From Smartphones to Satellites: Integrating Citizen Science and High-Resolution Environmental Data for Mapping Plant Invasions in Urban Mountain Landscapes

Predicting the distribution of invasive species in urban mountain ecosystems is important for biodiversity conservation and effective management. We developed a sequential two-stage species distribution modelling framework to characterize habitat suitability for the invasive vine Thunbergia alata in the Metropolitan District of Quito and to evaluate the contribution of citizen science records, systematic field surveys, and multi-source environmental data. In the first stage, 637 iNaturalist occurrence records retained after quality control were combined with 12 bioclimatic variables from WorldClim to develop a preliminary MaxEnt model. The resulting habitat suitability map was classified into five categories and used to design a stratified field survey across the predicted suitability gradient. We sampled 1,009 locations, obtaining 450 confirmed presences and 559 independent absence records. In the second stage, the field-derived occurrence data were used to develop a refined MaxEnt model incorporating the bioclimatic variables together with elevation, slope, land cover, VIIRS nighttime lights, NDVI, and the red spectral band. Environmental predictors were standardized to a common 5m. spatial resolution and screened for multicollinearity. Both models were evaluated using receiver-operating characteristic analyses and AUC, with 10 replicate runs and 25% of occurrence records withheld for testing. Mean AUC values were 0.851 ± 0.012 and 0.902 ± 0.008 for the preliminary and refined models, respectively. In the preliminary model, habitat suitability was primarily associated with broad-scale climatic variables, particularly precipitation seasonality and temperature-related variables. In the refined model, nighttime light intensity and elevation showed the highest contributions and permutation importance, while NDVI and climatic variables also contributed to model predictions. Independent field validation showed that 47.2% of absence observations occurred in the two lowest suitability classes, whereas 33.5% occurred in areas classified as highly or very highly suitable. The refined model therefore captured substantial spatial variation in habitat suitability while also revealing areas of mismatch between predicted suitability and field observations. These results demonstrate the value of sequentially integrating citizen science, systematic field observations, and fine-resolution environmental predictors to improve invasive species mapping. This framework provides a practical approach for supporting targeted monitoring and management of invasive plants in complex urban mountain ecosystems.