The occurrence of the Olive Quick Decline Syndrome (OQDS) caused by Xylella fastidiosa (Xf) in Apulia region (Italy), with the strain Co.Di.RO (Complesso del Disseccamento Rapido dell’Olivo) affecting mainly the olive trees, poses a serious threat for olive production in all Mediterranean countries. Xf is a regulated pathogen in Europe (EPPO A1 list) because it affects more than 350 plant species worldwide. Infected olive trees may die as a consequence of the multiplication of the bacterium inside the vascular system which restricts the water flow from the roots to the canopy of the tree. Around 95% of olive cultivation is concentrated in the Mediterranean region and Italy ranks second worldwide. Accordingly, Xf represents the main threat of olive trees worldwide due to the severe symptoms induced (mainly leaf scorch, dieback and quick decline of the tree), the long list of sap-feeding insects which may efficiently spread the pathogen, as the Philaenous spumarius in Apulia, and the large number of secondary hosts. Xf restricts the cultivation of olive trees and the preservation of the historical heritage of olive trees in the Mediterranean region. Currently, no control measures are fully effective in the control of the bacterium and in the management of the olive quick decline; therefore the early detection of infected trees, their immediate eradication and vector control strategies are the only means of avoiding or containing the risk of contamination. These measures could be more effective if the infection is identified at early stages of disease development, in order to mitigate the spread of the pathogen and infections to neighbouring trees. However, visual inspections in the field are time-consuming and expensive. To this aim, remote sensing could be a useful tool to detect water stress induced by Xf infection in olive trees at early stages. Recently, an increase in research occurred in the application of Geomatic techniques, due to a greater availability of Remote and Proximal Sensing (RS, PS) instruments which has led to significant progress in the monitoring of complex biological phenomena and relative data management for running in, stand-alone, or web-based Geographic Information System (GIS) platforms. In this way it is possible to integrate heterogeneous spatial data in a single operative environment. Such data can be obtained by means of direct methods or indirect methods. The resulting data can be used for the implementation of provisional models to identify a plant adversity in order to rationalize the intervention strategy. The first research of this work, the suitability of photointerpretation techniques to recognize and classify the plants damaged by OQDS in GIS environment was evaluated, for this purpose very high geometrical resolution aerial images were used by processing visible (VIS) and near infrared (NIR) data on a study area in South of Apulia region, which represents the first outbreak area of Xf. The remotely acquired radiometric measurements were aimed at identifying appropriate photo-types, morphologically suitable in detecting the alteration of olive trees associated to different levels of OQDS-like symptoms. The use of spatially defined images strengthened by the presence of the near infrared band has greatly facilitated the identification of signs of OQDS starting with key photo types which are well correlated to the expression of the disease. The technique made it possible to identify 20% of the photo interpreted OQDS-trees and infected by Xf. This achievement is the prerequisite to thoroughly examine and improve the methodology through the use of stereoscopic restitution in the GIS environment. However, a second research was aimed at assessing the potential of hyperspectral reflectance data (HR) to identify the infection of Xf in olive at early stages of development. Sampling was carried out on infected plants belonging to the two main olive varieties varieties (cvs. “Cellina di Nardò” and “Leccino”) grown in a commercial grove located in the outbreak area of Xf in south Apulia. Each sample was made of leaves collected from 10 twigs/tree with different levels of infection. The study focused on the: (i) the discrimination between infected asymptomatic and non infected leaves; (ii) the selection of the best wavelengths for highlighting this discrimination and (iii) the identification of bio-physiological indicators (vegetation indices) correlated to the OQDS induced by Xf. The discrimination of infected leaves has been made using pre-elaborated data acquired with a field spectroradiometer, in the spectral wavelengths range between 400 and 1830 nm. A heuristic approach to variable selection, used in literature (Lambda-Lambda R2 model - LLR2, Principal Component Analisys model - PCA and Wilks' Lambda) and a combined general purpose detection method, proposed in this research, named interval PCA Internal Clustering Validation, iPCA-ICV have been compared. The unsupervised method proposed, divides the spectrum of reflectance data into a determined number of intervals, calculates the PCA within them (iPCA) and validates the goodness of the groupings obtained (classes) through Cluster Validity index measurement. The discriminative ability of selected wavelengths by the two methods was assessed by generalized discriminant analysis based on canonical correlation and measurement of error type such as leave-one-out cross-validation, through confusion matrices. From both methods it was possible to discriminate leaves infected by Xf and to select specific narrowbands. However, the best discriminative power was obtained from iPCA-ICV for both varieties (error rates of 23.7% and of 22.02% respectively for cv. Cellina di Nardò and cv. Leccino), compared to the reference method (error rates equivalent to 42.47% and 22.02% respectively for cv. Cellina di Nardò and cv. Leccino). The two methods have shown differences in number and in the position in the narrowbands selected (each of 10 nm) between the two varieties. In particular, both agree with the VIS regions (close to the blue and the red) and that of Short Wave Infrared (SWIR) as portions of the spectrum increase the discrimination of Leccino, the variety less affected by the infection (23.1%), while, for Cellina, the species more affected (85.7% of positive findings). The iPCA-ICV identifies the absorption bands of water around 1180 and 1400 nm (and many bands of SWIR). The heuristic method identifies two bands of 705 and 805 nm, as determinants in the identification of Xylella. The identification of critical regions of the spectrum, therefore, is the first logical step towards the development of indicators of robust stress based on hyperspectral images. The band selection techniques, also, are extremely useful not only to improve the power of predictive models, but also for the interpretation of the data or design of specific sensors for Pest Disease Detection (PDD).
La presenza in Puglia (Italia) del Complesso del Disseccamento Rapido dell’Olivo (CoDiRO) causato da Xylella fastidiosa (Xf), il cui ceppo Co.Di.RO colpisce prevalentemente gli alberi di olivo, rappresenta una seria minaccia per la produzione olivicola in tutti i Paesi mediterranei. Xf è un patogeno regolamentato in Europa come organismo di quarantena (lista EPPO A1) perché colpisce più di 350 specie vegetali in tutto il mondo. La maggior parte degli olivi infetti muore a seguito della moltiplicazione del batterio all’interno del sistema vascolare che limita il flusso dell’acqua dalle radici alla chioma dell’albero. Circa il 95% della coltivazione olivicola è concentrata nella regione mediterranea e l’Italia è il secondo Paese produttore a livello mondiale. Quindi, Xf rappresenta una seria minaccia per l’olivo nel mondo a causa della gravità dei sintomi indotti (soprattutto bruscatura delle foglie, disseccamento dei rami e deperimento rapido dell’albero), della lunga lista di vettori che possono diffondere efficientemente il patogeno, come il Philaenous spumarius in Puglia, e l’elevato numero di ospiti secondari del patogeno. Xf rappresenta un limite per la coltivazione dell’olivo e per la tutela del patrimonio storico olivicolo nella regione mediterranea. Ad oggi, non esistono misure efficaci di controllo e di lotta diretta al batterio e al CoDiRO; quindi, l’identificazione precoce degli alberi infetti, la loro immediata eradicazione e le strategie di controllo dei vettori sono gli unici mezzi per impedire o limitare il rischio di contaminazione. Tali misure potrebbero essere più efficaci se l’identificazione dell’infezione avvenisse nei primi stadi di sviluppo della malattia, in modo da poter contenere la diffusione del patogeno e la sua trasmissione agli alberi circostanti. Comunque, i rilievi visivi in campo richiedono tempo e sono costosi. A questo scopo, il telerilevamento potrebbe essere uno strumento utile all’identificazione di stress idrici causati dai primi stadi dell’infezione di Xf negli alberi di olivo. In tempi recenti si è assistito ad un aumento della ricerca nelle applicazioni delle tecniche Geomatiche, favorito dalla maggiore disponibilità di strumenti di rilevazione da remoto e da vicino, che ha condotto ad un significativo avanzamento della possibilità di monitorare fenomeni biologici complessi e di gestire, in ambiente Geographic Information System (GIS), i relativi dati sia in modalità stand-alone che in rete. In tal modo è possibile integrare, in un unico ambiente operativo, dati spaziali eterogenei derivanti dall'impiego di metodi diretti, come le azioni di monitoraggio, o dall’utilizzo di metodi indiretti, come l'elaborazione dei dati telerilevati. I dati così prodotti possono essere utilizzati per l'implementazione di modelli previsionali nella difesa delle avversità sul territorio e per potere così adattare la strategia di intervento e razionalizzare la difesa delle colture. La prima ricerca in questo lavoro ha come obiettivo la valutazione dell’idoneità delle tecniche di fotointerpretazione per riconoscere e classificare piante colpite dal CoDiRO in ambiente GIS. A tal fine sono state utilizzate immagini da aereo ad alta risoluzione geometrica nel visibile e nell'infrarosso vicino relative ad un’area di studio nel sud della Regione Puglia, che rappresenta la prima area focolaio di Xf. Le misure radiometriche rilevate da remoto sono state orientate all'individuazione di appropriati fototipi, morfologicamente in grado di rilevare l’alterazione associata a diversi livelli di sintomi ascrivibili al CoDiRO. L’uso di immagini spaziali definite, rafforzato dalla presenza della banda nel vicino infrarosso, ha facilitato notevolmente l’identificazione dei segnali di CoDiRO a partire dai fototipi chiave che sono ben correlati all’espressione della malattia. La tecnica ha reso possibile l’identificazione del 20% di alberi fotointerpretati con CoDiRO ed infetti da Xf. Questo risultato rappresenta un buon presupposto per poter esaminare in modo approfondito e migliorare la metodologia attraverso la restituzione stereoscopica in ambiente GIS. La seconda ricerca è stata invece finalizzata all’accertamento del potenziale dei dati di riflettanza iperspettrale (HR) per poter identificare l’infezione di Xf nei primi stadi di sviluppo su olivo. I campionamenti hanno riguardato piante infette delle due principali varietà di olivo (cvs “Cellina di Nardò” e “Leccino”) coltivate in un campo commerciale localizzato nell’area focolaio di Xf nel Sud della Puglia. Ogni campione era composto da foglie raccolte da 10 rametti/albero con diversi livelli di infezione. Lo studio ha avuto come obiettivo la: (i) discriminazione tra foglie infette asintomatiche e foglie non infette; (ii) la selezione delle migliori bande per evidenziare tale discriminazione e il (iii) confrontato tra due metodi di selezione delle variabili a sostegno dell'analisi delle riflettanze iperspettrali. La discriminazione delle foglie infette asintomatiche da quelle non infette, utilizzando dati pre-elaborati acquisiti con uno spettroradiometro da campo, è stata definita nell’intervallo di lunghezze d’onda 400 - 1830 nm dello spettro. Un approccio euristico di selezione delle variabili, utilizzato in letteratura (Lambda-Lambda R2 model - LLR2, Principal Component Analisys model - PCA e Wilks' Lambda) e un combined general purpose detection method, proposto in questa ricerca, denominato interval PCA Internal Clustering Validation (iPCA-ICV) sono stati messi a confronto. Il metodo non supervisionato proposto, divide lo spettro dei dati di riflettanza in un numero determinato di intervalli, calcola la PCA all'interno di essi (iPCA) e convalida la bontà dei raggruppamenti ottenuti (classi) attraverso misure di Cluster Validity index. La capacità discriminante delle lunghezze d'onda selezionate dai due metodi è stata valutata mediante analisi discriminante generalizzata basata sulla correlazione canonica e sulla misura dell'errore di tipo leave-one-out cross-validation, attraverso matrici di confusione. Da entrambi i metodi è stato possibile discriminare foglie infette da Xylella fastidiosa e selezionare bande strette specifiche. Tuttavia, il miglior potere discriminante è stato ottenuto da iPCA-ICV per entrambe le varietà (percentuale di errore del 23.7% e del 22.02% rispettivamente per cv. Cellina di Nardò e cv. Leccino), rispetto al metodo di riferimento (percentuale di errore del 42.47% e del 22.02% rispettivamente per cv. Cellina di Nardò e cv. Leccino). I due metodi hanno evidenziato differenze nel numero e nella posizione delle bande strette selezionate (ciascuna di 10 nm) tra le due varietà. In particolare, entrambi concordano con le regioni del VIS (vicini al blu e al rosso) e dello Short Wave Infrared (SWIR) come porzioni dello spettro a maggior peso sulla discriminazione della Leccino, varietà meno colpita dall'infezione (23.1%), mentre, per la Cellina, varietà più colpita (85.7% di positività riscontrata), i due metodi risultano discordanti. Il iPCA-ICV individua le bande di assorbimento dell'acqua intorno a 1180, 1400 nm e in molte bande dello SWIR, il metodo euristico individua due bande a 705 e 805 nm, come determinanti nell'individuazione di Xylella. L'identificazione di regioni critiche dello spettro, dunque, costituisce il primo passo logico verso lo sviluppo di indicatori di stress robusti basati su immagini iperspettrali. Le tecniche di selezione delle bande, inoltre, risultano estremamente utili non solo per migliorare il potere dei modelli predittivi, ma anche per l'interpretazione dei dati o il design di sensori specifici Pest Desease Detection (PDD).
Defining optimal Hyperspectral Narrowbands as proximal sensing in the early detection of Xylella fastidiosa in olive trees / Gualano, Stefania. - (2017). [10.60576/poliba/iris/gualano-stefania_phd2017]
Defining optimal Hyperspectral Narrowbands as proximal sensing in the early detection of Xylella fastidiosa in olive trees
GUALANO, Stefania
2017-01-01
Abstract
The occurrence of the Olive Quick Decline Syndrome (OQDS) caused by Xylella fastidiosa (Xf) in Apulia region (Italy), with the strain Co.Di.RO (Complesso del Disseccamento Rapido dell’Olivo) affecting mainly the olive trees, poses a serious threat for olive production in all Mediterranean countries. Xf is a regulated pathogen in Europe (EPPO A1 list) because it affects more than 350 plant species worldwide. Infected olive trees may die as a consequence of the multiplication of the bacterium inside the vascular system which restricts the water flow from the roots to the canopy of the tree. Around 95% of olive cultivation is concentrated in the Mediterranean region and Italy ranks second worldwide. Accordingly, Xf represents the main threat of olive trees worldwide due to the severe symptoms induced (mainly leaf scorch, dieback and quick decline of the tree), the long list of sap-feeding insects which may efficiently spread the pathogen, as the Philaenous spumarius in Apulia, and the large number of secondary hosts. Xf restricts the cultivation of olive trees and the preservation of the historical heritage of olive trees in the Mediterranean region. Currently, no control measures are fully effective in the control of the bacterium and in the management of the olive quick decline; therefore the early detection of infected trees, their immediate eradication and vector control strategies are the only means of avoiding or containing the risk of contamination. These measures could be more effective if the infection is identified at early stages of disease development, in order to mitigate the spread of the pathogen and infections to neighbouring trees. However, visual inspections in the field are time-consuming and expensive. To this aim, remote sensing could be a useful tool to detect water stress induced by Xf infection in olive trees at early stages. Recently, an increase in research occurred in the application of Geomatic techniques, due to a greater availability of Remote and Proximal Sensing (RS, PS) instruments which has led to significant progress in the monitoring of complex biological phenomena and relative data management for running in, stand-alone, or web-based Geographic Information System (GIS) platforms. In this way it is possible to integrate heterogeneous spatial data in a single operative environment. Such data can be obtained by means of direct methods or indirect methods. The resulting data can be used for the implementation of provisional models to identify a plant adversity in order to rationalize the intervention strategy. The first research of this work, the suitability of photointerpretation techniques to recognize and classify the plants damaged by OQDS in GIS environment was evaluated, for this purpose very high geometrical resolution aerial images were used by processing visible (VIS) and near infrared (NIR) data on a study area in South of Apulia region, which represents the first outbreak area of Xf. The remotely acquired radiometric measurements were aimed at identifying appropriate photo-types, morphologically suitable in detecting the alteration of olive trees associated to different levels of OQDS-like symptoms. The use of spatially defined images strengthened by the presence of the near infrared band has greatly facilitated the identification of signs of OQDS starting with key photo types which are well correlated to the expression of the disease. The technique made it possible to identify 20% of the photo interpreted OQDS-trees and infected by Xf. This achievement is the prerequisite to thoroughly examine and improve the methodology through the use of stereoscopic restitution in the GIS environment. However, a second research was aimed at assessing the potential of hyperspectral reflectance data (HR) to identify the infection of Xf in olive at early stages of development. Sampling was carried out on infected plants belonging to the two main olive varieties varieties (cvs. “Cellina di Nardò” and “Leccino”) grown in a commercial grove located in the outbreak area of Xf in south Apulia. Each sample was made of leaves collected from 10 twigs/tree with different levels of infection. The study focused on the: (i) the discrimination between infected asymptomatic and non infected leaves; (ii) the selection of the best wavelengths for highlighting this discrimination and (iii) the identification of bio-physiological indicators (vegetation indices) correlated to the OQDS induced by Xf. The discrimination of infected leaves has been made using pre-elaborated data acquired with a field spectroradiometer, in the spectral wavelengths range between 400 and 1830 nm. A heuristic approach to variable selection, used in literature (Lambda-Lambda R2 model - LLR2, Principal Component Analisys model - PCA and Wilks' Lambda) and a combined general purpose detection method, proposed in this research, named interval PCA Internal Clustering Validation, iPCA-ICV have been compared. The unsupervised method proposed, divides the spectrum of reflectance data into a determined number of intervals, calculates the PCA within them (iPCA) and validates the goodness of the groupings obtained (classes) through Cluster Validity index measurement. The discriminative ability of selected wavelengths by the two methods was assessed by generalized discriminant analysis based on canonical correlation and measurement of error type such as leave-one-out cross-validation, through confusion matrices. From both methods it was possible to discriminate leaves infected by Xf and to select specific narrowbands. However, the best discriminative power was obtained from iPCA-ICV for both varieties (error rates of 23.7% and of 22.02% respectively for cv. Cellina di Nardò and cv. Leccino), compared to the reference method (error rates equivalent to 42.47% and 22.02% respectively for cv. Cellina di Nardò and cv. Leccino). The two methods have shown differences in number and in the position in the narrowbands selected (each of 10 nm) between the two varieties. In particular, both agree with the VIS regions (close to the blue and the red) and that of Short Wave Infrared (SWIR) as portions of the spectrum increase the discrimination of Leccino, the variety less affected by the infection (23.1%), while, for Cellina, the species more affected (85.7% of positive findings). The iPCA-ICV identifies the absorption bands of water around 1180 and 1400 nm (and many bands of SWIR). The heuristic method identifies two bands of 705 and 805 nm, as determinants in the identification of Xylella. The identification of critical regions of the spectrum, therefore, is the first logical step towards the development of indicators of robust stress based on hyperspectral images. The band selection techniques, also, are extremely useful not only to improve the power of predictive models, but also for the interpretation of the data or design of specific sensors for Pest Disease Detection (PDD).File | Dimensione | Formato | |
---|---|---|---|
29 ciclo-GUALANO Stefania.pdf
accesso aperto
Descrizione: Tesi di dottorato completa di frontespizio
Tipologia:
Tesi di dottorato
Licenza:
Creative commons
Dimensione
30.77 MB
Formato
Adobe PDF
|
30.77 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.