Abstract:
Invasive alien plant species are expanding globally due to climate change and human pressures, posing growing threats to biodiversity, ecosystem services and agricultural systems. One such species, Mesosphaerum suaveolens, is rapidly spreading across West Africa, where it disrupts rain-fed cropping systems including maize, a staple crop critical to food security in Burkina Faso. Despite its ecological and agronomic impacts, there's limited understanding of how climate change may affect its future distribution and potential interaction with maize cultivation zones. This study aimed to (i) determine the current distribution of M. suaveolens in Burkina Faso, (ii) forecast its future spread under different climate scenarios, and (iii) assess the spatial overlap between invasion risk and maize-growing areas. We compiled 3254 presence records of M. suaveolens and 51 environmental predictors. After reducing multicollinearity using a variance inflation factor threshold (VIF < 5), a subset of 9 uncorrelated predictors was retained. Three AI-based algorithms: Random Forest (RF), Support Vector Machine (SVM), and Convolutional Neural Network (CNN) were implemented in Python 3.10 to model current suitability. Models’ performance were evaluated using AUC and TSS, and only RF and CNN, which outperformed SVM, were used for future projections under eight combinations of general circulation models (MIROC6 and HadGEM3-GC31-LL), scenarios (SSP245 and SSP585), and time horizons (2050 and 2080). Maize suitability was modeled using RF, based on 547 presence records and 11 predictors (VIF-filtered). Current and future maize maps were intersected with M. suaveolens risk maps to assess spatial overlaps and quantify exposure levels. CNN projected a 42 to 46% decline in suitable habitats for M. suaveolens by 2080, while RF predicted minor changes (-2.6% to +0.8%). Currently, high-suitability maize areas cover 107,336 km² (39.77%), increasing to 133,362 km² (HadGEM3-GC31-LL) and 136,881 km² (MIROC6) by 2090 under SSP5-8.5. Currently, 62% of maize zones overlap with high invasion risk (CNN), decreasing to 14-16% by 2090, while RF estimates remain stable at ca. 3-4%. These results highlight the CNN model’s higher sensitivity to climate variability and the more conservative nature of RF projections. The combine use of SDMs and crop modeling in an AI framework offers a robust tool for anticipating invasive spread and informing climate-resilient agricultural planning. This study contributes to ecological forecasting and supports the achievement of SDGs 2, 13 and 15 through evidence-basedland-use strategies.
Description:
A Thesis submitted to the West African Science Service Center on Climate Change and Adapted Land Use and Université Joseph KI-ZERBO, Burkina Faso in partial fulfillment of the requirements for the Master of Science Degree in Informatics for Climate Change
