Proper determination of irrigation water consumption is essential in supporting sustainable water management in highly cultivated agricultural areas. This dissertation is a study that examines the estimation of irrigation volumes using a data-driven approach in maize fields in the city of Calcinato in the Chiese irrigation district and of Livraga in the Muzza irrigation Consortium, both located within the Pianura Padana (Po Valley) of Northern Italy using multi-year data. According to the availability of data, ground-based irrigation data were combined with environmental predictors of available data including soil moisture, rainfall, net radiation, evapotranspiration, and other aspects of meteorological data. Comparison and application of regression-based modelling techniques were applied such as linear regression, decision -tree regression, and neural -network regression. The models were trained on data of the year 2016 growing season and then tested with independent datasets of the years to test their strength and their ability to be transferred through time. The effect of various combinations of input variables on estimation of irrigation was tested using systematic testing. The standard statistical indicators were used to assess the model performance, and it allowed a systematic comparison between the modelling strategies and the predictor configurations. The findings suggest that the accuracy of irrigation estimation depends on the choice of the input variables as well as the modelling method with some variations noticed between years and datasets. Altogether, this paper emphasizes the prospects of machine-learning-driven regression techniques to estimate irrigation in the Chiese irrigation district and the Muzza irrigation Consortium and presents a methodology that can be used in irrigation monitoring and water management in other similar agroecosystems.
La corretta determinazione del consumo di acqua per l’irrigazione è essenziale per sostenere una gestione sostenibile delle risorse idriche nelle aree agricole ad elevata intensità colturale. La presente tesi analizza la stima dei volumi irrigui nei campi di mais situati nel comune di Calcinato, nel distretto irriguo del Chiese, e nell’area di Livraga appartenente al Consorzio di bonifica Muzza, entrambi localizzati nella Pianura Padana (Valle del Po), nel Nord Italia, utilizzando dati multi-annuali. In base alla disponibilità dei dati, le informazioni irrigue rilevate a terra sono state integrate con variabili ambientali e agro-meteorologiche, tra cui umidità del suolo, precipitazioni, radiazione netta, evapotraspirazione e altri parametri meteorologici. Sono state applicate e confrontate diverse tecniche di modellazione basate su regressione, quali regressione lineare, regressione ad albero decisionale e regressione tramite reti neurali. I modelli sono stati addestrati utilizzando i dati della stagione colturale 2016 e successivamente testati con dataset indipendenti degli anni seguenti, al fine di valutarne la robustezza e la trasferibilità temporale. L’effetto delle diverse combinazioni di variabili di input sulla stima dell’irrigazione è stato analizzato attraverso un processo di test sistematico. Le prestazioni dei modelli sono state valutate mediante indicatori statistici standard, consentendo un confronto coerente tra le diverse strategie di modellazione e le configurazioni dei predittori. I risultati suggeriscono che l’accuratezza della stima dell’irrigazione dipende sia dalla scelta delle variabili di input sia dal metodo di modellazione, con variazioni osservate tra anni e dataset. Nel complesso, questo studio evidenzia il potenziale delle tecniche di regressione basate su machine learning per la stima dei volumi irrigui nel distretto irriguo del Chiese e nel Consorzio Muzza, proponendo un quadro metodologico che può supportare il monitoraggio dell’irrigazione e la gestione delle risorse idriche in agroecosistemi simili.
Machine learning-based irrigation volume estimation in maize fields in Northern Italy
KOMATHURUTHIL PRADEEP, ADITHYA
2024/2025
Abstract
Proper determination of irrigation water consumption is essential in supporting sustainable water management in highly cultivated agricultural areas. This dissertation is a study that examines the estimation of irrigation volumes using a data-driven approach in maize fields in the city of Calcinato in the Chiese irrigation district and of Livraga in the Muzza irrigation Consortium, both located within the Pianura Padana (Po Valley) of Northern Italy using multi-year data. According to the availability of data, ground-based irrigation data were combined with environmental predictors of available data including soil moisture, rainfall, net radiation, evapotranspiration, and other aspects of meteorological data. Comparison and application of regression-based modelling techniques were applied such as linear regression, decision -tree regression, and neural -network regression. The models were trained on data of the year 2016 growing season and then tested with independent datasets of the years to test their strength and their ability to be transferred through time. The effect of various combinations of input variables on estimation of irrigation was tested using systematic testing. The standard statistical indicators were used to assess the model performance, and it allowed a systematic comparison between the modelling strategies and the predictor configurations. The findings suggest that the accuracy of irrigation estimation depends on the choice of the input variables as well as the modelling method with some variations noticed between years and datasets. Altogether, this paper emphasizes the prospects of machine-learning-driven regression techniques to estimate irrigation in the Chiese irrigation district and the Muzza irrigation Consortium and presents a methodology that can be used in irrigation monitoring and water management in other similar agroecosystems.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_KomathuruthilPradeep.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Master's thesis text
Dimensione
2.5 MB
Formato
Adobe PDF
|
2.5 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/10589/252062