Conference Agenda

Food security remains a critical global concern, exacerbated by climate change-induced disruptions in weather patterns, leading to more frequent and severe extreme weather events. Addressing these challenges demands innovative solutions in the agricultural sector. Remote Sensing (RS) offers a promising avenue for monitoring and mitigating agricultural stressors, yet, integrating the increasing quantity and quality of Earth Observation (EO) data holistically is a challenge. The EO4CerealStress project aims to overcome this obstacle by developing a new framework for processing these multisource data and monitoring multiple stressors in cereal-based agricultural systems. Leveraging optical sensors, along with in-situ measurements, the project focuses on detecting the impact of various stressors such as salinity, nutrient deficiency, water stress, lodging, etc. on crop productivity. This synergistic approach enables continuous monitoring and a comprehensive understanding of the complex interactions between different stress factors and their impact on crop yield. So far, intensive data collection has been conducted at three pilot sites in Austria, Italy, and Spain. The further analysis integrates data from new EO missions such as PRISMA, ENMAP, and Sentinel-2 measurements with the ground and airborne sensors like Headwall and Analytical Spectral Devices (ASDs) and utilizes vegetation radiative transfer models and machine learning algorithms to identify crop stress indicators, contributing to the Agricultural Science Precursor Experimental Dataset. The project will also explore the operational use of developed products and algorithms at new locations across Europe and Canada, ensuring that its outcomes translate into tangible solutions for agricultural challenges. Through its comprehensive approach and practical implementation, EO4CerealStress aims to lay the foundation for building resilience within agricultural systems.

Drought denotes a prolonged water supply deficit impacting various realms such as the atmosphere, soil, streamflow, groundwater, and economic activities. It has posed substantial challenges to Africa's food security and water resource inequality. Therefore, urgent efforts are needed to monitor agricultural drought in sub-Saharan Africa. Diverse drought indices have been developed, reliant on meteorological variables and remote sensing (RS) data. However, the efficacy of meteorological drought indices on a regional scale is hindered by the limited distribution of in-situ sites, and the indices derived from modeled data have not been evaluated in Africa. RS-based drought indices typically normalize indirect indicators derived from vegetation and land surface temperature (LST) anomalies; however, they face limitations in temporal sampling frequency and cloud cover. Additionally, current gap-free soil moisture (SM) products still grapple with coarse spatial resolutions, rendering them unsuitable for local irrigation management.

In this study, in-situ SM measurements from the International Soil Moisture Network (ISMN) in Africa serve as the ground truth for agricultural drought. The SMAP SM product, the ESA Soil Water Index (SWI), the Keetch-Byram Drought Index (KBDI), the Shortwave Infrared Transformed Reflectance (STR), the hybrid Scaled Drought Condition Index (SDCI), the Normalized Difference Water Index (NDWI), and the ECOSTRESS Evaporative Stress Index (ESI) are included for sensitivity analysis to soil moisture for different climate and land cover types. Results reveal SMAP's greater performance, followed by SWI. STR correlates with SM but includes scattered values. ECOSTRESS ESI effectively captures the spatial nuances of local drought stress at the seasonal scale; however, it is limited by the sampling frequency, impeding the variability analysis at intra-monthly scales. No single index universally excels, underscoring the need for refinement. Advocating for a high-resolution RS data-driven drought index, this study provides insights for future mission applications, offering a roadmap for enhanced drought monitoring in Africa.

Drought significantly impacts agricultural systems, affecting crop yields, food security, and socio-economic stability. Earth Observation (EO) data enhances drought monitoring, providing insights into crop conditions in near-real time. Yet, current monitoring primarily identifies drought hazards, not their impacts or risks. Understanding these requires context-specific information on management and cropping systems. Our study, conducted in Kenya, co-developed solutions with stakeholders to create EO-based products assessing drought risk and impacts, using Sentinel-2 and Sentinel-3 data and evaluating crop condition, evapotranspiration, and farming systems (irrigated/rainfed, mono/mixed cropping). The Sentinel-2 time series and vegetation indices were used to assess agricultural impacts by tracking crop changes and classifying drought-affected areas with a random forest method. National-scale maps for irrigated/rainfed areas were produced using random forest and harmonics, and Sentinel-2 and PlanetScope data fusion was tested to map mixed cropping systems using Convolutional Neural Networks. Crop yield data and biophysical predictors (SPI, NDVI, NDII, LST, albedo) informed a drought hazard model. Calibration of MODIS and Sentinel-3 data extended the time-series analysis for LST, NDVI, and NDII. The project linked drought hazard and impact data with information on farming systems, incorporating socio-economic and environmental data for a comprehensive risk assessment. Furthermore, Sentinel-2 and -3 data were used to derive daily 20-m evapotranspiration estimates using machine learning and energy balance models. The crop condition accuracy ranged from 75-90%, and farming systems classification accuracy was 97.87%. A static drought vulnerability map, combined with hazard/exposure data, visualized monthly drought risk at a 1 km resolution. The developed products showed high agreement with existing datasets, confirming their reliability in drought risk and impact assessment.

Drought Early Warning Systems (DEWS) are crucial components of a proactive, risk-based management approach. However, these systems often fall short in providing accurate and detailed impact-based seasonal forecasts at more localized spatial scales, such as small and moderate-sized basins or agricultural districts.

The InfoSequia-4CAST service, a project supported by the ESA-InCubed program, aims to deliver seasonal forecasts of risk of impact on crop yield and water supply. This is achieved by merging EO-based and climate indices and machine learning. Two decision tree-based ensemble forecasting methods, resting on the Fast and Frugal Trees and XGBoost techniques, have been tested in Mozambique and Spain. The design and evaluation of this tool and its outcomes have been collaboratively supported by local stakeholders.

Forecast models are trained and calibrated using a large dataset of enhanced drought predictors generated across several timescales, and with data collected from multiple sources and satellite sensors. For the crop yield pilot case tested in Mozambique, anomalies of the End-Of-Season Water Requirement Satisfaction Index (EOS-WRSI) is used as a proxy of crop performance. Water level observations in reservoirs serve as the basis for forecasting water supply scenarios in the Spanish case. InfoSequia-4CAST generates probabilities of failure predictions up to six months in advance, metrics of forecast performance and recommendations for early action. Outcomes are provided monthly to stakeholders to support the decision-making process. The system has been tested in a real-world operational context during the 2023-2024 season.

InfoSequia-4CAST has shown promising performance, meeting the key requirements previously set by local stakeholders. Further technical developments may improve InfoSequia-4CAST, including a) the employment of drought precursors able to better detect compound extremes and flash events, b) the combination of statistical and dynamic forecasting methods, and c) the retrieval of alternative and more accurate predictands when ground-based impact observations are lacking or unreliable.

The destructive insect known as Cotonet de les Valls (Delotococcus aberiae) in the province of Castellón (Spain) is causing significant economical losses in the Spanish agricultural sector, particularly in citrus fruits. The European Copernicus program has enabled the creation of numerous agricultural surveillance instruments by utilizing remote sensing technology. In this context, the objective of this research is to comprehend how light reflectivity changes based on the level of tree infection by examining temporal data from satellite. Sen2Like processor is used to retrieve the images. Furthermore, given the morphology of the studied crop (trees), shadows can introduce noise in terms of the spectral response. This is why the Bidirectional Reflectance Distribution Function (BRDF) is used to minimize the angular effects [1]. This investigation was carried out in the vicinity of Vall d'Uixó (Castellón, Spain) by analyzing around 25 hectares of land spread across various orange tree fields affected by cotonet, each with varying degrees of infestation, which were categorized as either healthy or diseased during the 2020-2021 season. Initially, we explored the connection between the cotonet infestation level and various optical bands (such as RED, NIR, SWIR, derived from Sen2Like), along with the Normalized Difference Vegetation Index (NDVI). To mitigate seasonal variations and concentrate on trend analysis, monthly linear regressions were applied to each group of fields and spectral range. The findings indicate that remote sensing data can be instrumental in the timely, objective, and cost-efficient management of the cotonet pest. It has been observed that it is feasible to distinguish between affected and healthy fields throughout the year using specific spectral ranges, with SWIR demonstrating particular efficacy, enabling differentiation throughout the latter half of the year. This study contributes to the advancement of novel surveillance tools for effective and sustainable measures against agricultural threats.

Advances in the quality and accessibility of Earth Observation (EO) information have led to rapid advances in data driven decision support, especially in pest risk. Historically, applications associated with pest management have focussed on the monitoring and detection of pest incursions, however in many cases early intervention is required before detectable damage has occurred. Where preventative action is needed, strong linkages with agricultural extension systems are required to understand how information can better inform preparedness and decision making. Here we describe the development of a suite of projects that use optical, radar and weather EO data products combined with ecological modelling methods to provide information to farmers and decision makers on when to intervene and where risks will be highest on a broader spatial scale. We describe how actors and decision makers are involved in the design process to ensure maximum impact of information. Firstly, we describe the Pest Risk Information SErvice (PRISE) which uses ERA5 weather data to produce advisories for smallholder farmers on when to intervene against pests commonly found in mixed maize growing systems in Africa. Secondly, we describe how spatial pest risk estimations can be improved by using Sentinel 2 datasets to improve the mapping of when and where irrigation occurs. Thirdly, we describe a framework for application of EO data to biosecurity decision making. A layered approach, overlapping temporal and spatial crop maps with environmental suitability modelling, can identify areas of high risk of the wheat blast pathogen. This information can guide where to use hyperspectral detection methods for emerging outbreaks in inaccessible areas.

.

Abstract: Agricultural productivity is influenced by various natural factors, making its assessment critical for ensuring food security. In this study, we introduce a robust empirical index, the Land Productivity Index (LPI), designed to evaluate the relative productivity of agricultural areas globally. The LPI integrates key environmental parameters including climate, soil attributes, and water availability, providing a comprehensive assessment of agricultural potential at a spatial resolution of 250 meters.

Priority 1: Monitoring stressors and changing growing conditions The LPI methodology enables the monitoring of stressors and changing growing conditions by incorporating parameters such as mean monthly temperature, evapotranspiration, soil pH, organic carbon content, and water holding capacity. Through this holistic approach, we identify areas prone to environmental stressors, facilitating targeted interventions for sustainable agriculture.

Priority 2: Supporting increased resilience and risk management To support increased resilience and risk management, we extend the LPI methodology by integrating the ECOCROP database of the Food and Agriculture Organization (FAO). By customizing the LPI for key crops, we assess crop suitability across different spatial and temporal scales. This enables us to provide valuable insights into irrigation advice, crop monitoring, and risk mitigation strategies, enhancing agricultural resilience in the face of evolving environmental challenges.

Priority 3: Analyzing historical trends and variations Furthermore, our study analyzes historical trends and variations in land productivity by assessing historical average conditions and deviations from these averages over time. By identifying spatial shifts and temporal trends, we gain a deeper understanding of agricultural dynamics, empowering policymakers and stakeholders to make informed decisions for sustainable agricultural development.

Conclusion: Our integrated approach combining Earth observations with crop suitability analysis offers a comprehensive framework for assessing agricultural productivity and resilience under changing environmental conditions. By addressing the priorities of monitoring stressors, supporting resilience, and analyzing historical trends, our methodology contributes to enhancing agricultural sustainability and food security on a global scale.

This abstract introduces a comprehensive framework aimed at enhancing climate adaptation strategies for maize cultivation, focusing on the evolution of the Plan-T platform developed as proof-of-concept for the Food and Agriculture Organization of the United Nations (FAO) in the framework of the ESA project EO4YEMEN.

The methodology foresees three main steps: crop seasonality assessment, crop yield estimation and identification of the optimal planting date.

The assessment of crop seasonality integrates diverse data sources, including meteorological, agronomic, and satellite data. Advanced techniques such as the integration of ECOSTRESS data (Evapotranspiration) and climate stressor analysis contribute to more accurate variety suitability mapping.

The enhancement of crop productivity estimation involves refining the AQUACROP agronomic model, incorporating new input data, and calibrating model parameters. This process includes refining varietal parametrization, modeling yield response, and validating outputs with field data.

Variety selection and optimal planting date assessment are based on detailed analysis of physiologic responses to stressors and soil water balance, utilizing climate and soil moisture data as well as short-term high resolution weather forecast from ECMWF.

The Plan-T service is offered through a dedicated web application that can be exploited on both desktop and portable devices. It allows selecting the site of analysis down to a single field, or manually providing coordinates. Once done, it automatically shows the soil chemical parameters and ranks the maize variety based on climate suitability and productivity. It finally allows assessing the optimal planting date for each variety.

Overall, this framework facilitates informed decision-making for climate adaptation in maize cultivation, providing valuable insights into variety suitability, planting dates, and potential crop productivity. The integration of advanced methodologies and tools within the Plan-T platform enhances its usability and effectiveness in supporting agricultural practices amidst changing climatic conditions.

Recent studies on the impact of climate change on agricultural systems show significant changes that will occur in the global food production chain. New climatic conditions are expected to have a severe impact on the productivity of main crop types such as wheat, maize, soybean, and corn [1]. This will alter the global map of agriculture in the 21st century, facilitating crop migration and changes in agricultural practices. This process is ongoing and can already be observed in historical cropland maps [2].

To study this process, we employed climate velocity principles, allowing us to provide spatial-temporal characteristics of climate change and generate flow fields for climate parameters [3]. This technique was applied to a collection of 20 years of MODIS-based winter wheat maps to uncover spatial-temporal trends in winter wheat production changes and estimate the directions of winter wheat expansion and migration. This data on "Winter Wheat Velocity," together with Climate Velocity maps built based on ERA-5 Cumulative Biological-Active Temperature for winter crops, allowed us to link changes in wheat production with changes in climate and estimate the percentage of wheat migration and expansion caused not only by agricultural intensification but also by environmental changes in Europe.

References:

[1] Jägermeyr, Jonas, et al. "Climate impacts on global agriculture emerge earlier in new generation of climate and crop models." Nature Food 2.11 (2021): 873-885.

[2] Sloat, Lindsey L., et al. "Climate adaptation by crop migration." Nature communications 11.1 (2020): 1243.

[3] Loarie, Scott R., et al. "The velocity of climate change." Nature 462.7276 (2009): 1052-1055.

Rising temperatures, unpredictable precipitation patterns and extreme climate events are disrupting crop production globally. This is a major challenge for rice production, the key staple food for over half of the world’s population and a primary calorie source for millions of vulnerable people. The TAPAS platform combines remote sensing data and AI to monitor both environmental stressors and cultivation conditions of crops, with potential to positively impact on crop production by enabling decision-making for more resilient rice cropping systems. We identify, quantify and track crop production anomalies using a combination of supervised and unsupervised algorithms. We combine archival and real time Earth Observation (EO) data augmented by machine learning and deep learning techniques to monitor rice production quality, yielding up-to-date analyses specific to climate-impacted hotspots. Applying classification algorithms to Landsat imagery using periodic spectral information allows us to monitor changes or patterns in annual land use mapping that are specific to rice cropping cycles. Historical MODIS data (2000-2014) is used to train Long Short Term Memory Network models (LSTM) for predicting such cycles from 2015 onwards. This establishes a ‘climate baseline’ for previous decades which allows us to determine crop cycle performance relative to this baseline by examining differences in predicted and observed NDVI. Our method allows a remote-sensing only approach to measuring climate change adaptation of crops, validating the use of deep learning and EO in extracting key insights into climate adaptation strategies. The role of AI in EO to model climate effects and interacting geo-factors has implications for food production sustainability and decision making in reaching Sustainable Development Goal 2 targets. Overall, our approach has the potential to support evidence-based scaling of climate smart agriculture practices leading to more resilient agrifood systems, while providing critical crop productivity specific intelligence that can help mitigate food insecurity.

Rice is the staple food for half the world's population. Around 90% of the world's rice is produced in Asia, particularly in the main river deltas. However, due to their geographic location and low altitude, Asian deltas are classified among the regions most vulnerable to the impacts of climate change. Additionally, impacts are exacerbated by the activities of rapidly growing populations.

Within the Space Climate Observatory program, the objective of the VietSCO project is to study such impacts on the Mekong Delta, one of the most important rice deltas in the world. Sentinel-1 data was used to determine rice area, crop calendar, number of crops per year, etc. and to assess the changes that have occurred over the past decade. ALOS-PALSAR data was used to detect the flooding status of rice fields, which determines methane emissions.

It is seen that the habitat suitable for rice has undergone drastic changes. Options have been identified to reduce and adapt to the risks of increasing salinity intrusion, drought and flooding, and land submergence. Among mitigation practices, we highlight practices that mitigate methane emissions from rice. For the Global Methane Pledge, rapidly reducing methane emissions from rice fields is considered the most effective strategy to reduce global warming. Work is currently underway to evaluate the use of ALOS PALSAR, calibrated by automatic water level devices, to locate continuously flooded rice fields, where incentive actions must be taken to reduce methane emissions through proper drainage practices.

The work carried out aimed to arouse the interest of local and national decision-makers in the use of Earth observation to monitor the evolution of rice production under the impacts of climate change and anthropogenic pressures. To this end, a platform has been created to allow decision-makers to visualize current and projected impacts according to scenarios. Finally, we will discuss the potential for application of the approach developed to other large deltas in Asia.

.

Smart-farming applications have seen a steep rise in the incorporation of EO data, mainly due to the availability of Sentinel 1 and 2, and to some extent the increased availability of affordable high-resolution satellite data. This increased uptake also results in ever increasing demands by the users in the level of detail as well as thematic accuracy of the products. While some of these demands can be met with high-resolution satellites, other demands and use cases remain too challenging. To fill this monitoring gap, Unmanned aerial vehicles (UAVs) or drones have become a popular and useful tool to identify and map (a)biotic stress at (sub)cm resolution. However, an important limitation remains the poor scalability due to the high costs and the need for manual, specialized, interventions. Here, a solution is proposed to circumvent these limitations, bringing UAV-based crop monitoring a step closer as a complementary companion to the satellite-based monitoring tools. As a first step, methodologies are presented with small, cheap drone systems that do not require elaborate training or licenses to operate. Secondly, a drone sampling methodology is proposed to optimize the cost-efficiency of data collection and processing, based on a direct georeferencing workflow to project the individual images on a regional Digital Surface Model. This allows for low or even negative image overlap which will be at least 70% more efficient compared to a traditional Structure-From-Motion workflow. Image analysis is performed on the individual images to identify plants or plant features like weeds, disease spots or flowers. To scale-up the adaptation of the drone sampling technology, the workflow has been integrated into MAPEO, VITO’s drone processing platform, culminating in the generation of consistently accurate and reliable data products. The overall workflow will be demonstrated with several use cases like agricultural insurances and invasive weed detection.

Spatially explicit information on land cover (LC) is commonly derived using remote sensing. The lack of training data remains a major challenge to produce accurate LC products. Although in-situ surveys are generally regarded as reliable information sources, there tend to be inconsistencies between in-situ LC data and the information derived from satellites due to the different view points.

Here, we develop a computer vision methodology to extract LC information from photos from the Land Use-Land Cover Area Frame Survey (LUCAS). A representative sample of 1120 photos covering eight major LC types across the European Union was selected. We then applied semantic segmentation to these photos using a neural network (Deeplabv3+) trained with the ADE20k dataset. For each photo, we extracted the original LC identified by the LUCAS surveyor, the segmented objects, and the pixel count for each ADE20k class. Using the latter as input features, we then trained a Random Forest model to classify the LC of the photo. Examining the relationship between the objects/features extracted by Deeplabv3+ and the LC labels provided by the LUCAS surveyors demonstrated how the LC classes can be decomposed into multiple objects, highlighting the complexity of LC classification from photographs. The results of the classification show a mean F1 Score of 89%. In a stage, we analyzed the semantic gap between this subset of LUCAS 2018 in-situ LC data and three high-resolution thematic LC products derived from satellite data, namely, Google’s Dynamic World, ESA’s World Cover, and Esri’s LC maps. Following an experimental analysis, we explore the feasibility of using geo-referenced street-level imagery to bridge the gap between information provided by field surveys and satellite data.

The spatial extent of croplands globally is a crucial input to global and regional agricultural monitoring systems. Although many new remotely sensed products are now appearing due to recent advances in the spatial and temporal resolution of satellite sensors, there are still issues with these products that are related to the definition of cropland used and the accuracies of these maps, particularly when examined spatially. To address the needs of the agricultural monitoring community, here we have created a hybrid map of global cropland extent at a 500 m resolution by fusing two of the latest high resolution remotely sensed cropland products: the European Space Agency’s WorldCereal and the cropland layer from the University of Maryland (GLAD croplands). Since the fallow class is included in the GLAD croplands map and we consider this relevant for food security applications we use this layer as in inital base map. We then aggregated the two products to a common resolution of 500 m to produce percentage cropland and compared them spatially, calculating two kinds of disagreement: density disagreement, where the two maps differ by more than 80%, and absence-presence of cropland disagreement, where one map indicates the presence of cropland while the other does not. Based on these disagreements, we selected continuous areas of disagreement, referred to in the paper as hotspots of disagreement, for manual correction by experts using the Geo-Wiki land cover application. The hybrid map was then validated using a stratified random sample based on the disagreement layer, where the sample was visually interpreted by a different set of experts using Geo-Wiki. The results show that the hybrid product improves upon the overall accuracy statistics of the individual layers, but more importantly, it represents a better spatially explicit cropland layer for early warning and food security assessment purposes.A version of the map has been used as part of JRC's early warning systems ASAP.

Ground truth reference data availability has long been a significant limiting factor for remote sensing data-driven model training. A vastly underutilized source of data comes from administrative data collected by mainly public service bodies. A demonstrated successful utilization is the EuroCrops project where we created training labels for crop type classification from farmers’ self-declarations on the crop types they cultivate. However, a key challenge of the project was the lack of domain-specific expert knowledge and incompatibilities between data released by different governments.

Recently, large language models (LLMs) have demonstrated remarkable abilities in information extraction and synthesis. We therefore propose training an LLM-based domain-specific foundation model to automate the data extraction and harmonization process from administrative data to ready-to-use training data.

This presentation will discuss a toy example of training a small language-based transformer model using the administrative data collected during EuroCrops with agriculture-based text data such as Wikipedia articles.

Preliminary experimentation with open-source not field-specific LLMs shows that they do offer some interesting domain-specific insights when decoding acronyms. However, they do not outperform traditional machine translation methods yet, such as Google translate, and at times output extremely believable but misleading results in the niche field of agriculture. Such bogus generations highlight the need for a grounded generation mechanism in the usage of LLMs.

.