LN distribution) and Pearson III distribution (assumption of LP III distribution). The historical sample was described better with log Pearson type III distribution than with lognormal distribution as far as both of the fuzzy linear regression models are concerned, while both Tanaka model and the modified fuzzy linear regression model produced similar results. Table 3 presents the total results and Figure 4 illustrates the fuzzy linear regressions which are based on the modified fuzzy linear regression model in the assumption of LN and LP III distributions.Table 3. Unbiased estimation of the mean value and the standard deviation, fuzzy coefficients and the suitability measures for both of the fuzzy regression models in the assumption of lognormal and log Pearson type III distributions.Figure 4. Observations, fuzzy and the conventional regression based on the modified fuzzy linear regression model in the assumption of a) LN distribution, b) LP III distribution.The historical sample was better described with log Pearson type III distribution than lognormal distribution in both of the fuzzy linear regression models, while both Tanaka model and the modified fuzzy linear regression model produced similar results. Table 4 presents the classification of hydrological drought in LN and LP III distributions for both of the fuzzy linear regression models. This classification was achieved by comparing each observed lnV to each fuzzified drought threshold by using the above mentioned measure G. Figure 5 illustrates the fuzzified thresholds of drought categories, while, the classification of the hydrological year ’94−’95 is also presented for illustration purposes.Table 4. Hydrological drought classification.Figure 5. Fuzzy thresholds of drought categories and (crisp) annual cumulative discharge for the hydrological years '94–'95 based on the modified fuzzy regression model in the case of a) lognormal distribution b) log Pearson III distribution.➢The hybrid fuzzy statistical method of fuzzy estimators was applied in the analysis and classification of hydrological drought in order to fuzzify Streamflow Drought index (SDI). The mathematical methodology of fuzzy estimators translates statistical information into fuzzy sets and enables the application of arithmetical operations via fuzzy sets. The unbiased estimators of the mean value and standard deviation of the sample are transformed from crisp numbers to fuzzy numbers because of a simple transformation of a 1−γ confidence interval to α−cuts (in the present application for γ=0.05, γ=0.1 and γ=0.2). By this way, the uncertainty arising from the point estimate of these parameters is better interpreted. Thus, with the use of the extension principle, algebraic operations among these statistical parameters can be performed even if they are described by different probability distributions, something which is too complicated in probability theory.The classification of hydrological drought was achieved by comparing fuzzified SDI to each (crisp interval) drought category. For this purpose, it was used a fuzzy measure of inclusion (measure C) which denotes the degree of inclusion of the fuzzified SDI into the examined drought category (Equation 6). Measure C(i,j) is based on measure of Kosko (1986) and has been adapted for this Thesis. It takes into account all information of the membership function as well. (6)The nominator of measure C(i,j) expresses the area included between the left and right thresholds of the examined drought category and the membership function of SDI (Figure 6). The denominator is the entire area defined by the membership function. In other words, it could be argued that measure C(i,j) expresses the proportion of the area of the fuzzified SDI included among the thresholds of drought category j. The sum of measures C(i,j) for all drought categories, as far as the same hydrological year i is concerned, is equal to unit.Analysis was carried out on the same sample of the annual cumulative streamflow of Evros (Maritsa or Meric) river. Moreover, fuzzified SDI was calculated for the reference periods of k=1 (Oct−Dec), k=2 (Oct−Mar) and k=3 (Oct−Jun) and for the hydrological year i. Table 4 below presents the results of the classification of annual hydrological drought (k=4, Oct−Sep) based on the fuzzy measure of inclusion Cj. Fuzzified SDI belongs to the category in which the fuzzy inclusion measure takes its highest value.Table 4. Classification of the annual hydrological drought based on the values of measure C.According to Table 4, there are some notably wet years whilst the majority of dry years are characterized as years of mild drought. However, some years, such as the hydrological year of 1993−’94, are not definite and not easy to be characterized since the fuzzyfied annual is considerably extended to three categories (Figure 6). It is worth mentioning that the classification of hydrological drought presented in Table 4 is similar to the results derived when the hybrid fuzzy frequency factor−based methodology is applied.Figure 6. Fuzzified annual SDI during 1993−'94 is extended to three drought (crisp) categories.Figure 7 presents the historical sequence of the fuzzified annual (reference period of k=4) SDI for γ=0.05 and γ=0.2, where it is observed that fuzziness increases as the γ−value decreases. In Figure 8 fuzzified SDI indices, for the reference period k=1, k=2 and k=3, are presented when γ=0.05. Analysis of the reference periods shows that the hydrological regime may change significantly during the hydrological year. Figure 9 presents a negative and a positive displacement of fuzzified SDI within the hydrological years 1993−’94 and 2004−’05, respectively. Figure 7. Fuzzified indices for a) γ=0.05 and b) γ=0.2.Figure 8. Fuzzified for reference periods of a) k=1, b) k=2, c) k=3 of all years (γ=0.05).Figure 9. Displacement of fuzzified indices to a) lower values during the moderate drought year of 1993−‘94 and b) higher values during the year of 2004−’05 characterized as non−drought year.➢The method of multiple fuzzy linear regression was applied in order for a fuzzy linear relation to be determined between groundwater and the main hydrometeorological variables of drought. The method was applied in two aquifer systems characterized by intense heterogeneity of their geological formations.The first case study is about a shallow unconfined aquifer at the eastern part of the Nestos river Delta in the Prefecture of Xanthi, Greece. In this study, a small sample size of observed groundwater depth of four wells is related to the hydrometeorological records at the same point in time j (Figure 10).Figure 10. Piezometric map of the unconfined aquifer at the eastern part of the Nestos river Delta and the examined wells (October 2008).The depth of groundwater (DGW) measured every three months (Gkiougkis, 2018) was related to the monthly records of rainfall (R), potential evapotranspiration (PET) and the three−months mean value of streamflow (Qmean) from October 2006 to October 2008 (Equation 7). Both Tanaka model (FLR−1 model) and modified fuzzy linear regression model (FLR−2 model) were used. Apart from values of the objective functions of the models, two more suitability measures were used: measure δ2 and a fuzzified version of the statistical measure Theil’s U (Equation 8). This measure is separately determined for each regression model based on the extension principle. Since suitability measure U is a fuzzy number, a fuzzy measure was used in order to determine which fuzzy linear regression model corresponds to the lowest value of the suitability measure U. As U−value gets lower, suitability of fuzzy linear regression model gets higher. (7)where are the fuzzy regression coefficients (which are STFN with a central value r and a semi−width w) of the constant term, the rainfall, the potential evapotranspiration and the streamflow, respectively. (8)where is the fuzzy output (fuzzy groundwater depth which is produced based on each fuzzy linear regression model) and, is the observed groundwater depth at the examined point in time j (which is crisp number).Tables 5, 6 below present the coefficients of the multiple fuzzy linear regression based on both of the models (FLR−1 and FLR−2) as far the examined wells are concerned. Table 7 presents the values of the objective functions and the values of the suitability measures. In Table 7, the terms LU, CU and RU denote the left boundary, the central value and the right boundary of the fuzzified statistical measure Theil’s U.Table 5. The fuzzy coefficients regarding the examined wells based on FLR−1 model.Table 6. The fuzzy coefficients regarding the examined wells based on FLR−2 model.The two multiple fuzzy linear regression models produced similar results. Most of the regression coefficients resulted to crisp numbers except the constant term. The multiple fuzzy linear relations having the lowest fuzziness were produced in well 194, while fuzziness increases as the distance from the coastal increases. In addition, the contribution of the streamflow increases as the distance from the coastal decreases. This can be explained because there are more coarse−grained materials at that area.Table 7. Suitability measures and the values of the objective functions J and F for both of the fuzzy linear regression models.It is worth mentioning that both FLR−1 model and FLR−1 model were additionally performed by using the observations of R and PET with a time lag of one, two and three months. However fuzziness got higher even from fuzziness of the worst well which is 177. Figure 11 graphically illustrates the results of the multiple fuzzy linear regressions for both of the models regarding well 194. All observed data are included into the produced fuzzy band. For illustration purposes, only fuzzy relation between DGW−PET is presented. Figure 12 presents the comparison between fuzzified measure U1 and U2 produced of results of FLR−1 and FLR−2, respectively, in the best well which is 194. Fuzzified measure U2 was marginally smaller than the U1 which means that FLR−2 worked marginally better than FLR−1.Figure 11. Graphic representation of the relation between DGW and PET according to the results of the applied multiple fuzzy linear regression in well 194, based on a) FLR−1 model and b) FLR−2.Figure 12. Comparison between the fuzzified Theil’s inequality coefficients and in well 194. The second case study is the aquifer system of the agricultural plain at the southeast of the city of Xanthi in the Prefecture of Xanthi, NE Greece (Figure 13), where the water table was directly related to meteorological drought by using Standardized Precipitation Index (SPI) and standardized version of Reconnaissance Drought Index (RDISt). The aquifers located to the upper geological formations of the study area are in the spectrum from unconfined up to semi−confined, and become less permeable and with a SE or E direction as Kosynthos river moves away and Vistonida Lake approaches. The water table values (WT) have been recorded by two water table loggers installed near Nea Kessani settlement (altitude 10m) and Vafeika settlement (altitude 19.5m) which are considered to be representative points of the wider area (Pliakas et al. 2015).Figure 13. Geological formations of the study area become from a fine up to silty clay in NW−SE direction towards Vistonida Lagoon. Drought indices were calculated based on monthly precipitation and temperature data available from 1996−’97 to 2015−’16, for several reference periods (in hydrological years, i.e. October−September). Then, based on Tanaka model, each value of the water table from 1998−’99 to 2006−’07 was related, separately, to the values of each drought index for the same time period. Apart from objective function J, the suitability of the model was checked by using objective function F of the modified fuzzy linear regression model as a suitability measure. In addition, it was used a fuzzy measure FD, which is a fuzzy version of a statistical measure similar to the determination coefficient R2. In contrast to measure F, the value of the fuzzy measure is desired to be the maximum. After several tests, the most satisfactory fuzzy linear regression that was achieved in both of the areas (Nea Kessani and Vafeika) has the following forms: (9) (10)where is the fuzzy estimate of the water table of January regarding the hydrological year i, are the values of the drought indexes for the reference period from October to December regarding the hydrological year i, and are the annual drought of the previous hydrological year i−1. The fuzzy regression coefficients were selected to be STFN. Based on the results of each multiple fuzzy linear regression described by the Equations (9) and (10), the fuzzy measure FD was calculated regarding the areas of Nea Kessani and Vafeika as follows: (11)where is the fuzzy estimate of water table, while is the observed water table regarding the hydrological year i, while the term of denotes the sample median.Algebraic operations of Equation (11) above are performed based on the extension principle. From the theorem of global existence for maxima and minima of functions with many variables, it is known that, if the domain of a real function is closed and bounded and the real function is continuous, then such a function will have its absolute minimum and maximum values at some points in the domain. It can be proved that the right boundary of the will always be equal to unit when the possibilistic model of Tanaka (1987) is used. The maximum value of Equation (11)is obtained when . This always holds since each observation is included into the support of the corresponding fuzzy estimate due to the inclusion constraints .In order to determine the greatest fuzzy measure FD, it was used a fuzzy comparison measure S (Nguyen, 2017) which takes into account the entire membership function. The higher the S−value the higher measure FD is and thus indicating the most satisfactory multiple fuzzy linear regression.The results of Equations (9) and (10), the suitability measures, and the values of the comparison measure regarding the areas of Nea Kessani and Vafeika are presented in the following Table 8. The results of the area of Nea Kessani are better than the area of Vafeika, while the results based on RDISt values are better than the corresponding ones based on SPI values in both of the areas. The drought of the first trimester significantly affects water table of January, while the contribution of the annual drought of the previous hydrological year is also significant. Figure 14 illustrates the fuzzy linear regression regarding the area of Nea Kessani in a 2D representation while in Figure 15, the results are presented in a 3D visualization. Figure 16 presents the comparison between fuzzy measure FD that was produced based on the results of Equations (9) and (10) regarding the area of Nea Kessani.Table 8. Total results of the multiple fuzzy linear regressions for both of the fuzzy linear regression models.Figure 14. Results of the multiple fuzzy linear regressions regarding the area of Nea Kessani area a) 2D representation between —WTi,Jan, b) 2D representation between —WTi,Jan, c) 2D representation between —WTi,Jan, d) 2D representation between —WTi,Jan.Figure 15. Results of the multiple fuzzy linear regressions regarding the area of Nea Kessani a) 3D visualization based on RDISt—WT, b) 3D visualization based on SPI—WT.Figure 16. Comparison between fuzzy measures concerning the multiple fuzzy linear regression with regard to the historical sample of the area of Nea Kessani based on RDISt (blue curve) and SPI (red curve) multiple fuzzy linear regressions.➢A hybrid fuzzy multicriteria model was developed. This model combines the widely used multicriteria method Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) with the classification method of fuzzy pattern recognition in order to assess vulnerability of systems to floods and droughts. The assessment of the vulnerability contains many different aspects and hence, the analysis can be performed using multicriteria methods. Fuzziness can be used in multicriteria analysis not only to express the uncertainty but also to provide a human−thinking similar method to combine different criteria.The decision is obtained through the measure (Equation 12), which denotes the degree to which alternative j pertains to the ideal solution by simultaneously taking into account the distances both of the ideal and anti−ideal solution. A basic property that must be verified in the fuzzy pattern recognition is that the sum of the membership values of the ideal and anti−ideal solution is equal to one (Equation 13) regarding each alternative. Measure is preferred to the measure used in conventional TOPSIS because its mathematical background is based on optimization. Here, fuzziness is used for a better interpretation of the multicriteria model and not in order to incorporate fuzzy information. (12)where are the distances of alternative j from the and the ideal and anti−ideal solution, respectively. (13)The methodology was applied in order to assess four alternatives regarding their vulnerability to floods. The case study refers to the Greek parts of the Nestos and Strymonas transboundary river basins where each was divided in areas with elevation <100m and areas with elevation >100m (Figure 17). Figure 17. Illustration of the case studies areas and the subdivision of the areas to the elevation threshold of 100 meters.The concept of vulnerability is composed of the criteria of exposure, sensitivity and the adaptive capacity which, in turn, are composed of eleven socioeconomic sub−criteria (Table 9), while the initial weights of the criteria and sub−criteria were predefined (Jun et al., 2013).It is noted that the proposed methodology can be used for the assessment of vulnerability to drought giving the possibility of classification to non−ordered categories.Table 9.Description and weights of utilized proxy variables (sub−criteria).Table 10 presents the weights of the main three criteria and the rating of the alternatives after the first composition of the sub−criteria. The final evaluation of the examined alternatives based on decision measure is presented in Table 11, while the distances of the ideal and anti−solution are presented as well. Area 1 (A1) and Area 2 (A2) represent the areas with elevation <100m and >100m, respectively, regarding the river basin of Strymonas river, while the Area 3 (A3) and Area 4 (A4) represent the areas with elevation <100m and >100m, respectively, regarding the river basin of Nestos river. The alternative with the highest vulnerability to floods is areas with elevation <100m regarding the river basin of Strymonas River (A3). The final rank ordering from the most to the least vulnerable alternative is A3>A1>A4>A2.Table 10. Pay−off matrix of the examined areas after the first composition of the proxy variables (sub−criteria).Table 11. Distance measures from the «ideal» and «anti−ideal» solutions and the membership degree of each examined area (alternative) at the «ideal solution» (high vulnerability).It is mentioned that the ranking is the same either if the maximum membership degree of each area at the ideal solution (μj+) is followed or following the minimum membership degree at the anti−ideal solution (μj−) because of Equation (9). It should be mentioned that generally this does not stand for the widely used TOPSIS.A hybrid fuzzy multicriteria methodology was developed to select the most discrete preferable recharge site in order to apply managed aquifer systems (MAR) by utilizing floodwaters. This was conducted in the context of a preventive planning against drought as green solution. For this purpose, it was used a fuzzy version of Analytic Hierarchy Process (FAHP) and Mamdani−type Fuzzy Inference Systems (FIS). Fuzziness was introduced in order to interpret uncertainty in the process of determining the criteria weights and the rating of the alternatives.The fuzzy version of the widely used multicriteria method of AHP that was applied is the logarithmic fuzzy preference programming (LFFP) because it guarantees a unique and optimist solution for the determination of criteria weights. Based on FAHP−LFFP method, crisp criteria weights are derived from a fuzzy pairwise comparison matrix. Pairwise comparisons between criteria are carried out based on a fuzzy scale which denotes the (fuzzy) dominance of one criterion onto the other (Table 12, Figure 18). Finally, from a mathematical point of view, the problem results to an optimization problem. Mamdani−type FIS is an autonomous tool of fuzzy logic based on a fuzzy rules if−then based system. With the use of FIS, imprecise knowledge and information result to a quantitative fuzzy conclusion. Mamdani−type FIS was used to fuzzify the classes to which the criteria were divided. It was also used because the value function of each criterion was unknown.The proposed methodology was applied in the aquifer system of the agricultural plain located at the southeast of the city of Xanthi in the Prefecture of Xanthi, NE Greece. The aquifer system is hosted in geological formations of high heterogeneity consisting mainly of loam, clays, sands, gravels, etc. Seven alternatives were evaluated under the consideration of nine criteria (Table 13).Table 12. Triangular fuzzy numbers (TFN) for the pairwise comparison of the criteria regarding their relative importance.Figure 18. Intensities of the relative importance between two criteria presented as triangular fuzzy numbers (TFN).Table 13. Crisp weights of criteria and the rating of alternatives with respect to the classes of criteria. Crisp classes have been fuzzified in FIS.The examined criteria were divided into classes based on the expert’s knowledge of the case study and then these classes were fuzzified with the use of FIS. An example of the form of the fuzzified classes is presented in the following Figure 19 where fuzzy classes regarding the criteria of storativity and transmissivity are illustrated.Figure 19. Fuzzy classes described by linguistic variables regarding the criteria of a) storativity, b) transmissivity.The final ranking order of the alternatives is presented in Table 14 and in Figure 20 below. In the rating estimation of each alternative where FIS was used, the widely used fuzzy implications of algebraic product and fuzzy min implication were both used.Most preferable sites are located at the north−western part of the study area, near Kosynthos river, and the most preferable is Al5 alternative. This mainly holds due to the hydrogeological conditions of the north−western part which are more favourable than the south−eastern part as far as groundwater recharge is concerned. However, this is not an absolute condition due to the intense diversity of geological formations. In addition, the final rank list was also affected by the contribution of the other criteria.The proposed methodology can be applied either on small or large samples. Its applicability requires lithological profile and hydraulic characteristics for both vadose and saturated zones, while it can be applied only in unconfined aquifers and where there is an underlying recharge axis. Furthermore, applying the selected type MAR requires the presence of a river whose excess waters can be utilized.Table 14. Final evaluation of the alternatives based on the fuzzy implication of algebraic product and fuzzy min implication.Figure 20. Final evaluation of the alternatives (recharge sites for applying MAR).In summary, the PhD thesis develops the following hybrid fuzzy methodologies in order to analyse and assess the extreme hydrological phenomena of flood and drought:Application of a hybrid fuzzy probability model to interpret the uncertainty in the coupling of the historical sample with the theoretical probability distribution in data sets of rainfall, maximum discharge and streamflow with the use of fuzzy regression and by applying the proposed measures of suitability.Application of a hybrid fuzzy statistical model to fuzzify SDI (Steamflow Drought Index), which is used to estimate intensity and categorize hydrological drought. Fuzziness is introduced into the mean and the standard deviation of the sample in order to interpret uncertainty of their point estimate. From a mathematical point of view, the approach is based on the method of fuzzy estimators and the extension of principle.Use of a multiple fuzzy linear regression model in order to determine the relationship between drought and groundwater aquifers hosted in strongly heterogeneous formations. The model applies when there is low data availability.Use of a hybrid fuzzy multi−criteria model to assess the vulnerability of systems to flood and drought since the concept of vulnerability contains many dimensions. Fuzziness is used to improve interpretation regarding the synthesis of criteria. Assessment of vulnerability contributes to risk assessment in the context of planning green solutions for the prevention, protection and preparedness of systems against these phenomena.Development of a hybrid fuzzy multi−criteria methodology for the selection of favorable sites to implement artificial groundwater recharge by utilizing floodwaters. This conjunctive use of excess surface water and groundwater is a green measure in the context of drought prevention. Fuzziness is introduced to express uncertainty which arises in the process of determining the criteria weights and the rating of alternatives. From a mathematical point of view, the approach is based on the methods of both FAHP−LFFP and Mamdani−type FIS.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2022

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Papadopoulos, Christoforos
  • Παπαδόπουλος, Χριστόφορος

Subjects

dc:subject × 36

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/51180

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Papadopoulos, Christoforos; Παπαδόπουλος, Χριστόφορος. Ασαφή υβριδικά μοντέλα για την ανάλυση και αξιολόγηση ακραίων υδρολογικών φαινομένων. Democritus University of Thrace (DUTH), 2022. http://hdl.handle.net/10442/hedi/51180