Invasive Annual Grass Mapping: Leveraging Land Surface Phenology and Bioclimatic Relationships

Abstract:

Many nonnative plant species have been introduced to the semi-arid landscape of the United States’ interior Pacific Northwest. However, some of the greatest ecological and economic impacts have resulted from exotic annual grass introduction. A recently introduced species, ventenata (Ventenata dubia), has been observed invading portions of this region that have typically been resistant to invasion by other exotic grass species. Land managers need spatially explicit information about populations of ventenata to address the effects of the invasion and its potential to change the landscape, including alterations to fire behavior. Remote sensing data have been successfully used to locate invasive plants by leveraging their unique phenological characteristics. Many of these studies employed MODIS or Landsat data to characterize annual or long-term phenology. MODIS has been used for mapping invasive species like cheatgrass because of its high temporal resolution allowing for detailed phenological characterization. However, the spatial resolution of this sensor reduces its ability to detect smaller populations in heterogenous landscapes. While Landsat has a higher spatial resolution, the temporal resolution is typically inadequate for capturing annual phenological patterns. We used spatio-temporal data fusion to generate a dense time series of land surface reflectance to characterize annual phenology at 30m resolution. We leveraged 30m phenology in conjunction with biophysical variables to model the presence of ventenata populations using a random forests model. The phenological characteristics of ventenata helped to differentiate it from surrounding native species while climatic conditions helped to separate ventenata from other common invasive species present in the region. Results showed that ventenata exhibited a unique phenological pattern when compared to other land cover types in the region and incorporating phenological information resulted in improved model performance. Mapped predictions will assist land managers in planning fire or weed management activities.