Nitrogen (N) fertilization is crucial in today’s crop production. Over the last decades, several proximal and remote-sensing-based approaches have been proposed to improve nitrogen use efficiency (NUE). In addition, crop simulation models (CSMs) have been utilized to determine yield potential and optimal N fertilization rates, considering the weather's effect during a given growing season. In the past, CSMs have been run using datasets of weather variables observed over 20 to 100 years (historical weather data). Historical data works well when the observed weather is close to the historical average but under or overestimates the crop's actual crop production, and N needs when outlier conditions occur. To this, crop models have been run using a combination of the observed and seasonal forecasts. In this thesis work, presented as a collection of articles, three main aspects were evaluated across three years of experiments. The first experiment tested two different crop model-based approaches in winter wheat. The Decision Support System for Agrotechnology Transfer (DSSAT) CSM was run using (i) historical weather data (1992-2018) and (ii) using a combination of observed weather data (up to the fertilization dates) and seasonal weather forecasts integrated with proximal sensing information. Results showed that feeding the CSM with historical and seasonal forecasts had greater yields, protein content and N efficiencies. In particular, coupling the model with in-season plant N estimates captured the spatial variability at the field level and provided the highest N efficiencies. In the second experiment, the performance of the two aforementioned crop model-based approaches was also evaluated in corn. The DSSAT CSM was run using (i) historical weather data (1992-2018) and (ii) using a combination of observed weather data (up to the fertilization dates) and seasonal weather forecasts integrated with proximal sensing information. The CSM approaches did not consistently outperform uniform fertilization due to two aspects: first, the use of proximal sensing early in the season (at the V6 stage) was able to capture only a limited part of the spatial variability; second, the historical dataset and seasonal forecasts well not represented the observed rainfall timing and amounts, leading to underestimation of the actual N leaching. In the third experiment, an automatic model calibration (AMC) approach was tested for providing user-independent model calibration in winter wheat and corn. AMC performances were compared, using RMSE and d-index, to those obtained with the manually calibrated and a default cultivar. Biomass, leaf area index, N uptake and grain yields collected over two growing seasons for each crop were the parameters used for the calibration. Results showed that the AMC aligned better to the observed values than the other two cultivars, suggesting AMC could be used for providing accurate, user-independent and less time-consuming cultivar genetic coefficients in corn and winter wheat.
Le concimazioni azotate sono uno degli aspetti più importanti dell’agricoltura di oggi. Negli scorsi decenni, diverse metodologie, legate ad informazioni derivanti da sensori prossimali e remoti, sono state proposte per una migliore gestione dell’azoto, e per migliorarne l’efficienza d’uso. I modelli di simulazione colturale sono stati utilizzati per definire il potenziale produttivo di un’area, e, in funzione di questo, dosaggi ottimali di azoto da apportare in maniera sito-specifica. Ovviamente, le stime fatte da un modello di simulazione colturale dipendono in buona parte dal dataset meteo che viene utilizzato. L’utilizzo di dataset meteo climatologico, dove 20-100 anni di dati meteo storici vengono forniti al modello, risulta appropriato quando le variabili meteo osservate (temperatura minima e massima, radiazione solare e precipitazione) non si discostano molto dai valori medi storici. Nel caso contrario, il modello andrà a sovrastimare o sottostimare lo sviluppo della coltura. Per cercare di aumentare l’accuratezza delle informazioni meteo fornite al modello, spesso si ricorre a dataset meteo che sono l’integrazione di dati osservati e previsioni stagionali. Nel presente lavoro di tesi, che viene presentato come una collezione di articoli, tre principali aspetti sono stati studiati. Nel primo capitolo, il modello di simulazione DSSAT è stato utilizzato per simulare la crescita e lo sviluppo del frumento tenero, secondo due diverse metodologie: (i) il modello è stato utilizzato con un dataset meteo climatologico (dal 1992 al 2018), e (ii) il modello è stato utilizzato con un dataset che combinava dati meteo osservati, fino alla data di concimazione, e previsioni stagionali, dalla data di concimazione in poi. Rese più elevate, maggior contenuto proteico della granella e migliori efficienze d’uso dell’azoto hanno contraddistinto le metodologie basate sul modello di simulazione. Nel secondo capitolo, il modello è stato utilizzato con le stesse due metodologie testate su frumento, in questo caso su mais. In questo caso, il modello non ha portato a performance consistentemente migliori rispetto alla concimazione uniforme, a causa di: (i) sensori prossimali utilizzati in fase troppo precoce, portando alla definizione di una minima parte della variabilità spaziale; (ii) frequenza ed entità degli eventi piovosi osservati non presente né nel dataset climatologico né in quello previsionale. Nel terzo capitolo, una metodologia di auto calibrazione del modello di simulazione è stata testata su frumento tenero e mais. I risultati ottenuti dalla cultivar auto calibrata, da quella calibrata manualmente, e da una cultivar di default sono stato confrontati con i valori misurati di biomassa, azoto assimilato, leaf area index, e produzione finale. La cultivar auto calibrata ha mostrato le migliori performances, suggerendo che la seguente metodologià potrà essere utilizzata per fornire calibrazioni robuste e veloci sia su mais che su frumento.
INTEGRAZIONE TRA MODELLI DI SIMULAZIONE COLTURALE, PREVISIONI METEO STAGIONALI E DATI PROSSIMALI PER ORIENTARE LE CONCIMAZIONI A DOSAGGIO VARIABILE / Gobbo, Stefano. - (2023 Mar 24).
INTEGRAZIONE TRA MODELLI DI SIMULAZIONE COLTURALE, PREVISIONI METEO STAGIONALI E DATI PROSSIMALI PER ORIENTARE LE CONCIMAZIONI A DOSAGGIO VARIABILE
GOBBO, STEFANO
2023
Abstract
Nitrogen (N) fertilization is crucial in today’s crop production. Over the last decades, several proximal and remote-sensing-based approaches have been proposed to improve nitrogen use efficiency (NUE). In addition, crop simulation models (CSMs) have been utilized to determine yield potential and optimal N fertilization rates, considering the weather's effect during a given growing season. In the past, CSMs have been run using datasets of weather variables observed over 20 to 100 years (historical weather data). Historical data works well when the observed weather is close to the historical average but under or overestimates the crop's actual crop production, and N needs when outlier conditions occur. To this, crop models have been run using a combination of the observed and seasonal forecasts. In this thesis work, presented as a collection of articles, three main aspects were evaluated across three years of experiments. The first experiment tested two different crop model-based approaches in winter wheat. The Decision Support System for Agrotechnology Transfer (DSSAT) CSM was run using (i) historical weather data (1992-2018) and (ii) using a combination of observed weather data (up to the fertilization dates) and seasonal weather forecasts integrated with proximal sensing information. Results showed that feeding the CSM with historical and seasonal forecasts had greater yields, protein content and N efficiencies. In particular, coupling the model with in-season plant N estimates captured the spatial variability at the field level and provided the highest N efficiencies. In the second experiment, the performance of the two aforementioned crop model-based approaches was also evaluated in corn. The DSSAT CSM was run using (i) historical weather data (1992-2018) and (ii) using a combination of observed weather data (up to the fertilization dates) and seasonal weather forecasts integrated with proximal sensing information. The CSM approaches did not consistently outperform uniform fertilization due to two aspects: first, the use of proximal sensing early in the season (at the V6 stage) was able to capture only a limited part of the spatial variability; second, the historical dataset and seasonal forecasts well not represented the observed rainfall timing and amounts, leading to underestimation of the actual N leaching. In the third experiment, an automatic model calibration (AMC) approach was tested for providing user-independent model calibration in winter wheat and corn. AMC performances were compared, using RMSE and d-index, to those obtained with the manually calibrated and a default cultivar. Biomass, leaf area index, N uptake and grain yields collected over two growing seasons for each crop were the parameters used for the calibration. Results showed that the AMC aligned better to the observed values than the other two cultivars, suggesting AMC could be used for providing accurate, user-independent and less time-consuming cultivar genetic coefficients in corn and winter wheat.File | Dimensione | Formato | |
---|---|---|---|
tesi_Stefano_Gobbo.pdf
embargo fino al 23/03/2026
Descrizione: tesi_Stefano_Gobbo
Tipologia:
Tesi di dottorato
Dimensione
4.23 MB
Formato
Adobe PDF
|
4.23 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.