Synergistic use of satellite, legacy, and in situ data to predict spatiotemporal patterns of the invasive Lantana camara in a savannah ecosystem

Lilly Theresa Schell^1*Emma Evers²Sarah Schönbrodt-Stitt¹Konstantin Müller¹Maximilian Merzdorf¹Drew Arthur Bantlin²Insa Otte¹

• ¹Department of Remote Sensing, Institute of Geography and Geology, University of Würzburg, Würzburg, Bavaria, Germany
• ²Conservation and Research Department, Akagera National Park, Kayonza, Rwanda

ModelingModeling species distributions is critical for managing invasive alien species, as reliable information on habitat suitability is essential for effective conservation and rehabilitation strategies. In thisThics study, we modeled models the suitable habitat and potential distribution of the notorious invader Lantana camara (LC) in the Akagera National Park (1,122 km²), a savannah ecosystem in Rwanda, a savannah ecosystem. Spatio-temporal patterns of Lantana camara LC from 2015 to 2023 were predicted at a 30-m spatial resolution using a presence-only species distribution model in Google Earth Engine, implementing a Random Forest classification algorithm and set up in the Google Earth Engine. The model incorporated Sentinel-1 SAR, Sentinel-2 multispectral data, anthropogenic socio-ecological parameterspredictors, and in situ presence data of Lantana camara. A maximum of 33% of the study area was predicted as a suitable Lantana camara LC habitat in 2023, with higher vulnerability in the central, northern, and southern Akagera National Park. The Cchange detection analysis revealed an increase in habitat suitability in the north-eastern sector and a decrease in the southwestern part of the park over the study period. . The model's predictive performance was robust, with AUCROC values ranging from 0.93 to 0.98 and AUCPR values ranging from 0.79 to 0.94. Key factors influencing Lantana camara LC habitat suitability in the study area included presence ofare the road network, the , elevation, and soil nitrogen levels. Additionally, the red edge, shortwave, and near-infrared spectral bands were identified as important essential predictors, highlighting the efficacy effectivenessefficacy of combining remote sensing and anthropogenic socio-ecological data with machine learning techniques to predict invasive species distributions. These findings offer provide valuable guidance for developing effective conservation strategies to protect savannah ecosystems and mitigate the spread of Lantana camara LC spread in the future.