Browsing by Subject "random forest classifier"

Bamboos are very relevant for ecosystem dynamics in tropical and subtropical forests worldwide. Madagascar holds 35 endemic species of woody bamboo, mainly found in forested areas. Six species of bamboo lemurs endemic to the island are highly specialized to exploit such bamboo species. Bamboo lemurs are highly threatened with extinction, and recent translocations of individuals to protected areas have failed, resulting in the death or emigration of all translocated individuals. The distribution, dynamics, and conservation status of the bamboo species they depend on is poorly known, while rising deforestation rates and direct exploitation of bamboo present important threats. Remote-sensing techniques, used often to map vegetation classes over large scales can potentially be used to map the extent of bamboo in a forested area, given the distinct structural and phenological characteristics of these grasses. In this study, I integrated ecological and remote sensing approaches to understand the diversity of bamboo species in the Ranomafana region of Madagascar, and to determine whether their distribution can be mapped through remoted sensing, if their occurrence is limited to particular habitat types, whether the national park presents suitable patches of bamboo for bamboo lemurs, and whether vegetation changes over the last 15 years have diminished the potential availability of bamboo in the area. I conducted an expedition to Ranomafana National Park (RNP), in south-eastern Madagascar, in November 2023, and collected 123 vegetation plots, including 45 plots revisited after an expedition in 2008. I used this field data to train a random forest classifier conducting supervised classifications based on Sentinel-2 imagery. I used these classifications to map the extent of different vegetation types and two genera of bamboo in RNP and its surroundings and evaluated their spatial and temporal distribution. The classification model accurately predicted the distribution of genus Cathariostachys, an endemic giant bamboo which is the main food source for bamboo lemurs in Ranomafana. Plots with Cathariostachys grew in the highest densities of all bamboo assessed in the field plots, and its distribution was projected over secondary and degraded forest areas, including areas outside RNP. Instead, the distribution of the endemic climbing bamboo of genus Sokinochloa, a secondary food source for bamboo lemurs, was captured with lower accuracy, mostly in the forested areas inside RNP, where it grew in low densities in the field plots. Comparing forest extent based on remote sensed classifications from 2008 and 2023, I show that a big proportion of forest has been lost in unprotected areas. RNP provided protection against forest loss, but the promotion of primary forest conservation and the disturbances through illegal activities are likely not providing habitat to secondary forest dwelling Cathariostachys bamboo. Small, sparse Sokinochloa patches within the continuous forest can assist bamboo lemur connectivity when moving through clustered patches of Cathariostachys. Nonetheless, because bamboo lemurs are affected by poaching in unprotected areas, the availability of good-quality Cathariostachys bamboo stands inside RNP is crucial for their conservation. However, no evidence of sexual reproduction or establishment of new patches of Cathariostachys was found in the field. Therefore, habitat restoration measures focused on securing new growth of this bamboo genus inside RNP could have a great impact on the conservation of bamboo lemurs. The success of such conservation measures could be assessed remotely with the mapping model developed in this study.