Predicting the co-invasion of two Asteraceae plant genera in post-mining landscapes using satellite remote sensing and airborne LiDAR.

The Asteraceae plant family includes the most widespread weedy invaders in Europe, which may jointly inhibit natural succession in degraded land under restoration. The complex local drivers of co-invasions hinder remote sensing (RS) monitoring efforts, as the links between the ecological and the spectral habitat properties are largely unknown. We proposed a comprehensive framework for machine learning modeling of the co-invasion of two Erigeron spp. and two Solidago spp. in post-mining landscapes of S Poland, using both field data and a combination of Sentinel-2, Landsat 7 and airborne LiDAR RS predictors. Stochastic Gradient Boosting best captured the non-linear dependencies (Accuracy = 0.670-0.886, AUC = 0.675-0.923), and generally outcompeted two other classifiers (Random Forest and Support Vector Machines with a Radial Basis Function Kernel). The field-based functional diversity metrics were the strongest predictors, corroborating improved resistance to invasions by native plant functional richness. In terms of RS data, the most favorable conditions for co-invasion were identified by a narrow range of reflectance in the red-edge interval of a Sentinel-2 image, and constrained by LiDAR-derived vegetation height (for Erigeron spp.) and by high land surface temperatures (for Solidago spp.). The highest share of patches suitable for co-invasion was consistently found in the low vegetation land cover class, between 36% and 64% cover. We therefore advise considering particular management actions, such as increasing the supply of native seed, thus improving local community resistance to invasions. The proposed methods and openly available RS predictors may facilitate targeted monitoring and cost-effective management interventions.