In the last decades, water resources became increasingly stressed by different phenomena, including climate change, pollution, industrialization, population growth and urbanization. Along with population increase, urbanization is significantly changing the spatial distribution of water demands. Urban water distribution systems must meet an increasing demand and, in response to that, urban planners need to improve urban water management plans and develop resilient water distribution systems. Understanding and forecasting water demand become a very important issue in different sectors of urban planning, and as demonstrated by the increasing number of the studies on urban water consumption. The main objective of these studies is to detect models able to able to provide an exhaustive description of the urban water use and to analyze which and how different external factors influence water use and the changes, in order to collect much information as possible. The first phase of this work focuses on detection of the changes in urban withdrawal between the 80s and 2000 in Milan. It results that the withdrawal of the city overall decreased, during the considered period. Also, the withdrawals of the lifting stations became more stable in the second part of the studied period. It results that there were not any events which influenced the city overall withdrawals, but around the 1990 and the 2010 numerous suburbs stations showed changes in the withdrawal trend, probably as response to some urban population movements. The second phase focused on attribution, meaning to investigate which variables mostly influenced the withdrawals of the stations. The withdrawals of four lifting stations were modelled following the Box-Jenkins procedure. The selected stations were Corsico, Crescenzago, Salemi, Suzzani: these stations were chosen according to monthly data availability. The withdrawals were modelled first considering different intervals, according to change in the trend, then considering the complete dataset. The best models turned out to be the AutoRegressive models AR(p) of order 2 or 3. Considering the complete dataset all these models showed values of R2 greater than 0.60 both in calibration and validation phases, while considering the different intervals, some intervals showed very low R2 coefficient (less than 0.20). In order to investigate the drivers of the changes in withdrawals and improve the models, some external variables were added to the models. In particular, the withdrawals of the nearest stations, the monthly temperature and the monthly precipitation were added. The level of cross-correlation of the external variables with past withdrawals were low: these levels showed that the variables could not give a significant improvement to the models. In fact, the obtained ARX models confirmed this assumption. Comparing the model performances, the withdrawals of the nearest stations gave a slightly better performance than the other variables, while between the climate variables the higher contribution was given by the number of exceptionally hot days. Instead there was not a clear response to the question if it is better to use the amount of precipitation or its occurrence in the month, because each station showed different response. According to the obtained results, some aspects of the work need further developments. For example, having access to data series at high resolution, both in temporal and spatial resolution, should improve the analysis of urban water consumption. Also, it may be test if using a more sophisticated kind of model (e.g. the Artificial Neural Networks) it is possible to obtain better results in terms of precision.
Negli ultimi decenni le risorse idriche sono state notevolmente stressate da diversi fenomeni quali il cambiamento climatico, l'inquinamento, l'industrializzazione, l'aumento della popolazione mondiale, l'urbanizzazione. Insieme all'aumento della popolazione, l'urbanizzazione sta modificando la distribuzione spaziale della domanda d'acqua. I sistemi di distribuzione idrica urbana devono andare incontro a una domanda sempre crescente e, di conseguenza, i pianificatori hanno bisogno di migliorare i piani di gestione delle risorse idriche urbane e di sviluppare sistemi di distribuzione idrica più elastici. La necessità di comprendere e prevedere la domanda di acqua assume un ruolo importante in diversi settori della gestione urbana e, di conseguenza sono aumentati gli studi che analizzano il consumo urbano di acqua. Questi studi hanno l'obiettivo di individuare modelli in grado di fornire una descrizione esaustiva dell'uso di acqua urbano e di investigare come i differenti fattori esterni influenzino l'uso d'acqua e i cambiamenti al fine di ottenere il maggior numero di informazioni. La prima parte di questo lavoro è concentrata sull'investigazione dei cambiamenti nel prelievo d'acqua urbano di Milano tra gli anni 80 e il 2000. Risulta che il prelievo della città è diminuito lungo il periodo studiato. Inoltre, i prelievi delle singole stazioni di sollevamento si sono stabilizzati nella seconda parte del periodo di studio. Risulta che non ci sono stati eventi che hanno influenzato i consumi generali della città, ma tra il 1990 e il 2010 numerose stazioni periferiche hanno mostrato cambiamenti nel trend del prelievo, probabilmente in risposta a delle migrazioni interne della popolazione urbana. La seconda parte è concentrata sull'individuazione delle variabili che hanno maggiormente influenzato i prelievi delle stazioni. I prelievi di quattro stazioni sono stati modellati seguendo la procedura Box-Jenkins. Le stazioni selezionate sono Corsico, Crescenzago, Salemi, Suzzani: le stazioni sono state scelte sulla base della disponibilità dei dati di prelievo mensili. I prelievi sono stati modellati prima considerando diversi intervalli, creati sulla base dei cambiamenti di trend, poi considerando il dataset completo. I modelli migliori si sono rivelati i modelli AutoRegressive AR (p) di ordine 2 o 3. Considerando il dataset completo tutti i modelli hanno mostrato valori di R2 superiori a 0.60 sia in calibrazione che in validazione, mentre considerando i diversi intervalli in alcuni si è ottenuto un coefficiente R2 basso (minore di 0.20). Per indagare i principali driver dei cambiamenti nei prelievi e migliorare i modelli, alcune variabili esterne sono state aggiunte ai modelli. In particolare, sono stati aggiunti i prelievi delle stazioni più vicine, la temperatura mensile e le precipitazioni mensili. La cross-correlazione tra le variabili esterne e i prelievi era bassa, quindi introdurre le variabili non avrebbe portato a un miglioramento significativo dei modelli. Infatti, i modelli ARX ottenuti hanno confermato questa ipotesi. Comparando le performance dei modelli, i prelievi delle stazioni più vicine a quelle modellizzate hanno ottenuto una performance leggermente migliore rispetto alle altre variabili, mentre tra le variabili climatiche il contributo maggiore è dato dal numero di giorni del mese eccezionalmente caldi. Invece non c'è una risposta chiara alla domanda se è meglio usare la quantità di precipitazione o la sua occorrenza nel mese, perché ogni stazione ha mostrato un comportamento diverso. Sulla base dei risultati ottenuti, alcuni aspetti di questo lavoro richiedono ulteriori sviluppi. Per esempio, poter accedere a serie di dati a risoluzione maggiore, sia in termini temporali che in termini spaziali, permetterebbe di analizzare con miglior precisione il consumo idrico della città. Inoltre, si potrebbe verificare se modellizzare i dati con un altro tipo di modello, più sofisticato (per esempio le reti neurali artificiali), permetterebbe di raggiungere risultati migliori in termini di performance.
Analysis and modelling of water withdrawals in the city of Milan
INVERNIZZI, LIVIA
2018/2019
Abstract
In the last decades, water resources became increasingly stressed by different phenomena, including climate change, pollution, industrialization, population growth and urbanization. Along with population increase, urbanization is significantly changing the spatial distribution of water demands. Urban water distribution systems must meet an increasing demand and, in response to that, urban planners need to improve urban water management plans and develop resilient water distribution systems. Understanding and forecasting water demand become a very important issue in different sectors of urban planning, and as demonstrated by the increasing number of the studies on urban water consumption. The main objective of these studies is to detect models able to able to provide an exhaustive description of the urban water use and to analyze which and how different external factors influence water use and the changes, in order to collect much information as possible. The first phase of this work focuses on detection of the changes in urban withdrawal between the 80s and 2000 in Milan. It results that the withdrawal of the city overall decreased, during the considered period. Also, the withdrawals of the lifting stations became more stable in the second part of the studied period. It results that there were not any events which influenced the city overall withdrawals, but around the 1990 and the 2010 numerous suburbs stations showed changes in the withdrawal trend, probably as response to some urban population movements. The second phase focused on attribution, meaning to investigate which variables mostly influenced the withdrawals of the stations. The withdrawals of four lifting stations were modelled following the Box-Jenkins procedure. The selected stations were Corsico, Crescenzago, Salemi, Suzzani: these stations were chosen according to monthly data availability. The withdrawals were modelled first considering different intervals, according to change in the trend, then considering the complete dataset. The best models turned out to be the AutoRegressive models AR(p) of order 2 or 3. Considering the complete dataset all these models showed values of R2 greater than 0.60 both in calibration and validation phases, while considering the different intervals, some intervals showed very low R2 coefficient (less than 0.20). In order to investigate the drivers of the changes in withdrawals and improve the models, some external variables were added to the models. In particular, the withdrawals of the nearest stations, the monthly temperature and the monthly precipitation were added. The level of cross-correlation of the external variables with past withdrawals were low: these levels showed that the variables could not give a significant improvement to the models. In fact, the obtained ARX models confirmed this assumption. Comparing the model performances, the withdrawals of the nearest stations gave a slightly better performance than the other variables, while between the climate variables the higher contribution was given by the number of exceptionally hot days. Instead there was not a clear response to the question if it is better to use the amount of precipitation or its occurrence in the month, because each station showed different response. According to the obtained results, some aspects of the work need further developments. For example, having access to data series at high resolution, both in temporal and spatial resolution, should improve the analysis of urban water consumption. Also, it may be test if using a more sophisticated kind of model (e.g. the Artificial Neural Networks) it is possible to obtain better results in terms of precision.| File | Dimensione | Formato | |
|---|---|---|---|
|
2019_10_Invernizzi.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
2.33 MB
Formato
Adobe PDF
|
2.33 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/10589/150007