A multimodality test outperforms three machine learning classifiers for identifying and mapping paddocks using time series satellite imagery

Abstract

Rotational grazing in paddocks is a strong indicator of intensive grassland management. Uncertainty in the extent of grassland management intensity is a reported source of uncertainty in greenhouse gas budgeting. This article outlines a method of detecting paddock locations in Sentinel 2 satellite imagery using a statistical multimodality test. The test was compared to three machine learning algorithms (support vector machine, random forest and extreme gradient boosting). Photographic records of the Eurostat 2018 LUCAS survey were used as ground truth data to confirm the presence or absence of paddocks. The multimodality test achieved an overall accuracy of 88.4% versus the best machine learning accuracy of 82.4%. The test was also used to map paddock occurrence at a regional scale in the Republic of Ireland. Overall map accuracy was 85.7% versus validation data. The test can be applied in temperate grasslands with pre-mapped field boundaries where rotational grazing is practiced.

Keywords:

• Land use
• grassland
• intensification
• Earth observation
• remote sensing

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The data that support the findings of this study are available upon request.

Additional information

Funding

This research was funded by Teagasc VistaMilk SFI Research Centre, Moorepark, Co. Cork.