Assessing vegetation health and status is becoming a known issue for helping farmers’ decisions and actions. The use of remote sensing techniques in agriculture has become widely popular during the past years. Remote sensing data can greatly contribute to constantly monitor the earth’s surface and to estimate important vegetation parameters. This work exploits EO data, acquired from both airborne and spaceborne sensors, and hybrid machine learning techniques to estimate biophysical variables, such as Leaf Chlorophyll Content (LCC), Canopy Chlorophyll Content (CCC), Leaf Nitrogen Content (LNC), Canopy Nitrogen Content (CNC) and Leaf Area Index (LAI). The newly-released PROSAIL-PRO model was used to generate a database with hundreds of simulated vegetation reflectance spectra; this database was then used to train machine learning regression algorithms for the estimation of crop biophysical and biochemical variables. The validation was carried out using ground data from two maize crops, located near Grosseto. During July 2018, two field campaigns were carried out to measure BVs in 87 plots within the two maize fields. The best performing algorithms were then applied to satellite datasets, including the multispectral sensor Sentinel-2 and simulations of hyperspectral sensor PRISMA, recently launched by the Italian Space Agency, for the production of BVs maps.
La valutazione della salute e dello stato della vegetazione sta diventando un problema noto per aiutare le decisioni e le pianificazioni degli agricoltori. Negli ultimi anni, le tecniche di telerilevamento terrestre stanno diventando sempre più popolari ed utilizzate. I dati telerilevati ricoprono un ruolo fondamentale: essi contribuiscono al monitoraggio costante della superficie terrestre e alla stima di importanti parametri vegetativi. Questo lavoro sfrutta i dati EO (Earth Observation), acquisiti da sensori sia aerei che satellitari, e tecniche ibride di machine learning (ML) per stimare importanti variabili biofisiche, come il Leaf Chlorophyll Content (LCC), Canopy Chlorophyll Content (CCC), Leaf Nitrogen Content (LNC), Canopy Nitrogen Content (CNC) and Leaf Area Index (LAI). Il recente RTM (Radiative Transfer Model) PROSAIL-PRO è stato utilizzato per generare un database con centinaia di spettri simulati di vegetazione; questo database è stato poi utilizzato per addestrare algoritmi di regressione di ML per la stima delle variabili biofisiche e biochimiche (VB). La validazione è stata effettuata utilizzando i dati di campo di due colture di mais, situate vicino a Grosseto. Nel mese di luglio 2018, sono state condotte due campagne per misurare i parametri biofisici e biochimici in 87 zone di campionamento all'interno dei due campi di mais. Gli algoritmi più performanti sono stati poi applicati ai dataset satellitari, provenienti dal sensore multispettrale Sentinel-2 e le simulazioni del sensore iperspettrale PRISMA, recentemente lanciato dall'Agenzia Spaziale Italiana, per la produzione di mappe di VB.
Hybrid approaches for the estimation of biophysical variables of agronomic interest from hyperspectral and multispectral data : applications for maize crops
RANGHETTI, MARINA
2019/2020
Abstract
Assessing vegetation health and status is becoming a known issue for helping farmers’ decisions and actions. The use of remote sensing techniques in agriculture has become widely popular during the past years. Remote sensing data can greatly contribute to constantly monitor the earth’s surface and to estimate important vegetation parameters. This work exploits EO data, acquired from both airborne and spaceborne sensors, and hybrid machine learning techniques to estimate biophysical variables, such as Leaf Chlorophyll Content (LCC), Canopy Chlorophyll Content (CCC), Leaf Nitrogen Content (LNC), Canopy Nitrogen Content (CNC) and Leaf Area Index (LAI). The newly-released PROSAIL-PRO model was used to generate a database with hundreds of simulated vegetation reflectance spectra; this database was then used to train machine learning regression algorithms for the estimation of crop biophysical and biochemical variables. The validation was carried out using ground data from two maize crops, located near Grosseto. During July 2018, two field campaigns were carried out to measure BVs in 87 plots within the two maize fields. The best performing algorithms were then applied to satellite datasets, including the multispectral sensor Sentinel-2 and simulations of hyperspectral sensor PRISMA, recently launched by the Italian Space Agency, for the production of BVs maps.File | Dimensione | Formato | |
---|---|---|---|
Marina Ranghetti - Hybrid approaches for the estimation of biophysical variables of agronomic interest from hyperspectral and multispectral data applications for maize crops.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
7.25 MB
Formato
Adobe PDF
|
7.25 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/173885