Sustainable Agriculture and Water Resources
Starting from the outmost results of the previous Dragon4 (32275) initiative this new proposal intends to explore the application of Sino-EU optical data into agronomic models to predict crop performance and to monitor and forecast crop pests and diseases.The team, as in the previous project, is composed by the Italian team with the National Research Council and two Universities, while the Chinese one with the Chinese Academy of Science and the Beijing Research center for Information Technology in Agriculture. The project will explore and verify pre-operative algorithms and processing chains using ESA/Chinese multi-frequency EO data to (a) develop innovative and advanced methods for crop plant key parameters retrieval at different growth stages (plant pigment, equivalent water thickness, dry matter, nitrogen and biomass); (b) estimate crop yields and grain quality using agronomic models by integrating multi-source information and by different assimilation techniques; (c) identify early crop stress both at the leaf and at the canopy level also by inferring the agricultural soil properties; (d) develop new dynamic methods and models to monitor and forecast crop pests and disease. This will be achieved by using ESA Products (Sentinel-2), TPM (Landsat-8, SPOT 7, WorldView) and Chinese product (GF-6, GF-1), taking also advantage of the spectral resolution of the ASI PRISMA hyperspectral imagery and the next generation ESA FLEX and Sentinel-5 missions. Moreover, the project foresees to explore the possibility offered by the DIAS systems.The Sino-EU synergy, beside the use of the EO mission data, will be exploited on jointly selected agricultural test sites in China and in Italy. Firstly, the Team will jointly explore the capabilities of the operational mapping of vegetation variables through RTMs, which is a challenging task due to the ill-posedness of the inversion and to the influence of several hampering factors (e.g., the canopy structure, the influence of the atmosphere, the illumination conditions and the sun-sensor geometries). For these reasons, the use of regularization strategies (cost functions, multiple solution) to reduce the uncertainties in the quantitative estimation of vegetation variables. To this aim, approaches exploiting physically-based radiative transfer models (RTM), will be compared with other methods. Secondly, data assimilation of multivariate and multi-scale remotely sensed variables into agricultural models (i.e. crop growing, pest e disease) will be explored. The project proposes to advance those approaches that tackle multiple scales and multiple variables, i.e. employing concurrently two or more variables for the assimilation. Agricultural model uncertainties will be assessed using global sensitivity analysis methods. Different assimilation algorithms (deterministic and stochastic) based on the EnKF and PSO methods. These methods will update the state variables and/or parameters of the crop models, to estimate variables of agronomic interests, such as crop yield and grain protein quality.Finally, the integration of multi-source data to retrieve crop physicochemical parameters, monitor crop pests and diseases habitat and then forecast damaged areas and levels at both farm and national scales will be explored. Machine learning methods, such as decision tree, SVM and deep learning algorithms will be applied to learn from field samples and devise complex models to detect the relationship between pest/disease and features. New dynamic methods and models will be compared with traditional methods using ground samples. Cross cutting validation activities will provide the data and for the retrieval algorithms validation and for the data assimilation approaches. Field experiments will be implemented. Ground hyperspectral data, agronomic management data, agrometeorological data, and soil data will be collected. Crop yield and quality variables (e.g. protein content) will be measured at harvest.
