Mostafa Adibian
Evaluating LSTM Networks for Local and Regional Rainfall-Runoff Modelling in North-Western Italian Alpine Catchments.
Rel. Alberto Viglione. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Per L'Ambiente E Il Territorio, 2026
|
|
PDF (Tesi_di_laurea)
- Tesi
Accesso limitato a: Solo utenti staff fino al 16 Settembre 2027 (data di embargo). Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (20MB) |
Abstract
This thesis evaluates Long Short-Term Memory (LSTM) networks for daily rainfall-runoff modelling in 37 gauged mountainous catchments in the Piedmont and Valle d'Aosta regions of north-western Italy. Local catchment-specific LSTM models and a regional pooled LSTM model are compared against available benchmark rainfall-runoff simulations, using daily precipitation, mean temperature, and potential evapotranspiration as the only dynamic forcing inputs. Performance is assessed using NSE, KGE, RMSE, MAE, and Flow Duration Curve (FDC) diagnostics for high-, middle-, and low-flow regimes. A proof-of-concept experiment on a CAMELS catchment verified the LSTM workflow and assessed input sensitivity through progressive variable removal. Validation NSE remained above 0.85 across all tested configurations, supporting the reduced-input design for the Alpine phase, though these results cannot be transferred directly to Alpine conditions.
The main Alpine analysis covered all 37 catchments, of which 34 produced valid test-period results; three were excluded because target-value filtering during sequence construction yielded no usable test samples
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
URI
![]() |
Modifica (riservato agli operatori) |
