This study develops three different artificial intelligence (AI) models in order to investigate the effects of climate change on groundwater resources using historical records of precipitation, temperature and groundwater levels together with regional climate projections. In particular, the Non-linear Autoregressive Neural Network (NARX), the Long-Short Term Memory Neural Network (LSTM) and the Convolutional Neural Network (CNN) were compared. Considering an aquifer located in northern Italy as a case study, the neural networks were trained to replicate observed groundwater levels by taking as input precipitation and temperature records, and in the case of the NARX also antecedent groundwater levels, on a monthly scale. The trained networks were used to infer groundwater levels until the end of the century based on precipitation and temperature projections provided by an ensemble of 13 Regional Climate Models (RCMs) from the EURO-CORDEX initiative. Two emission pathways were considered: the RCP4.5 and RCP8.5. All the AI models show good performance metrics during the training phase, but NARXs perform poorly compared to the other models during validation and testing. For the future, the NARX and LSTM models predict a decline in groundwater levels, especially for the RCP8.5 scenario, while slight changes are expected using the CNN. As NARXs are not deep learning techniques and CNNs may not be able to extrapolate values outside the training range, LSTMs appear to be better suited for climate change impact evaluations.

Artificial intelligence models to evaluate the impact of climate change on groundwater resources / Secci, Daniele; Giovanna Tanda, Maria; D'Oria, Marco; Todaro, Valeria. - In: JOURNAL OF HYDROLOGY. - ISSN 0022-1694. - 627:(2023). [10.1016/j.jhydrol.2023.130359]

Artificial intelligence models to evaluate the impact of climate change on groundwater resources

Secci, Daniele
Methodology
;
Giovanna Tanda, Maria
Methodology
;
D'Oria, Marco
Methodology
;
Todaro, Valeria
Methodology
2023-01-01

Abstract

This study develops three different artificial intelligence (AI) models in order to investigate the effects of climate change on groundwater resources using historical records of precipitation, temperature and groundwater levels together with regional climate projections. In particular, the Non-linear Autoregressive Neural Network (NARX), the Long-Short Term Memory Neural Network (LSTM) and the Convolutional Neural Network (CNN) were compared. Considering an aquifer located in northern Italy as a case study, the neural networks were trained to replicate observed groundwater levels by taking as input precipitation and temperature records, and in the case of the NARX also antecedent groundwater levels, on a monthly scale. The trained networks were used to infer groundwater levels until the end of the century based on precipitation and temperature projections provided by an ensemble of 13 Regional Climate Models (RCMs) from the EURO-CORDEX initiative. Two emission pathways were considered: the RCP4.5 and RCP8.5. All the AI models show good performance metrics during the training phase, but NARXs perform poorly compared to the other models during validation and testing. For the future, the NARX and LSTM models predict a decline in groundwater levels, especially for the RCP8.5 scenario, while slight changes are expected using the CNN. As NARXs are not deep learning techniques and CNNs may not be able to extrapolate values outside the training range, LSTMs appear to be better suited for climate change impact evaluations.
2023
Artificial intelligence models to evaluate the impact of climate change on groundwater resources / Secci, Daniele; Giovanna Tanda, Maria; D'Oria, Marco; Todaro, Valeria. - In: JOURNAL OF HYDROLOGY. - ISSN 0022-1694. - 627:(2023). [10.1016/j.jhydrol.2023.130359]
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11381/2974316
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 5
  • ???jsp.display-item.citation.isi??? 2
social impact