Using Explainable Artificial Intelligence (XAI) to Improve LANDFIRE Existing Vegetation Type (EVT) Product

Abstract:

Abstract:

USGS LANDFIRE’s Existing Vegetation Type (EVT) represents the current distribution of the terrestrial ecological systems classification. A terrestrial ecological system is defined as a group of plant community types that tend to co-occur within landscapes with similar ecological processes, substrates, and/or environmental gradients. The LANDFIRE team uses decision tree models, field measurements of vegetation, Landsat imagery, elevation, and biophysical gradient data to map EVT with each lifeform – tree, shrub, and herbaceous.

Current machine learning applications and algorithms have developed promise to produce autonomous systems that automatically perceive, learn, predict and act on their own. However, the effectiveness of these systems is limited by the machine’s current inability to explain their decisions, algorithmic paths and actions to human users. The purpose of this presentation is to share results of applying XAI to EVT black-box models and demonstrate the tools developed to assist scientists/analysts understand and trust prediction outcomes of vegetation type which to streamlines development of the LANDFIRE EVT product.

Keywords: Machine Learning, Black-Box Model, Explainable Artificial Intelligence, XAI, Global Interpretability, Local Interpretability, United States Geological Survey, USGS, LANDFIRE Existing Vegetation Type