Use Cases in Crop Type Mapping
Outline
The “Use Cases in Crop Type Mapping” section of the UN Handbook (Chapters 15–20) illustrates how the theoretical foundations of Earth Observation (EO) are applied in real-world statistical operations. This section showcases diverse examples from National Statistical Offices (NSOs) and international initiatives, demonstrating the transition from experimental remote sensing to operational agricultural statistics. The section is divided into six chapters, each focusing on a specific country or platform.
Chapter 12, “Crop Monitoring with Sentinel-1 and Sentinel-2 Images in Poland”, is developed by experts of Statistics Poland (GUS). This use case represents a mature, operational system integrated into the European Common Agricultural Policy (CAP) monitoring framework. The system utilizes a fusion of optical (Sentinel-2) and radar (Sentinel-1) data. The inclusion of Sentinel-1 is critical for overcoming cloud cover, a frequent challenge in Northern Europe. It employs machine learning algorithms applied to the result of image segmentation method. The analysis looks at the temporal signature of crops throughout the growing season rather than single snapshots. Target crops are commodities including winter wheat, rape, maize, barley, and rye.
Chapter 13, “Crop Classification in Mexico” presents work done at INEGI** (National Institute of Statistics and Geography of Mexico), which combines image time series from the Harmonized Landsat-Sentinel (HLS) product and Sentinel-1 SAR data. The algorithm uses image segmentation to extract homogeneous polygons from images. Descriptive statistics for each polygon are then used as input to a trained classification model based on the extreme gradient boosting algorithm. The pipeline includes a “prediction mode” to use an already trained model to classify a subsequent year. A key feature is the use of temporal metrics (e.g., greenness trends over time) to distinguish irrigated from rainfed agriculture. The method targets strategic crops for Mexican food security, such as maize, beans, wheat, alfalfa, and chili. The approach aims to produce sub-national (municipal) crop area estimates to supplement the Agricultural Census and traditional surveys, particularly in complex landscapes with small plots.
Chapter 14, “Crop Classification in Zimbabwe” highlights results by the EOSTAT project, led by FAO, which aims to build capacity in developing nations for independent crop monitoring. The methods uses the Sen4Stat toolboox developed at UCLouvain, which is based on the classification of image time series data using a random forest model. The application combines Sentinel-2 time series with in-situ data collection. The chapter emphasizes the integration of field data (collected via mobile tablets/ODK) with satellite processing to train robust models. The primary focus is on maize, the country’s staple food crop, mostly grown by smallholder farmers.
Chapter 15: “Efficient and Reliable Paddy Rice Classification” is authored by researchers of Wuhan University, China. This chapter addresses the specific challenge of mapping paddy rice in complex Asian landscapes. Rice has a unique spectral signature due to the transplanting phase, followed by rapid greening. The method uses Sentinel-1 SAR data to detect this flooding signal through monsoon clouds. The work proposes a new phenology-based vegetation index (PRICOS) for mapping paddy rice, which obtains high classification performance. An important contribution of the work is providing the means for accurate estimation of rice acreage for national food supply monitoring.
Chapter 16, “Crop Classification and Land Use Mapping in Chile” is developed by experts from the National Institute of Statistics (INE) in Chile. This chapter illustrates the use of Earth observation to modernize the agricultural census. The approach uses Sentinel-2 imagery and DEM to classify land cover types. Target crops are a wide variety of horticultural and field crops relevant to Chile’s export-oriented agriculture (e.g., fruit trees, vineyards, annual crops).
Chapter 17, “Agricultural Mapping with Digital Earth Africa”, moves beyond a single country to a continental platform. Digital Earth Africa provides an “Open Data Cube” accessible to African institutions and experts. It describes the use of the DE Africa Sandbox (a Jupyter Notebook environment) which allows users to run pre-packaged R/Python workflows on analysis-ready data without downloading files. The DEAfrica Sandbox allows users to upload local training data and generate their own crop maps using the platform’s computing power, allowing NSOs with limited computing infrastructure to generate their own agricultural statistics using a standardized, scientifically validated cloud platform.