Mowing Detection Intercomparison Exercise (MODCiX) – Evaluation of Grassland Mowing Detection Algorithms across Europe

This dataset contains reference data that were used to compare the performance of various mowing detection algorithms in the Mowing Detection Intercomparison Exercise (MODCiX). The geopackage contains the spatial locations of managed grassland areas across Europe along with observed or reported dates of mowing activity. The reference data were gathered from different sources and were reprojected (EPSG:3035), harmonized and filtered. Based on the way of intial data acquisition and reported confounding factors, a quality label was assigned to each reference location. Further details can be found in Schwieder et al., 2026.

The full dataset (modcix_reference_data_2017-2021_public.gpkg) cotains information for the years 2017 - 2021 for selected regions in:

• France
• Germany
• Italy
• Sweden
• Switzerland

It has a total of 3177 features (geometries) with 12 attributes:

• MOD_ID: unique MODCiX ID
•  Mow_1 - Mow_6: Dates of reported or observed mowing activity in the format "YYYY-MM-DD"
• NMow: Number of mowing events
• Region: Label for each region
• Year: Label for each year
• Label: Quality label based on the initial way of data acquisition and/or other known issues (1: high certainty - 3: confounding factors)
• Comment: Additional comments
• geom: geometry infortmation

Evaluation of additional algorithms

For the independent evaluation and comparison of additional algorithms, we prepared a stratified random split in 30% training (*_train.gpck) and 70% validation (*_valid.gpck) data, which can be used to optimize your algorithm. Once you predicted the mowing events for the reference areas, you can fill them in the CSV template provided, add a unique name (Group), define the data (OPT, SAR, OPT_SAR) and method domain (RBA, ML) and upload it to https://modcix.thuenen.de. Please note that if you add your results to the already existing values in the template, you can directly compare your results to the performance of other algorithms in the evaluation app. In this version the table contains the predictions of NOA (National Observatory of Athens; retrained with the shared subset of trainig data) and TIRBA (Thünen Institute rule-based algorithm; subset of the predictions that were shown in the paper). 

If you want to share and permanently include your results for the comparison exercise, you can send your results via e-mail so that they can be included in follow-up versions. Please note that we reserve the right to include in the updated table only results from algorithms that involve significant changes, in order to keep the comparison clear and concise.

-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Acknowledgements: The authors would like to thank ARPEA, particularly Elena Xausa and Gianluca Cantamessa, for their support in facilitating access to the data for the Piedmont region. We thank the field operators from SITES Röbäcksdalen Field Station for collecting the Swedish dataset. We further acknowledge Annika Ludwig for her valuable assistance in interpreting webcam images in South Tyrol.

-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Schwieder, Marcel and Lobert, Felix and Weber, Dominique and Reinermann, Sophie and Asam, Sarah and Sarvia, Filippo and De Petris, Samuele and Borgogno-Mondino, Enrico and Muhuri, Arnab and Oppelt, Natascha and Atzberger, Clement and Tsardanidis, Iason and Kontoes, Haris and Godechal, François and Lucau-Danila, Cozmin and Planchon, Viviane and Garioud, Anatol and Huet, Célestin and Valero, Silvia and Mallet, Clement and Morel, Julien and Rossi, Mattia and Vuolo, Francesco and Dujakovic, Aleksandar and Schaumberger, Andreas and Klingler, Andreas and Holtgrave, Ann-Kathrin and Venter, Zander and Ruth, Sonnenschein and de Vroey, Mathilde and Radoux, Julien and Buck, Oliver and Franke, Anna Katharina and Schumacher, Uta and Ostrowski, Andreas and Hostert, Patrick and Erasmi, Stefan, Mowing Detection Intercomparison Exercise (MODCiX) – A Cross-European Evaluation of Grassland Mowing Detection Algorithms. Available at SSRN: https://ssrn.com/abstract=5480735 or http://dx.doi.org/10.2139/ssrn.5480735