This study presents a technique for assessing the Sentinel-2 and UAV-derived NDVI competence in monitoring and predicting the yield in three different processing tomato farms of Northern Italy. It investigates the pattern of NDVI change from the available Sentinel-2 cloud free image. It also compares the NDVI histogram derived from Sentinel-2 and UAV to checked their differences. A 2*2 m² area inside of a 10*10 m² Sentinel-2-based pixel area was used to take a yield sample data. Raster image of the field location was extracted using Sentinel-2-Copernicus Data Space Ecosystem, Copernicus Browser then was uploaded in Sentinel Application Platform (SNAP) software to derive the field NDVI. Additionally, the DJI MAVIC 3M UAV was also used to get it with a 2×2 m² pixel area resolution. All three fields were gridded with 2×2 m² grids using QGIS software. Then, in-depth relations between the NDVI of the sample location and the sample yield, i.e., correlation, coefficient of determination (Adj. R²), root mean square error (RMSE), relative root mean square error (RRMSE) and P value from July to September 2024, were extracted using MS-Excel. NDVI from specific dates that had the best relation to yield was chosen to create a linear regression for yield prediction at the sample level. This same relation was then used to predict the field yield. Relating sample yield with Predicted sample yield and total field yield with predicted field yields, relative root mean square error (RRMSE) was calculated to define prediction accuracies at sample and field level. Furthermore, using yield records, the influence of date of transplantation on life cycle and productivity was analyzed. It shows both Sentinel-2 and UAV were able to represent NDVI-based fields internal variability and yield related prediction. For the yield prediction, Sentinel-2 was found only good but not excellent enough, UAV was found excellent but their acquisitions numbers were found not sufficient enough. Based on these three field conditions, this study suggests 15 to 30 days before harvest as the best time for yield prediction. In the field of Castle with late rainfall and damaged tomatoes, low harvest with highest RRMSE of 16.07% was recorded. It supports the crop insurance payback because of environmental damages. Limited UAV based NDVI was found best predictor only in the field of CEO with RRMSE of 6.05% at the sample level and 7.85% in the field level while with Sentinel-2-derived NDVI, RRMSE was found 12% at sample level and 9.36% at field level. RRMSE in another field which didn’t have late rainfall damage, i.e. field Andrea, was found 4.89% with Sentinel-2-derived NDVI at the sample level and 10.13% at the field level. It provides the evidence of better consistency and prediction capability of UAV based prediction. Similarly, the field of Andrea and the CEO, which were adjacent to each other, with similar field condition had similar tomato variety but with different date of transplantation, their productivity was recorded as 110 t/h and 64 t/h and growing days of 110 and 130 days respectively. It provides the evidence of the impact of the date of transplantation on productivity and growing days. It suggests transplantation of tomato considering the temperature forecast report.
Questo studio presenta una tecnica per valutare la competenza NDVI derivata da Sentinel-2 e UAV nel monitoraggio e nella previsione della resa in tre diverse aziende agricole di pomodori da lavorazione del Nord Italia. Indaga il modello di variazione NDVI dall'immagine libera da nuvole Sentinel-2 disponibile. Confronta anche l'istogramma NDVI derivato da Sentinel-2 e UAV per verificarne le differenze. Un'area di 2*2 m² all'interno di un'area pixel basata su Sentinel-2 di 10*10 m² è stata utilizzata per acquisire dati di campione di resa. L'immagine raster della posizione del campo è stata estratta utilizzando Sentinel-2-Copernicus Data Space Ecosystem, Copernicus Browser è stata quindi caricata nel software Sentinel Application Platform (SNAP) per derivare l'NDVI del campo. Inoltre, è stato utilizzato anche il DJI MAVIC 3M UAV per ottenerlo con una risoluzione dell'area pixel di 2×2 m². Tutti e tre i campi sono stati suddivisi in griglie con griglie di 2×2 m² utilizzando il software QGIS. Quindi, relazioni approfondite tra l'NDVI della posizione del campione e la resa del campione, ovvero correlazione, coefficiente di determinazione (Adj. R²), errore quadratico medio (RMSE), errore quadratico medio relativo (RRMSE) e valore P da luglio a settembre 2024, sono state estratte utilizzando MS-Excel. L'NDVI da date specifiche che avevano la migliore relazione con la resa è stato scelto per creare una regressione lineare per la previsione della resa a livello di campione. Questa stessa relazione è stata quindi utilizzata per prevedere la resa del campo. Correlando la resa del campione con la resa del campione prevista e la resa totale del campo con le rese del campo previste, è stato calcolato l'errore quadratico medio relativo (RRMSE) per definire l'accuratezza della previsione a livello di campione e di campo. Inoltre, utilizzando i record di resa, è stata analizzata l'influenza della data del trapianto sul ciclo di vita e sulla produttività. Dimostra che sia Sentinel-2 che UAV sono stati in grado di rappresentare la variabilità interna dei campi basata sull'NDVI e la previsione correlata alla resa. Per la previsione della resa, Sentinel-2 è stato trovato solo buono ma non abbastanza eccellente, UAV è stato trovato eccellente ma i loro numeri di acquisizioni non sono stati trovati abbastanza sufficienti. Sulla base di queste tre condizioni di campo, questo studio suggerisce da 15 a 30 giorni prima del raccolto come il momento migliore per la previsione della resa. Nel campo di Castle con piogge tardive e pomodori danneggiati, è stato registrato un raccolto basso con il più alto RRMSE del 16,07%. Supporta il rimborso dell'assicurazione del raccolto a causa dei danni ambientali. L'NDVI basato su UAV limitato è stato trovato il miglior predittore solo nel campo di CEO con RRMSE del 6,05% a livello di campione e del 7,85% a livello di campo mentre con NDVI derivato da Sentinel-2, RRMSE è stato trovato al 12% a livello di campione e al 9,36% a livello di campo. RRMSE in un altro campo che non ha subito danni da pioggia tardiva, vale a dire il campo Andrea, è stato trovato al 4,89% con NDVI derivato da Sentinel-2 a livello di campione e al 10,13% a livello di campo. Fornisce la prova di una migliore coerenza e capacità di previsione della previsione basata su UAV. Allo stesso modo, il campo di Andrea e del CEO, che erano adiacenti l'uno all'altro, con condizioni di campo simili avevano varietà di pomodoro simili ma con data di trapianto diversa, la loro produttività è stata registrata come 110 t/h e 64 t/h e giorni di crescita rispettivamente di 110 e 130 giorni. Fornisce la prova dell'impatto della data di trapianto sulla produttività e sui giorni di crescita. Suggerisce il trapianto di pomodoro considerando il rapporto di previsione della temperatura.
Analisi delle previsioni di resa del pomodoro con Sentinel-2 e NDVI basato su UAV: uno studio di caso nelle aziende agricole del Nord Italia
KAFLE, AVINASH
2023/2024
Abstract
This study presents a technique for assessing the Sentinel-2 and UAV-derived NDVI competence in monitoring and predicting the yield in three different processing tomato farms of Northern Italy. It investigates the pattern of NDVI change from the available Sentinel-2 cloud free image. It also compares the NDVI histogram derived from Sentinel-2 and UAV to checked their differences. A 2*2 m² area inside of a 10*10 m² Sentinel-2-based pixel area was used to take a yield sample data. Raster image of the field location was extracted using Sentinel-2-Copernicus Data Space Ecosystem, Copernicus Browser then was uploaded in Sentinel Application Platform (SNAP) software to derive the field NDVI. Additionally, the DJI MAVIC 3M UAV was also used to get it with a 2×2 m² pixel area resolution. All three fields were gridded with 2×2 m² grids using QGIS software. Then, in-depth relations between the NDVI of the sample location and the sample yield, i.e., correlation, coefficient of determination (Adj. R²), root mean square error (RMSE), relative root mean square error (RRMSE) and P value from July to September 2024, were extracted using MS-Excel. NDVI from specific dates that had the best relation to yield was chosen to create a linear regression for yield prediction at the sample level. This same relation was then used to predict the field yield. Relating sample yield with Predicted sample yield and total field yield with predicted field yields, relative root mean square error (RRMSE) was calculated to define prediction accuracies at sample and field level. Furthermore, using yield records, the influence of date of transplantation on life cycle and productivity was analyzed. It shows both Sentinel-2 and UAV were able to represent NDVI-based fields internal variability and yield related prediction. For the yield prediction, Sentinel-2 was found only good but not excellent enough, UAV was found excellent but their acquisitions numbers were found not sufficient enough. Based on these three field conditions, this study suggests 15 to 30 days before harvest as the best time for yield prediction. In the field of Castle with late rainfall and damaged tomatoes, low harvest with highest RRMSE of 16.07% was recorded. It supports the crop insurance payback because of environmental damages. Limited UAV based NDVI was found best predictor only in the field of CEO with RRMSE of 6.05% at the sample level and 7.85% in the field level while with Sentinel-2-derived NDVI, RRMSE was found 12% at sample level and 9.36% at field level. RRMSE in another field which didn’t have late rainfall damage, i.e. field Andrea, was found 4.89% with Sentinel-2-derived NDVI at the sample level and 10.13% at the field level. It provides the evidence of better consistency and prediction capability of UAV based prediction. Similarly, the field of Andrea and the CEO, which were adjacent to each other, with similar field condition had similar tomato variety but with different date of transplantation, their productivity was recorded as 110 t/h and 64 t/h and growing days of 110 and 130 days respectively. It provides the evidence of the impact of the date of transplantation on productivity and growing days. It suggests transplantation of tomato considering the temperature forecast report.| File | Dimensione | Formato | |
|---|---|---|---|
|
thesis start-Rev-finall.pdf
non disponibili
Descrizione: This study presents a technique for assessing the Sentinel-2 and UAV-derived NDVI competence in monitoring and predicting the yield in three different processing tomato farms of Northern Italy.
Dimensione
5.71 MB
Formato
Adobe PDF
|
5.71 MB | Adobe PDF | Richiedi una copia |
È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/28564