This study focuses on the comparison of hybrid methods of estimation of biophysical variables such as leaf area index (LAI), leaf chlorophyll content (LCC), fraction of absorbed photosynthetically active radiation (FAPAR), fraction of vegetation cover (FVC), and canopy chlorophyll content (CCC) from Sentinel-2 satellite data. Different machine learning algorithms were trained with simulated spectra generated by the physically-based radiative transfer model PROSAIL and subsequently applied to Sentinel-2 reflectance spectra. The algorithms were assessed against a standard operational approach, i.e., the European Space Agency (ESA) Sentinel Application Platform (SNAP) toolbox, based on neural networks. Since kernel-based algorithms have a heavy computational cost when trained with large datasets, an active learning (AL) strategy was explored to try to alleviate this issue. Validation was carried out using ground data from two study sites: one in Shunyi (China) and the other in Maccarese (Italy). In general, the performance of the algorithms was consistent for the two study sites, though a different level of accuracy was found between the two sites, possibly due to slightly different ground sampling protocols and the range and variability of the values of the biophysical variables in the two ground datasets. For LAI estimation, the best ground validation results were obtained for both sites using least squares linear regression (LSLR) and partial least squares regression, with the best performances values of R² of 0.78, rott mean squared error (RMSE) of 0.68 m² m⁻² and a relative RMSE (RRMSE) of 19.48% obtained in the Maccarese site with LSLR. The best results for LCC were obtained using Random Forest Tree Bagger (RFTB) and Bagging Trees (BagT) with the best performances obtained in Maccarese using RFTB (R² = 0.26, RMSE = 8.88 μg cm⁻², RRMSE = 17.43%). Gaussian Process Regression (GPR) was the best algorithm for all variables only in the cross-validation phase, but not in the ground validation, where it ranked as the best only for FVC in Maccarese (R² = 0.90, RMSE = 0.08, RRMSE = 9.86%). It was found that the AL strategy was more efficient than the random selection of samples for training the GPR algorithm.

Crop growth models play an important role in agriculture management, allowing, for example, the spatialized estimation of crop yield information. However, crop model parameter calibration is a mandatory step for their application. The present work focused on the regional calibration of the Aquacrop-OS model for durum wheat by assimilating high spatial and temporal resolution canopy cover data retrieved from VENµS satellite images. The assimilation procedure was implemented using the Bayesian approach with the recent implementation of the Markov chain Monte Carlo (MCMC)-based Differential Evolution Adaptive Metropolis (DREAM) algorithm DREAM_(KZS). The fraction of vegetation cover (fvc) was retrieved from the VENµS satellite images for two years, during the durum wheat growing seasons of 2018 and 2019 in Central Italy. The retrieval was based on a hybrid method using PROSAIL Radiative Transfer Model (RTM) simulations for training a Gaussian Process Regression (GPR) algorithm, combined with Active Learning to reduce the computational cost. The Aquacrop-OS model was calibrated with the fvc data of 2017–2018 for the Maccarese farm in Central Italy and validated with the 2018–2019 data. The retrieval accuracy of the fvc from the VENµS images were the Coefficient of Determination (R²) = 0.76, Root Mean Square Error (RMSE) = 0.09, and Relative Root Mean Square Error (RRMSE) = 11.6%, when compared with the ground-measured fvc. The MCMC results are presented in terms of Gelman–Rubin R statistics and MR statistics, Markov chains, and marginal posterior distribution functions, which are summarized with the mean values for the most sensitive crop parameters of the Aquacrop-OS model subjected to calibration. When validating for the fvc, the R² of the model for year (2018–2019) ranged from 0.69 to 0.86. The RMSE, Relative Error (RE), Relative Variability (α), and Relative Bias (β) ranged from 0.15 to 0.44, 0.19 to 2.79, 0.84 to 1.45, and 0.91 to 1.95, respectively. The present work shows the importance of the calibration of the Aquacrop-OS (AOS) crop water productivity model for durum wheat by assimilating remote sensing information from VENµS satellite data.

The present work reports the global sensitivity analysis of the Aquacrop Open Source (AOS) model, which is the open-source version of the original Aquacrop model developed by the Food and Agriculture Organization (FAO). Analysis for identifying the most influential parameters was based on different strategies of global SA, density-based and variance-based, for the wheat crop in two different geographical locations and climates. The main objectives were to distinguish the model’s influential and non-influential parameters and to examine the yield output sensitivity. We compared two different methods of global sensitivity analysis: the most commonly used variance-based method, EFAST, and the moment independent density-based PAWN method developed in recent years. We have also identified non-influential parameters using Morris screening method, so to provide an idea of the use of non-influential parameters with a dummy parameter approach. For both the study areas (located in Italy and in China) and climates, a similar set of influential parameters was found, although with varying sensitivity. When compared with different probability distribution functions, the probability distribution function of yield was found to be best approximated by a Generalized Extreme Values distribution with Kolmogorov–Smirnov statistic of 0.030 and lowest Anderson–Darling statistic of 0.164, as compared to normal distribution function with Kolmogorov–Smirnov statistic of 0.122 and Anderson–Darling statistic of 4.099. This indicates that yield output is not normally distributed but has a rather skewed distribution function. In this case, a variance-based approach was not the best choice, and the density-based method performed better. The dummy parameter approach avoids to use a threshold as it is a subjective question; it advances the approach to setting up a threshold and gives an optimal way to set up a threshold and use it to distinguish between influential and non-influential parameters. The highly sensitive parameters to crop yield were specifically canopy and phenological development parameters, parameters that govern biomass/yield production and temperature stress parameters rather than root development and water stress parameters.

The experimentation carried out within this project aimed at comparing various methods used to estimate leaf area index (LAI), which is a key variable to describe the vigour and productive potential of a crop. The direct or “destructive” method requires the removal of the biomass of the sample and is a very laborious and burdensome technique. The indirect estimates consist in the acquisition of a series of readings taken directly in the field with specific instruments, mainly based on the measurement of the quantity of light intercepted by the foliage and on the use of complex mathematical models. Indirect methods allow for a quick estimation of the LAI and there is a strong interest in developing economic and user-friendly tools. The tools compared in the present study included two state-of-the-art, rather expensive, commercial instruments, i.e. the Sunscan ceptometer (Delta-T, UK) and the LAI-2000 (Li-Cor, USA). These were compared to inexpensive alternatives offered by hemispheric photography from a smartphone equipped with a fish-eye lens and from a smartphone app called “Pocket LAI”. All the methods tested showed the ability to provide an effective estimate of the LAI even if sometimes with an underestimation of the value measured by the direct method.

Novel approaches and algorithms to estimate crop physiological processes from Earth Observation (EO) data are essential to develop more sustainable management practices in agricultural systems. Within this context, this paper presents the results of different research activities carried out within the ESA-MOST Dragon 4 programme. The paper encompasses two research avenues: (a) the retrieval of biophysical variables of crops and yield prediction; and (b) food security related to different crop management strategies. Concerning the retrieval of variables, results show that LAI, derived by radiative transfer model (RTM) inversion, when assimilated into a crop growth model (i.e., SAFY) provides a way to assess yields with a higher accuracy with respect to open loop model runs: 1.14 t·ha⁻¹ vs 4.42 t·ha⁻¹ RMSE for assimilation and open loop, respectively. Concerning food security, results show that different pathogens could be detected by remote sensing satellite data. A k coefficient higher than 0.84 was achieved for yellow rust, thus assuring a monitoring accuracy, and for the diseased samples k was higher than 0.87. Concerning permanent crops, neural network (NN) algorithms allow classification of the Pseudomonas syringae pathogen on kiwi orchards with an overall accuracy higher than 91%.

Globally, croplands represent a significant contributor to climate change, through both greenhouse gas emissions and land use changes associated with cropland expansion. They also represent locations with significant potential to contribute to mitigating climate change through alternative land use management practices that lead to increased soil carbon sequestration. In spite of their global importance, there is a relative paucity of tools available to support field- or farm-level crop land decision making that could inform more effective climate mitigation practices. In recognition of this shortcoming, the Simple Algorithm for Yield Estimate (SAFY) model was developed to estimate crop growth, biomass, and yield at a range of scales from field to region. While the original SAFY model was developed and evaluated for winter wheat in Morocco, a key advantage to utilizing SAFY is that it presents a modular architecture which can be readily adapted. This has led to numerous modifications and alterations of specific modules which enable the model to be refined for new crops and locations. Here, we adapted the SAFY model for use with spring barley, winter wheat and winter oilseed rape at selected sites in Ireland. These crops were chosen as they represent the dominant crop types grown in Ireland. We modified the soil–water balance and carbon modules in SAFY to simulate components of water and carbon budgets in addition to crop growth and production. Results from the modified model were evaluated against available in situ data collected from previous studies. Spring barley biomass was estimated with high accuracy (R² = 0.97, RMSE = 95.8 g·m⁻², RRMSE = 11.7%) in comparison to GAI (R² = 0.73, RMSE = 0.44 m²·m⁻², RRMSE = 10.6%), across the three years for which the in situ data was available (2011–2013). The winter wheat module was evaluated against measured biomass and yield data obtained for the period 2013–2015 and from three sites located across Ireland. While the model was found to be capable of simulating winter wheat biomass (R² = 0.71, RMSE = 1.81 t·ha⁻¹, RRMSE = 8.0%), the model was found to be less capable of reproducing the associated yields (R² = 0.09, RMSE = 2.3 t·ha⁻¹, RRMSE = 18.6%). In spite of the low R² obtained for yield, the simulated crop growth stage 61 (GS61) closely matched those observed in field data. Finally, winter oilseed rape (WOSR) was evaluated against a single growing season for which in situ data was available. WOSR biomass was also simulated with high accuracy (R² = 0.99 and RMSE = 0.52 t·ha⁻¹) in comparison to GAI (R² = 0.3 and RMSE = 0.98 m²·m⁻²). In terms of the carbon fluxes, the model was found to be capable of estimating heterotrophic respiration (R² = 0.52 and RMSE = 0.28 g·C·m⁻²·day⁻¹), but less so the ecosystem respiration (R² = 0.18 and RMSE = 1.01 g·C·m⁻²·day⁻¹). Overall, the results indicate that the modified model can simulate GAI and biomass, for the chosen crops for which data were available, and yield, for winter wheat. However, the simulations of the carbon budgets and water budgets need to be further evaluated—a key limitation here was the lack of available in situ data. Another challenge is how to address the issue of parameter specification; in spite of the fact that the model has only six variable crop-related parameters, these need to be calibrated prior to application (e.g., date of emergence, effective light use efficiency etc.). While existing published values can be readily employed in the model, the availability of regionally derived values would likely lead to model improvements. This limitation could be overcome through the integration of available remote sensing data using a data assimilation procedure within the model to update the initial parameter values and adjust model estimates during the simulation.