3  Introduction

3.1 Background

Earth Observation (EO) stands at a pivotal juncture in its capacity to transform agricultural statistics. For National Statistical Offices (NSOs) – our primary audience – this handbook arrives at a critical moment. The integration of EO into agricultural statistics is not just a technical opportunity; it is a policy imperative. As countries strive to meet Sustainable Development Goals (SDGs), ensure food security, and respond to climate volatility, the ability to generate timely, reliable, and spatially explicit agricultural data has become essential.

While satellite data availability has expanded dramatically—most notably through the Copernicus Sentinel missions, which now provide near-daily global coverage at 10–20m resolution (ESA, 2023; Drusch et al., 2012)— the operational use of these resources by NSOs remains limited. This persistent gap is striking, given the extensive scientific literature showcasing advanced techniques for land cover and crop type mapping. Only 25–35% of national agricultural monitoring systems in Africa, South Asia, and Southeast Asia have operational EO-based capabilities for crop area and yield estimation—reflecting a notable gap between EO’s technical potential and its institutional adoption in developing economies (Whitcraft, A. K. et al., 2020).

3.2 Improving the use of EO in agricultural statistics

The reasons for the disconnect between EO’s potential and its adoption by NSOs are multifaceted. First, data abundance itself introduces complexity. Platforms like the Copernicus Data Space Ecosystem and Digital Earth Africa offer petabyte-scale Analysis-Ready Data (ARD), but turning this into actionable agricultural statistics requires high-performance computing and structured processing chains. While countries such as Poland (ESA, 2023; CloudFerro, 2025) and Brazil (Gomes et al, 2021) have established national HPC ecosystems to process continental-scale EO data cubes, many NSOs lack comparable resources.

Second, research-grade EO solutions often falter in operational settings. Deep learning models may achieve over 90% accuracy under experimental conditions (Rubwurm et al, 2024)), but their deployment in national programs is challenged by inconsistent training data, cloud interference, and the need to reconcile statistical rigor with time-sensitive reporting cycles (De Simone et al., 2025). Operational maturity demands more than algorithmic innovation—it requires adaptation, institutional readiness, and governance.

To meet these challenges, the global EO community has coalesced around building robust toolchains. ESA’s Sen4Stat platform—built on GDAL and incorporating Orfeo Toolbox components—offers a reproducible pipeline for crop mapping and yield forecasting (Bontemps et al., 2023). Similarly, Brazil’s SITS R package enables scalable time-series classification and is now deployed in Chile’s national land accounts (De Simone et al., 2025). But beyond individual tools, a more integrated coordination framework is emerging.

At the center of this effort is GEOGLAM (Group on Earth Observations Global Agricultural Monitoring)—an open international community of practice uniting the world’s leading programs in crop monitoring. GEOGLAM orchestrates a diverse set of platforms and toolboxes for agricultural monitoring such as GEOGLAM Crop Monitor, ESA Sen4Stat, ESA WorldCereal, NASA Harvest GLAM, JRC ASAP, Aircas CropWatch, FAO EOSTAT, FAO GIEWS, FAO WaPOR. These platforms and toolboxes collectively advance the capacity of countries to monitor agricultural production through shared data, methodological standardization, and transparency. Importantly, GEOGLAM is increasingly focused on strengthening the statistical rigor of crop monitoring—aligning more closely with NSO requirements for official area and yield estimation. This handbook complements that ambition, offering tools, case studies, and software that help bridge EO-based crop monitoring with statistical reporting.

In parallel, the Food and Agriculture Organization (FAO) has played a critical role in enabling country-level capacity development. The UN Task Team on Earth Observations for Agricultural Statistics, established under the UN Committee of Experts on Big Data and Data Science and the UN Expert Group on Rural, Agricultural and Food Security Statistics, has driven collective progress in EO adoption by NSOs. Through documented contributions to the UN Statistical Commission, the Task Team has promoted South–South cooperation, capacity building, and joint projects across regions. Recently, it has deepened collaboration with the UN Global Platform, particularly through increased engagement with the UN Big Data Regional Hubs. In this context, the Global Hub for Big Data in China has taken a leading role in supporting this handbook initiative as a key deliverable for operationalizing EO within national statistical systems (UNCEBD, 2025).

Accuracy, in this context, is non-negotiable. As shown by Olofsson et al. (2024), classification errors above 15% can lead to over 20% distortion in area statistics—errors that compromise national reporting and policy formulation. This handbook addresses these issues across the full statistical workflow: from building ARD data cubes to implementing bias-corrected estimators, quantifying uncertainty, and generating statistically valid crop area measures. Real-world deployments, such as Chile’s 30m-resolution crop maps and Indonesia’s use of active learning to cut labeling costs by 40% while maintaining 90% detection accuracy, demonstrate the feasibility of integrating EO into official systems.

3.3 Purpose of this Handbook

Drawing on lessons from earlier efforts—including the 2017 FAO Handbook on Remote Sensing for Agricultural Statistics co-authored with Jacques Delincé (FAO, 2017)—this volume presents a practical, use-case-driven guide for EO adoption in statistical production. It reflects a significant paradigm shift from theoretical principles to field-tested applications.

What sets this handbook apart is its emphasis on reproducibility and reusability. Each chapter includes executable R and Python scripts that allow NSOs to directly implement the methods discussed. Whether using Sen4Stat’s GDAL workflows, SITS classifiers, NASA Harvest’s ARYA crop yield model (Becker-Reshef et al., 2023), or PRESTO, a transformer-based model for time-series feature extraction (Tseng et al., 2024), the focus is on transparency and adaptability. These are not theoretical tools—they are field-tested systems embedded in national workflows.

The Handbook showcases diverse cases contributed by a wide range of national statistical offices and research teams, including crop monitoring with SAR images in Poland (GUS), crop classification in Mexico (INEGI), Senegal (DAPSA and FAO EOSTAT), Zimbabwe (ZIMSTAT and FAO EOSTAT), Chile (INE-Chile), and digital workflows supported by Digital Earth Africa. Yield estimation use cases span countries such as Finland (LUKE), Indonesia (BPS), Poland (GUS), Colombia (DANE and FEDEARROZ), Ukraine (University of Strasburg and Ukraine Statistics) and China (Institute of Remote Sensing and Digital Earth, Zhejiang University, Chinese Academy of Science). These contributions represent collaborative efforts involving institutions such as Université Catholique de Louvain (UCLouvain), University of Strasbourg (UNISTRA), FAO EOSTAT, Statistics Indonesia, IBGE, INEGI, and VITO, among others, alongside international research bodies and space agencies.

Each contribution has been shaped through close cooperation between NSOs and geospatial experts. Their expertise, spanning the academic, operational, and policy realms, ensures that the methodologies described are both advanced and implementable.

Beyond showcasing results, the Handbook places special emphasis on methodological components critical for statistical integrity. These include quality assessment of in situ data, evaluation of survey design, validation of classification models through confusion matrices, and estimation of area statistics corrected for classification bias. For instance, recent advances described by De Simone et al. (2025) propose rigorous quality control protocols for training data using self-organizing maps (SOM) and SMOTE-based sample balancing, offering a path to significantly higher classification accuracy.

Innovation is also explored in chapters addressing frontier topics such as crop yield estimation (UNISTRA), field boundary mapping (e.g., IBGE), and the role of drones and EO in disaster risk reduction. Of relevance is the contribution on the WorldCereal project (Van Tricht et al., 2023), coordinated by VITO, which highlights the potential of self-supervised learning and global pretraining strategies for democratizing access to high-performance EO models in data-scarce regions. These developments reflect continuing validation efforts in operational EO workflows now being adopted in countries like Senegal and Zimbabwe.

The Handbook also acknowledges the evolution of toolboxes supporting these efforts. The Sen4Stat system, developed by UCLouvain with ESA support, has become a key asset for NSOs aiming to operationalize EO workflows for agricultural monitoring (Bontemps et al., 2024). Its uptake in Senegal and Zimbabwe illustrates its growing relevance. In a sense, the Handbook captures the evolution from research frontier to operational solutions. It represents the joint aspirations of NSOs, the geospatial science community, and international organizations to reimagine agricultural statistics for the era of big data and planetary monitoring. Throughout, the Handbook emphasizes transparency, trust, and standardization. These are not merely technical ideals but are foundational to the credibility of agricultural statistics in national and global decision-making. By promoting harmonized protocols and open-source tools, this volume seeks to foster a common language between statisticians and EO specialists, thereby enabling more consistent and comparable agricultural data worldwide.

The journey from satellite pixels to policy-relevant statistics is complex—but no longer aspirational. With 15 real-world use cases, field-tested methodologies, and ready-to-use software, this handbook aims to help NSOs move from exploration to execution, and from data collection to decision-making.

We invite readers to see this Handbook not only as a reference but as a launchpad. The time has come to embed EO firmly into the statistical systems that underpin food security, rural development, and environmental sustainability. Let this volume guide that journey—methodologically rigorous, globally informed, and grounded in practice.

3.4 Future of satellite mission and its relevance to NSOs

As the EO landscape evolves, so too must national strategies. In 2025, NASA announced the discontinuation of Landsat Next—once intended to continue the legacy of the world’s longest-running EO mission. This unexpected restructuring highlights the fragility of EO continuity and the urgent need for risk mitigation. To this end, the handbook includes a dedicated chapter to help NSOs develop resilient, multi-source EO strategies that reduce dependency on any single mission or platform.

Fortunately, efforts are underway to secure long-term EO data availability. The European Space Agency, through its Sentinel Expansion and Copernicus Next Generation missions, is investing in hyperspectral, L-band radar, and thermal infrared sensors designed to ensure continuity and enhance capacity. These investments, aligned with GEOGLAM’s coordination, provide a sustainable pathway for countries to anchor EO in their statistical systems.

3.5 Summary

The journey from satellite pixels to policy-relevant statistics is complex—but no longer aspirational. With 15 real-world use cases, field-tested methodologies, and ready-to-use software, this handbook aims to help NSOs move from exploration to execution, and from data collection to decision-making.

References

Becker-Reshef, I., Barker, B., Humber, M., Hosseini, M., & Justice, C. (2023). NASA Harvest’s Open Crop Yield Models: ARYA Performance and Global Applications. Frontiers in Sustainable Food Systems. https://doi.org/10.3389/fsufs.2023.1194332

De Simone, L., Pizarro, E., Paredes, J., et al. (2025). Quality Control of Training Samples for Agricultural Statistics Using Earth Observation. Statistical Journal of the IAOS, 0(0). https://doi.org/10.1177/18747655251338033

Olofsson, P., Foody, G. M., Herold, M., Stehman, S. V., Woodcock, C. E., & Wulder, M. A. (2014). Good Practices for Estimating Area and Assessing Accuracy of Land Change. Remote Sensing of Environment, 148, 42–57. https://doi.org/10.1016/j.rse.2014.02.015

FAO. (2017). Handbook on Remote Sensing for Agricultural Statistics. Food and Agriculture Organization of the United Nations.

UNCEBD. (2025). Report of the Task Team on Earth Observations for Agricultural Statistics to the UN Statistical Commission. United Nations Committee of Experts on Big Data.

Van Tricht, K., Degerickx, J., Gilliams, S., et al. (2023). WorldCereal: a dynamic open-source system for global-scale, seasonal, and reproducible crop and irrigation mapping. Earth System Science Data, 15(12), 5491–5515. https://doi.org/10.5194/essd-15-5491-2023

Bontemps, S., Deffense, N., Nørgaard, B., & Defourny, P. (2024). Sen4Stat – Sentinels for Agricultural Statistics, D16.0 Concept Paper.

ESA, 2023. European Space Agency. Copernicus Sentinel Missions Overview. Available at: https://sentinels.copernicus.eu

Drusch, M., et al., 2012. Sentinel-2: ESA’s Optical High-Resolution Mission for GMES Operational Services. Remote Sensing of Environment, 120, 25–36. https://doi.org/10.1016/j.rse.2011.11.026

Whitcraft, A. K., Becker-Reshef, I., Killough, B., & Justice, C. O. (2020).Meeting Earth Observation Requirements for Global Agricultural Monitoring. Remote Sensing of Environment, 239, 111901. https://doi.org/10.1016/j.rse.2020.111901

Lesiv, M., et al., 2020. Estimating the Global Distribution of Field Size Using Crowdsourcing. Global Change Biology, 25(1), 174–186. https://doi.org/10.1111/gcb.14492

CloudFerro. (2025, February 26). CloudFerro to provide NSIS-Cloud services for Poland [Press release]. CloudFerro. Retrieved from https://cloudferro.com/news/cloudferro-to-provide-nsis-cloud/:contentReferenceoaicite:0

European Space Agency. (2020, November 12). Sentinel data enables new system for agricultural monitoring in Poland. European Space Agency (ESA) – Observing the Earth: Copernicus News. Retrieved from https://www.esa.int/Applications/Observing_the_Earth/Copernicus/Sentinel_data_enables_new_system_for_agricultural_monitoring_in_Poland

Gomes, V. C. F., Carlos, F. M., Queiroz, G. R., Ferreira, K. R., & Santos, R. (2021). Accessing and processing Brazilian earth observation data cubes with the Open Data Cube platform. ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, V-4-2021, 153–159. https://doi.org/10.5194/isprs-annals-V-4-2021-153-2021:contentReferenceoaicite:2

Rußwurm, M., Wang, S., Kellenberger, B. et al. Meta-learning to address diverse Earth observation problems across resolutions. Commun Earth Environ 5, 37 (2024). https://doi.org/10.1038/s43247-023-01146-0

Tseng, G., Cartuyvels, L., Zvonkov, I., Purohit, S., Rolnick, D., & Kerner, H. (2024). Lightweight, pre-trained transformers for remote sensing timeseries (arXiv preprint arXiv:2304.14178). arXiv. https://arxiv.org/abs/2304.14178

Foust, J. (2025, May 21). NASA’s budget crisis presents an opportunity for change. SpaceNews. https://spacenews.com/nasas-budget-crisis-presents-an-opportunity-for-change/?utm_source=chatgpt.com