The extent of trees in the tropics

Abstract

Remote sensing-based assessments of tree cover and extent provide insight into global and local land use planning. However, there remains considerable uncertainty about the extent of drylands forests and trees in areas with non-forest land use. Here, we present a 10-meter tree extent map for 4.35 billion hectares in the tropics (-23.44° to 23.44° latitude) based on multitemporal composites of Sentinel-1 and Sentinel-2 analyzed with convolutional neural networks. We identify 2,719 million hectares of tropical land with ≥10% tree cover, including 680 million hectares in drylands. We find that 52% of tropical cropland and 46% of tropical urban areas had ≥10% tree cover in 2020. Compared to previous datasets, we demonstrate improved ability to map fragmented and open canopy forests at small spatial scales, especially in drylands, urban areas, and on cropland. As such, this data is expected to enable improved monitoring of tree dynamics at local scales.

Calculating Indicators From Global Geospatial Data Sets for Benchmarking and Tracking Change in the Urban Environment

“…This indicator uses 10 m resolution tree cover data for the year 2020 from the Trees in Mosaic Landscapes data set 6 (Brandt et al 2022), which gives tree extent within each 10 m pixel. It also applies the gridded population for 2020 at 100 m resolution from the WorldPop project as retrieved from Google Earth Engine.…”

“…This NDVI provides an estimate of the location of all types of vegetation within the area of interest. This is distinct from tree cover data (Brandt et al 2022) used in other indicators, which intentionally includes only trees. We then compare these areas to built-up areas in the city as defined by the built-up class from the 2020 ESA WorldCover.…”

“…The limitations of Brandt et al (2022) mean that certain cities may have missing tree cover data due to a lack of cloud-free imagery for 2020 or a lack of valid Sentinel-1 radar imagery, which can happen at the borders of satellite orbits and is common in coastal cities. Cloud removal algorithms used to generate tree cover data sets often have a high quantity of false negatives in cities (pixels that look like buildings but are, in fact, clouds) because clouds and buildings share bright reflectance values.…”

“…The persistence of cloudy images means that most tree cover data sets underestimate trees in cities because these false negative clouds hide trees in some of the pixels. Brandt et al (2022) employ a more aggressive approach to cloud removal in cities, resulting in fewer false negatives but less available data. In this data set, trees are defined as woody vegetation with a height of more than 5 m, or woody vegetation with a height of more than 3 m having a defined crown with at least a 5 m diameter.…”

“…In the tropics, the data set was found to have 97 percent user accuracy and 96 percent producer accuracy. Additional limitations and uncertainty related to the tree cover data set is described in the methods paper for Trees in Mosaic Landscapes (Brandt et al 2022).…”