The socio-economic costs of drought are progressively increasing worldwide due to the undergoing alteration of hydro-meteorological regimes induced by climate change. Although drought management is largely studied in the literature, traditional drought indexes often fail in detecting critical events in highly regulated systems, which generally rely on ad-hoc index formulations and cannot be generalized to different contexts. In this study, we contribute a novel framework for the design of a basin-customized drought index. In contrast to the ad-hoc approach for index formulation, the proposed framework is generalizable and portable across different basins and contexts. The resulting index represents a surrogate of the state of the basin and is computed by combining the available information about the water circulating in the system to reproduce a target variable, namely the water deficit. The water deficit is formulated in order to be representative for the drought condition of the basin, for instance, as a supply deficit or a soil water deficit. In order to select the more relevant variables to explain the deficit and combine them into an index we use an advanced feature extraction algorithm called Wrapper for Quasi Equally Informative Subset Selection (W-QEISS). W-QEISS relies on a multi-objective evolutionary algorithm to find Pareto-efficient subsets of variables by maximizing the wrapper accuracy, minimizing the number of selected variables (cardinality) and optimizing relevance and redundancy of the subset. The accuracy is evaluated trough the calibration of a pre-defined model (i.e., an extreme learning machine) of the water deficit for each candidate subset of variables, with the index selected from the resulting solutions balancing high accuracy and low cardinality. Two case studies are considered, the Spanish Júcar river basin and the Italian Lake Como basin. The Júcar case study serves as validation for the proposed methodology, as an index with similar purposes (the State Index) has already been designed for the basin by experts knowledge, and has proven to be effective in triggering drought management measures. The index constructed by W-QEISS outperforms the official State Index in terms of accuracy and cardinality while maintaining similarities in the variables choice, thus aligning with the Spanish experts’ judgement. Building on these positive results, the same procedure is applied to the Lake Como case study, a regulated lake primarily operated for irrigation supply to four downstream agricultural districts. In the absence of an institutional drought monitoring system, we constructed the customized index for the Lake Como using all the hydrological variables from the existing monitoring system as well as the most common drought indicators at multiple time aggregations. The soil moisture deficit in the root zone, computed by a distributed-parameter water balance model of one agricultural district, is used as target variable. Numerical results show that the framework succeeds in constructing a combined drought index that reproduces the soil moisture deficit and is highly correlated to the annual agricultural yield. Further, to assess the operational value of the proposed index, we re-designed the operation of the lake dam by including the index among the inputs of the operating policy (Informed Policy) and by contrasting the corresponding improvement with the performance of a policy relying on the storage and the day on the year only (Baseline Policy). Results show that the Informed policy outperforms the Baseline policy of nearly 20%, thus the index represents a good candidate to both trigger and inform drought management strategies.
I costi socio-economici legati a magre e siccità sono in crescita a livello mondiale a causa delle alterazioni dei regimi idro-meteorologici indotti dal cambiamento climatico. Nonostante la gestione delle magre sia un tema molto studiato in letteratura, gli indicatori tradizionali risultano spesso inadeguati nell’individuare gli eventi critici in sistemi fortemente regolati, i quali generalmente fanno uso di indici ad-hoc non generalizzabili a contesti differenti. Questo studio si propone di formalizzare un approccio strutturato per lo sviluppo di un indice di magra bacino-specifico. A differenza dell’approccio ad-hoc per la formulazione di indici, l’approccio proposto è invece generalizzabile e applicabile a diversi bacini e contesti. L’indice risultante rappresenta un surrogato dello stato del bacino ed è ottenuto combinando l’informazione disponibile sulla disponibilità d’acqua nel sistema, nel tentativo di simulare una variabile-obiettivo, ossia il deficit idrico. Il deficit idrico viene formulato in modo che sia rappresentativo delle condizioni di magra del bacino, ad esempio come difetto di fornitura rispetto ad una domanda idrica o come deficit di umidità del suolo. La selezione delle variabili rilevanti per simulare il deficit, e del modo più efficace di combinarle in un indice, viene compiuta attraverso un avanzato algoritmo di estrazione di variabili chiamato Wrapper for Quasi Equally Informative Subset Selection (W-QEISS). W-QEISS basa il suo funzionamento su un algoritmo genetico a molti obiettivi per selezionare dei sottoinsiemi di variabili efficienti in senso paretiano, bilanciando le esigenze di massimizzazione dell’accuratezza, minimizzazione della cardinalità, e di ottimizzazione di rilevanza e ridondanza del sottoinsieme. L’accuratezza è valutata tramite la calibrazione di una classe di funzioni predefinita (i.e., una extreme learning machine) per la riproduzione del deficit idrico, ripetuta per ogni insieme di variabili candidato. L’indice viene selezionato tra le soluzioni risultanti, bilanciando accuratezza e cardinalità. Due casi di studio sono considerati: il bacino del fiume Júcar in Spagna e del lago di Como in Italia. Il caso di studio dello Júcar viene impiegato per la validazione della metodologia proposta, dal momento che un indice con un simile scopo (l’ Indice di Stato) è già stato definito per il bacino attraverso il giudizio degli esperti, e si è dimostrato efficace nell’azionare misure di gestione delle magre. L’indice costruito con il W-QEISS performa meglio dell’ Indice di Stato in termini di accuratezza e cardinalità, pur presentando somiglianze con quest’ultimo nella scelta delle variabili, e quindi allineandosi con il giudizio degli esperti. Sulla base di questi risultati positivi, la stessa procedura viene applicata al caso di studio del lago di Como, un lago gestito principalmente per l’irrigazione di quattro distretti agricoli a valle. Data la mancanza di un sistema istituzionalizzato di monitoraggio di magre, l’indice bacino-specifico è stato costruito utilizzando tutte le variabili idrologiche osservate attraverso il sistema di monitoraggio e i più comuni indicatori di magra a diverse aggregazioni temporali. La variabile-obiettivo utilizzata è il deficit di umidità del suolo nella zona delle radici della coltivazione, stimata attraverso un modello di bilancio idrico a parametri distribuiti di un distretto agricolo. I risultati numerici dimostrano l’efficacia dell’approccio proposto nel creare un indicatore combinato che riproduca il deficit di umidità del suolo e che sia anche altamente correlato al raccolto agricolo annuale. Inoltre, per quantificare il valore operativo dell’indice proposto, abbiamo ridefinito la politica di gestione della diga del lago includendo l’indice tra i suoi input (politica Informata) e comparando il miglioramento corrispondente con la performance di una politica basata solo sull’invaso e il giorno dell’anno (politica di Base). I risultati mostrano che la politica Informata ha una performance superiore di quasi il 20% rispetto alla politica di Base, dunque l’indice rappreseta un buon candidato per azionare e informare strategie di controllo delle magre.
Design and application of drought indexes in highly regulated Mediterranean water systems
ZANIOLO, MARTA
2015/2016
Abstract
The socio-economic costs of drought are progressively increasing worldwide due to the undergoing alteration of hydro-meteorological regimes induced by climate change. Although drought management is largely studied in the literature, traditional drought indexes often fail in detecting critical events in highly regulated systems, which generally rely on ad-hoc index formulations and cannot be generalized to different contexts. In this study, we contribute a novel framework for the design of a basin-customized drought index. In contrast to the ad-hoc approach for index formulation, the proposed framework is generalizable and portable across different basins and contexts. The resulting index represents a surrogate of the state of the basin and is computed by combining the available information about the water circulating in the system to reproduce a target variable, namely the water deficit. The water deficit is formulated in order to be representative for the drought condition of the basin, for instance, as a supply deficit or a soil water deficit. In order to select the more relevant variables to explain the deficit and combine them into an index we use an advanced feature extraction algorithm called Wrapper for Quasi Equally Informative Subset Selection (W-QEISS). W-QEISS relies on a multi-objective evolutionary algorithm to find Pareto-efficient subsets of variables by maximizing the wrapper accuracy, minimizing the number of selected variables (cardinality) and optimizing relevance and redundancy of the subset. The accuracy is evaluated trough the calibration of a pre-defined model (i.e., an extreme learning machine) of the water deficit for each candidate subset of variables, with the index selected from the resulting solutions balancing high accuracy and low cardinality. Two case studies are considered, the Spanish Júcar river basin and the Italian Lake Como basin. The Júcar case study serves as validation for the proposed methodology, as an index with similar purposes (the State Index) has already been designed for the basin by experts knowledge, and has proven to be effective in triggering drought management measures. The index constructed by W-QEISS outperforms the official State Index in terms of accuracy and cardinality while maintaining similarities in the variables choice, thus aligning with the Spanish experts’ judgement. Building on these positive results, the same procedure is applied to the Lake Como case study, a regulated lake primarily operated for irrigation supply to four downstream agricultural districts. In the absence of an institutional drought monitoring system, we constructed the customized index for the Lake Como using all the hydrological variables from the existing monitoring system as well as the most common drought indicators at multiple time aggregations. The soil moisture deficit in the root zone, computed by a distributed-parameter water balance model of one agricultural district, is used as target variable. Numerical results show that the framework succeeds in constructing a combined drought index that reproduces the soil moisture deficit and is highly correlated to the annual agricultural yield. Further, to assess the operational value of the proposed index, we re-designed the operation of the lake dam by including the index among the inputs of the operating policy (Informed Policy) and by contrasting the corresponding improvement with the performance of a policy relying on the storage and the day on the year only (Baseline Policy). Results show that the Informed policy outperforms the Baseline policy of nearly 20%, thus the index represents a good candidate to both trigger and inform drought management strategies.| File | Dimensione | Formato | |
|---|---|---|---|
|
2017-4-Zaniolo.pdf
non accessibile
Descrizione: Thesis text
Dimensione
3.8 MB
Formato
Adobe PDF
|
3.8 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/133677