Assessment of Reforestation for the Subtropical Humid Forest in South-East Brazil; A comparison of datasets to assess discrepancies in area, evaluate agreement in overlap and analyse the accuracy for monitoring reforestation

Organised by Laboratory of Geo-information Science and Remote Sensing

Mon 13 June 2016 10:00 to 10:30

Room Gaia 2

By Froede Vrolijk (the Netherlands)

Tropical forests provide important ecosystem services, contain high biodiversity and sequester large amounts or carbon. These forests are declining globally, posing serious negative consequences for environmental sustainability. In recent years however, a trend towards forest regrowth is described in literature. In Brazil, large tracts of forest land have historically been converted into cattle pasture and agricultural land. These land change patterns were followed by land abandonment, which resulted in large areas of secondary forest growth. These forests grow rapidly and sequester large amounts of carbon. Planted forests also play an important role in Brazil, where tree plantations are expanding fast, with approximately 5000 km2 per year. Moreover, initiatives such as REDD+ aim to promote sustainable management of forests and enhance forest carbon stocks, in addition to reducing deforestation and forest degradation. Reforestation is now believed to be one of the most prominent solutions for combatting environmental issues. The aim of this study is to quantify annual forest regrowth according to existing datasets, and to asses their monitoring methods. Additionally, we assess the type of forest (natural or plantation) that is detected by each dataset. Our goal is to reveal the discrepancies between the datasets and validate the accuracy of each dataset for monitoring forest regrowth in south-east Brazil. The JRC land cover dataset, the FAO land use dataset and Hansen forest gain dataset were used to calculate the amount of regrowth in the study area. Additionally, a time series method was applied to a sample area of the study region. The regrowth area calculations show the highest amount of regrowth for the Hansen dataset (1304 km2), whereas JRC detects 295 km2, and the FAO dataset shows 28 km2 of change to forest land per year. The inter-comparison between the datasets reveals the agreement between datasets for assigning a regrowth label, using paired comparisons. The overlay of the JRC and Hansen datasets shows the highest percentage in overlap (54.25 %), followed by FAO – Hansen (20.25 %; using FAO as a reference) and JRC – FAO (0.54 %) when using the JRC dataset as a reference. The validation of the datasets using visual analysis in the entire study area was implemented for the JRC and FAO datasets, and shows an accuracy of 65.4 percent for the JRC dataset and 13.6 percent for the FAO dataset. Moreover, the validation of the Hansen dataset and time series method was implemented for the selected sampling site. The Hansen dataset shows an overall accuracy of 86 percent, and the time series approach shows an accuracy of 64 percent. The type of regrowth (natural, plantation or mixed) that is detected varies largely for each dataset. The JRC and Hansen datasets mostly detect plantation forests, covering 54 and 94 percent, respectively. The FAO dataset and time series method mostly detect natural forest, covering 45 percent and 76 percent, respectively. We conclude that there is a strong need for large-scale validation of datasets, in order to assess the uncertainty that accompanies each dataset. The validation of the datasets in this study shows a commission error only, and does not include the omission error. Further research is needed to quantify omission errors.

Keywords: Tropical secondary forest, regrowth, visual analysis, reforestation, dataset comparison, time series analysis, REDD+