研究目的
To explore the applicability of Sentinel-2A for mapping cropping practice, including weed infestation, in Northern Italy using NDVI time-series analysis.
研究成果
Sentinel-2A is applicable for mapping cropping practice with reasonable accuracy, capable of detecting high weed infestations and distinguishing crop phenology. The study provided a detailed map of cropping practices in Northern Italy, highlighting areas with high weed infestation. Future work could benefit from higher temporal resolution with Sentinel-2B or combining with Sentinel-1 data to overcome cloud cover limitations.
研究不足
The temporal resolution of Sentinel-2A is limited by cloud cover, making it difficult to detect precise phenological events like harvest dates. The spatial resolution may not detect low levels of weed infestation. The study relies on unsupervised classification with limited ground truth data, and accuracy for certain classes (e.g., broadleaved trees) is lower.
1：Experimental Design and Method Selection:
The study used an unsupervised classification approach with k-means clustering on NDVI time-series data from Sentinel-2A to map cropping practices. The methodology included analyzing temporal patterns to distinguish crop types and weed infestation levels.
2：Sample Selection and Data Sources:
Five case study fields in Northern Italy with varying levels of common ragweed infestation were selected based on in situ observations. Data included 291 Level 1C Sentinel-2A images covering two tiles (T32TMR and T32TLQ) in 2016, and Corine Land Cover data for non-irrigated agricultural land.
3：List of Experimental Equipment and Materials:
Sentinel-2A satellite imagery, Google Earth for visual assessment, WorldView 2 RGB images from DigitalGlobe for validation, and software tools including Intel Data Analytics Abstraction Library (DAAL) for clustering.
4：Experimental Procedures and Operational Workflow:
Downloaded and preprocessed Sentinel-2A images by removing clouds and artifacts using mask files. Calculated NDVI for each pixel. For case studies, manually removed cloud-affected data points and plotted NDVI time series. For regional mapping, applied k-means clustering to classify pixels into land use classes based on NDVI time series, with accuracy assessment using Google Earth.
5：Data Analysis Methods:
Used Davies-Bouldin index to determine optimal cluster numbers. Visual inspection and comparison with ground truth and high-resolution images for class labeling. Accuracy assessment involved stratified random sampling and confusion matrix analysis.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
