Eastern Texas Ecological Conservation: Mapping Change in Longleaf Pine Distribution Across Eastern Texas to Inform Conservation and Restoration Efforts

Once widespread in the southeastern U.S., longleaf pine (Pinus palustris) and its ecosystem declined to only 3% of its original extent due to logging, fire suppression, and encroachment from industrial pine plantations. The latter are typically dominated by loblolly pine (Pinus taeda) and frequently harvested and replanted. Distinguishing longleaf from loblolly pine is challenging; they exhibit similar structure, growth patterns and spectral signatures and share similar habitat. Current longleaf conservation action heavily relies on manual site survey. NASA DEVELOP collaborated with the Texas Longleaf Team, Texan by Nature, and Texas A&M Forest Service to map the current and historical distribution of longleaf in eastern Texas using NASA Earth observations and machine learning. We trained a Random Forest classification model on Landsat 9 Operational Land Imager (OLI)-2 reflectance data and derived spectral indices, and applied the model to historical imagery from Landsat 8 OLI and Landsat 5 Thematic Mapper (TM) dating back to 1985. Compared to reference data, the model achieved an overall agreement of 90% with a kappa index of agreement (k) of 0.79. Consideration of current and historical models suggests a 21.5% average increase in longleaf pine per decade. While these results may depict actual conservation outcomes, the ability to apply a classifier across the study area was hindered by longleaf pine stands occurring among similar pine tree species and seral stages. The results may be improved by augmenting the training data with more sites spread across the study area. Despite training data and map accuracy uncertainties, our study demonstrated the feasibility of using NASA Earth observations with machine learning to locate potential longleaf pine stands for informing conservation and restoration efforts.