LandTrendr smoothed spectral profiles enhance woody encroachment monitoring
2021
Secondary succession (SS) is one of the main consequences of the abandonment of agricultural and forestry
practices in rural areas, causing -among other processes- woody encroachment on former pastures and croplands.
In this study we model and monitor the spatial evolution of SS over semi-natural grassland communities in the
mountain range of the Pyrenees in Spain, during the last 36 years (1984-2019). Independent variables for
‘annual-based’ and ‘period-based’ modeling were drawn from a suite of Surface Reflectance Landsat images,
LandTrendr (LT)-algorithm-adjusted images and LT outputs. Support vector machine (SVM) classifiers were
trained and tested using all possible variable combinations of all the aforementioned datasets. The best modeling
strategy involved yearly time series of LT-adjusted Tasseled Cap Brightness (TCB) and Wetness (TCW) axes as
predictors, attaining a F1-score of 0.85, a Matthew Correlation Coefficient (MCC) of 0.67 and an AUC 0.83.
Woodlands encroached above 480,000 ha of grasslands and crops during the study period. A model using LT
outputs for the whole period also denoted good performance (F1-score = 0.85, MCC = 0.75) and estimated a
similar area of woodland expansion (~509,000 ha), but this ‘period’ approach was unable to provide temporal
information on the year or the encroachment dynamics. Our results suggest an overall proportion of 66% for the
Pyrenees being affected by SS, with higher intensity in the west-central part, decreasing towards the eastern end.
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
67
References
1
Citations
NaN
KQI