Abstract

The modernization of agricultural statistics necessitates the integration of innovative data sources to enhance timeliness, accuracy, and spatial granularity while reducing the response burden. This presentation outlines a comprehensive operational system developed for estimating crop areas, assessing crop conditions, and forecasting yields using Earth Observation (EO) data within the framework of official statistics in Poland.
The methodology employs a multi-sensor approach, fusing high-resolution optical data (Sentinel-2) with radar data (Sentinel-1) to ensure continuous monitoring capabilities independent of cloud cover. For crop area estimation, the system utilizes object-based image classification (OBIA) combined with advanced machine learning algorithms, such as Random Forest and Support Vector Machines (SVM). This robust framework supports a multi-temporal estimation cycle. It begins with early spring estimates for winter (including rapeseed) and spring crops, followed by detailed summer estimates in July and September, which allow for the classification of 37 distinct crop species. Finally, an autumn estimate is conducted specifically to distinguish between maize for grain and maize for silage.
Furthermore, the presentation describes a module for crop condition monitoring and yield prediction. This component analyzes time series of vegetation indices derived from Sentinel-3 and MODIS, integrated with agrometeorological data (ERA-5). The system calculates phenological parameters and applies regression models to forecast yields relative to long-term averages.
The results demonstrate that integrating remote sensing with administrative sources offers
a flexible and cost-effective solution for agricultural statistics, delivering precise data critical for decision-making processes and food security monitoring. Authors: Przemysław Slesiński, Natalia Kotulak