Proximal and Remote sensing integration for improved accuracy and representativeness in crop yield gap estimation

Food loss and waste have become a critical global challenge, intensifying environmental pressure on land and water resources and generating avoidable greenhouse gas emissions across the entire food supply chain. According to the Food and Agriculture Organization, food loss and waste affect society by compromising all four dimensions of food security (utilization, access, availability, and stability). One major strategy to prevent unsustainable agricultural expansion and environmental degradation, while simultaneously increasing food production to feed a growing global population, is to enhance agricultural output from existing croplands by closing crop yield gaps. Consequently, the quantification of yield gaps and accurate yield prediction play crucial roles in ensuring food security. Addressing these challenges requires advanced digital tools capable of supporting informed agricultural decision-making. Two key tools, crop growth models (CGMs) and remote sensing (RS), have been widely adopted in agricultural research. Traditionally applied independently, these approaches are increasingly being integrated, leveraging the complementary strengths of RS data and crop model simulation capabilities. This integration is particularly important for sustainable agricultural management and food security in data-scarce and resource-limited farming systems. Crop growth models provide mechanistic representations of plant growth and development, enabling yield prediction under varying environmental and management conditions. On the other hand, remote sensing plays a crucial role in vegetation monitoring by delivering spatially and temporally explicit information on vegetation condition, health, and dynamics. However, the effective deployment of CGMs requires localized calibration, typically based on in situ measurements, literature-derived parameters, or expert knowledge, often obtained from plot-scale experiments. Although plot-scale data are highly accurate, they often lack representativeness because experimental designs tend to minimize natural variability, such as soil and topographic heterogeneity, using blocking, contiguous layouts, and other controlled setups. As a result, yields obtained from plot-scale experiments frequently exceed those observed at the field scale due to inherent locational bias. This scale mismatch has long raised questions about the validity of extrapolating results from plot-scale studies to entire fields or regions, and about the extent to which such findings can reliably inform field-scale conclusions. In this context, the integration of remote sensing data, although less precise than plot-level measurements, provides the advantage of capturing the spatial variability of vegetation parameters and yield, thereby increasing representativeness and strengthening model parameterization and yield forecasting at the field or larger scale. In fact, RS enables efficient, large-scale agricultural monitoring with consistent revisit times, making it a potential alternative to manual model calibration. Numerous studies have integrated RS-derived crop variables into CGMs through data assimilation techniques to improve calibration and the accuracy of estimates. Data assimilation is valuable for reducing uncertainty in model inputs and compensating for missing biophysical processes, minimizing discrepancies between simulations and observations. In this context, lower observational error is more important than higher spatial resolution, and the choice of assimilation method significantly affects crop growth simulations. Over the past decade, advances in RS technology, including expanded spectral capabilities and radar/optical sensors, have enabled timely, consistent estimation of crop biophysical variables at field and regional scales. Despite these developments, data assimilation applications remain constrained by data availability and quality. For example, although relatively high-resolution remote sensing (RS) data can yield accurate estimates of crop variables, their application is often constrained by scale effects, revisit frequency, and the availability of cloud-free imagery. Consequently, many studies rely on coarse resolution but high-temporal-frequency datasets or medium-resolution sensors with lower revisit rates. Nonetheless, satellite data offering high spectral, temporal, and spatial resolution, such as those from the PlanetScope, Landsat-8, Sentinel-2, and Sentinel-3 missions, have recently become freely accessible for both research and operational applications. In addition, multispectral sensors on unmanned aerial vehicles (UAVs) and affordable in-field sensors now offer high spatial and temporal resolution data. However, the selection of a high-resolution RS dataset is largely determined by the user’s required scale, data accessibility, and accuracy. UAVs offer flexible, low-cost, field-scale imagery but limited spatial coverage. Conversely, high-resolution satellites cover larger areas but are costlier, prone to cloud interference, and have lower temporal resolution. Integrating high-resolution RS into crop models improves spatial representation and yield estimation accuracy. Some studies enhance reliability further by combining multiple high-resolution RS sources to refine spatial and temporal estimates of crop state variables. The CGMs lack vegetation reflectance and spatial characteristics, which are a key basis of remote sensing satellite data. To address this lack radiative transfer models (RTMs) have been increasingly coupled with crop models. RTMs simulate the interaction of electromagnetic radiation with vegetation canopies and the atmosphere, enabling the retrieval of leaf- and canopy-level variables through model inversion with limited ground data. However, there is a rising need to investigate reliable retrieval techniques to achieve quantification and retrieve spatiotemporal information about crop traits. Frequently assimilated variables include leaf area index (LAI), aboveground biomass, canopy nitrogen accumulation, phenology, and chlorophyll content. Recent literature indicates that only a limited number of RTMs have been integrated with CGMs. Specifically, four RTMs (PROSAIL, 6S, SLC, and FLiES) have been coupled with ten different CGMs, including APSIM, WheatGrow/InfoCrop-Wheat, Oryza, and WOFOST. Based on this scientific background, this research lays the methodological and conceptual groundwork for the future integration of crop growth models (DSSAT) with radiative transfer models (SCOPE) to enhance the quantification and prediction of crop yield gaps under current and future environmental conditions by integrating multi-scale empirical evidence, remote sensing observations, and process-based crop modeling. Specifically, the research objectives are to: 1. Systematically quantify, through a global meta-analysis, the effects of water stress and combined water and biotic stress on crop yield gaps. 2. Evaluate the inversion of the SCOPE radiative transfer model for estimating key crop biophysical variables, leaf area index, leaf chlorophyll content, and canopy chlorophyll content, using multispectral and hyperspectral UAV imagery, with a focus on durum wheat and potato. 3. Assess and simulate the impacts of climate change on durum wheat yield across representative Mediterranean environments using the CERES-Wheat model. The meta-analysis shows that water stress reduced yields by 36% per hectare and 45% per plant, with Pulses, Cereals, and Horticultural crops most affected—Pulses exhibiting losses up to 62% per plant. Biotic stress further amplified these losses, as observed in melon, where fungal infections increased yield gaps from 24% to 63%. Climate zones also played a significant role, with reductions reaching up to 38% in warm temperate regions with dry summers. The results of the first chapter of this thesis underscore the need for field-based data to refine global yield gap estimates and to strengthen local agronomic research for climate adaptation. The inversion of SCOPE showed good agreement between vegetation parameters extracted from remote sensing and those measured in the field, and optimization of the cost function further improved accuracy. These results highlight the potential of SCOPE inversion for robust, physically based monitoring of crop biophysical and biochemical traits. Modeling the impact of climate change on durum wheat yields across representative Mediterranean environments enabled accurate yield predictions under future climate scenarios. Simulations identified irrigation as a key adaptation strategy to reduce yield losses and highlighted the importance of integrated management, supported by process-based crop models, for developing climate-resilient Mediterranean cropping systems.