Examining changes in woody vegetation cover in a human-modified temperate savanna in Central Texas between 1996 and 2022 using remote sensing

Abstract

Savanna ecosystems across the globe have experienced substantial changes in their vegetation composition. These changes can be attributed to three main processes: (1) encroachment, which refers to the expansion of woody plants into open areas, (2) thicketization, which is characterized by the growth of sub-canopy woody plants, and (3) disturbance, defined here as the removal of woodland cover due to both natural forces and human activities. In this study, we utilized Landsat surface reflectance data and Sentinel-1 SAR data to track the progression of these process from 1996 to 2022 in the significantly modified Post Oak Savannah ecoregion of Central Texas. Our methodology employs an ensemble classification algorithm, which combines the results of multiple models, to develop a more precise predictive model, along with the spectral–temporal segmentation algorithm LandTrendr in Google Engine (GEE). Our ensemble classification algorithms demonstrated high overall accuracies of 94.3 and 96.5% for 1996 and 2022, respectively, while our LandTrendr vegetation map exhibited an overall accuracy of 80.4%. The findings of our study reveal that 9.7% of the overall area experienced encroachment of woody plants into open area, while an additional 6.8% of the overall area has transitioned into a thicketized state due to the growth of sub-canopy woody plants. Furthermore, 5.7% of the overall area encountered woodland disturbance leading to open areas. Our findings suggest that these processes advanced unevenly throughout the region, resulting in the coexistence of three prominent plant communities that appear to have long-term stability: a dense deciduous shrubland in the southern region, as well as a thicketized oak woodland and open area mosaic in the central and northern regions. The successional divergence observed in these plant communities attests to the substantial influence of human modification on the landscape. This study demonstrates the potential of integrating passive optical multispectral data and active SAR data to accurately map large-scale ecological processes.