Modelling susceptibility to Parthenium hysterophorus invasion in KwaZulu-Natal Province, South Africa using physical, climatic and remotely sensed derived variables.
Adeola, Arogoundade Mariama.
MetadataShow full item record
Invasive alien plants (IAP) are considered as one of the major causes of global change. Parthenium hysterophorus is recognized as one of the world’s most aggressive, harmful and extremely resilient invasive plant species. It has adverse impacts on the environment, economies, biodiversity, human health and agriculture. Identification and modelling of areas vulnerable to Parthenium invasion is critical for proactive control and site- specific management of its spread. This study sought to test the performance of Maxent algorithm in modelling habitats susceptible to Parthenium invasion using selected environmental and physical variables and remotely sensed data. Specifically, the study sought to identify key physical and bio-climatic variables that influence the distribution of Parthenium. Furthermore, the study sought to determine the value of the freely available Sentinel 2 multispectral instrument (MSI) datasets in concert with environmental variables in modelling habitat susceptible to Parthenium invasion. The Maximum Entropy model (MaxEnt) machine learning algorithm was used to model Parthenium invasion using presence - only records (n = 274). Results showed that landscapes characterized by low elevation, close proximity to roads and high precipitation were the most susceptible to Parthenium invasion. An Area under curve (AUC) value of 0.946 was attained, indicating that the model derived using the aforementioned optimal physical and bio-climatic variables performed better than random. Based on the high AUC values, results also showed that all the model scenarios derived from spectral bands and environmental variables, vegetation indices and environmental variables and a combination of spectral bands, vegetation indices and environmental variables performed better than random, with AUC values of 0.976, 0.970 and 0.974, respectively. The higher accuracy exhibited by the optimal model (bands and environmental variables) can be attributed to the integration of red edge band centered at 705 nm in Sentinel 2 MSI and environmental variables in predicting areas susceptible to Parthenium. Overall, these results demonstrate the potential of integrating the freely available Sentinel 2 MSI data and environmental variables to improve the mapping of habitat susceptibility to Parthenium invasion. These results could be beneficial for early detection, site -specific weed management and long-term monitoring.