A framework for mapping conservation agricultural fields using optical and radar time series imagery
(2025) Remote Sensing of Environment : an interdisciplinary journal — Vol. 328, p. 114858 (2025)
Files
1-s20-S0034425725002627-main.pdf
Open Access - Adobe PDF
- 12.82 MB
Details
- Authors
- Abstract
- The importance of conservation agriculture (CA) is undeniable, both for improving soil health and offering a viable path towards achieving carbon neutrality. However, to date, survey statistics on the extent of conservation agriculture were based on farmer declarations or field inspections. This is a major impediment to the promotion or monitoring of conservation agriculture. Here, we collected the management practices of a total of 247 fields under conservation agriculture in the Walloon region of Belgium in 2020–2021, with the aim of developing a classification model for the prediction of conservation agriculture by combining remotely sensed data with census data. We identified seven variables in the model, linked to each of the three main principles of conservation agriculture (crop diversification, maximum soil cover and minimum mechanical soil disturbance). The number of different annual crops and cereals in the rotation was obtained from the agricultural census. For the extent of soil cover, the Google Earth Engine (GEE) platform was used to obtain a time series of optical remote sensing images (2015–2020, Sentinel-2, Landsat-7, Landsat-8) and precipitation data. We then analyzed the variation of spectral indices such as the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Tillage Index (NDTI) and constructed indicators to distinguish between bare soil and cover crop. For minimum mechanical soil disturbance, in addition to the above data, radar data (Sentinel-1) were also obtained from the GEE platform to establish a tillage practice model. Subsequently, the Random Forest (RF) classification method was used to construct a classification model distinguishing fields under conservation from those under conventional practices. The results of a ten-fold cross-validation showed a good overall accuracy of 92 %. The model was utilized to classify the farming systems in all croplands of the Hesbaye region of Belgium. The results show that 15.5 % (2875 fields) out of 18,516 cropland fields can be classified as conservation agriculture. These fields tend to adopt non-inversion tillage and have diverse crop rotations.
- Affiliations
Citations
Zhou, Y., Ferdinand, M., van Wesemael, J., Dvorakova, K., Baret, P., Van Oost, K., & van Wesemael, B. (2025). A framework for mapping conservation agricultural fields using optical and radar time series imagery. Remote Sensing of Environment : an interdisciplinary journal, 328, 114858. https://doi.org/10.1016/j.rse.2025.114858 (Original work published 2025)