High-resolution mapping of orchard distribution across Italy
Accurate large-scale orchard mapping is crucial for agricultural development and sustainable resource management. However, achieving this on a broad scale presents significant challenges, primarily due to the high costs and complexities associated with in-situ sampling and georeferencing. Remote sen...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-06-01
|
| Series: | Science of Remote Sensing |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2666017225000434 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate large-scale orchard mapping is crucial for agricultural development and sustainable resource management. However, achieving this on a broad scale presents significant challenges, primarily due to the high costs and complexities associated with in-situ sampling and georeferencing. Remote sensing is valuable for accurate mapping of orchard, yet certain areas remain a challenge due to their heterogeneous spatial and temporal nature, influenced by climate, soil, local management practices and other phenomena, that make each orchard unique, especially in highly fragmented landscapes. In this study, we propose a method for large-scale crop mapping at high spatial resolution, using Italy as a case study. We utilized a dense spectro-temporal cube of Sentinel-2 and Sentinel-1 images with pyramidal-based sampling, employing available georeferenced polygons. A probabilistic XGB classifier was applied, followed by a Bayesian approach using prior data from the national statistical system to calibrate the final classification probabilities.The final scope of this study is to enhance the specificity of certain subclasses within Corine Land Cover class 2 in Italy, with a particular focus on the most prevalent orchards. Our method, which combines machine learning with Bayesian calibration, has proven effective in identifying more specific crops that are often present in particular regions and therefore underrepresented at the national level. By increasing the granularity of these subclasses in a LULC coarse dataset, this approach provides improved support for agricultural management, landscape planning, and related sectors, benefiting agricultural authorities, research institutions, and farmers. Additionally, the resulting mapped dataset is made publicly available, promoting broader applications and facilitating further research in agricultural monitoring and management. |
|---|---|
| ISSN: | 2666-0172 |