RETRIEVAL OF MAIZE BIOPHYSICAL VARIABLES FROM MULTISPECTRAL AND HYPERSPECTRAL EO DATA USING A HYBRID APPROACH

Assessing crops health and status is becoming relevant to support farmers' decisions and actions for a sustainable agriculture. The use of remote sensing techniques in agriculture has become widely popular during the past years. Earth Observing (EO) data can greatly contribute to constantly monitor crops phenology and to estimate important vegetation biophysical parameters. This work presents a hybrid approach, which exploits the PROSAIL-PRO model and Machine Learning (ML) algorithms, to estimate maize biophysical variables, such as Canopy Chlorophyll Content (CCC) and Leaf Area Index (LAI). The test site is represented by two maize fields located near Grosseto (Tuscany, IT), where two field campaigns were carried out in July 2018. During the same period, the airborne sensor Hyplant-DUAL acquired two images of the test site. These images were used to simulate PRISMA and Sentinel-2 data in order to investigate the difference of the retrieval performance between hyperspectral and multispectral EO data. Results show similar performance between Sentinel-2 and PRISMA. The ML algorithms, providing the best performance (GPR and NN) within the hybrid framework, were then applied to actual Sentinel-2 images. The retrieval results for LAI and CCC were compared to estimations assessed through the ESA S2Toolbox. The comparison showed that the proposed method provides better results than those achieved through S2Toolbox, for both LAI (R = 0.85 and MAE = 0.39; S2Toolbox: R = 0.35 and MAE = 0.87) and CCC (R = 0.73 and MAE = 0.20; S2Toolbox: R = 0.29 and MAE = 0.68).