COS 34-3
Mapping lianas in tropical forests using high-resolution imaging spectroscopy
Increasing size and abundance of lianas relative to trees are among the pervasive changes in neotropical forests, and may lead to reduced forest carbon stocks. Yet the liana growth form is chronically understudied in forest censuses, resulting in few data on the scale, cause, and impact of increasing lianas. Satellite and airborne remote sensing provide a potential way to map and monitor lianas at large spatial and rapid temporal scales, compared with plot-based forest censuses. Recent advances in imaging spectroscopy and classification algorithms present new opportunities to discriminate liana cover at the tree canopy scale. We combined high-resolution airborne imaging spectroscopy (from the Carnegie Airborne Observatory) and a ground-based tree canopy census in central Panama to investigate whether tree canopies supporting lianas could be discriminated from tree canopies with no lianas. We performed supervised binary classifications using support vector machine algorithms, testing trees supporting five different liana canopy cover threshold values (severe: >80%, heavy: >60%, moderate: >40%, mild: >20%, thin: ≥1%) against trees with 0% liana cover.
Results/Conclusions
We achieved testing accuracies of greater than 90% in discriminating trees with severe liana canopy cover from trees with 0% liana cover. Accuracy remained high (>73%) for the heavy and moderate liana cover threshold values, but fell below 67% for the mild and thin threshold values. When the severe SVM model was applied to the full image of the study site, we found a 2.3% false-positive error rate when validated against an independent plot-level dataset of liana canopy cover. Using our landscape-scale liana cover classification map, we show that 11.9%-18.0% of the 585 ha study site has severe liana canopy cover. Given the relative increase in lianas and their negative effect on tree growth and mortality, our finding of severe liana cover across a large fraction of the landscape has broad implications for ecosystem function and forest carbon storage.