Dataset

Global Forest/Non-Forest maps

The global forest/non-forest maps (FNF) were generated by a Random Forest machine learning-based classification method, with the re-processed global 25m resolution PALSAR-2 mosaic dataset (Ver. 2) as input. Here, the "forest" is defined as the tree covered land with an area larger than 0.5 ha and a canopy cover of over 10 %, in accordance with the FAO definition of forest. The classification results are presented in four categories, with two categories of forest areas: forests with a canopy cover of 90 % or more and forests with a canopy cover of 10 % to 90 %, depending on the density of the forest area.

The previous version of the global forest/non-forest maps - currently the case for the PALSAR FNF maps - were published in three categories: forest, non-forest, and water. A threshold method was used to determine forest/non-forest for each year, which caused discontinuities when examining the same area over time, and changes could not be accurately captured.

Therefore, in the newly released (Ver 2) version, data for the past year is used in addition to the data for the target year to be created. For example, when creating the 2020 FNF, the global 25m resolution mosaics of 2019 and 2020 are stacked and used as input data for classification. When stacked, the target year mosaic is weighted 3 times more than the past year's mosaic. In this way, the stacked data can represent the forest change by retaining the effect of the target year on the areas where the forest has changed.

Accuracy validation of the classification results was conducted for the four-category case and for the three-category case where the two forest areas were integrated. The results indicate that the overall accuracy was at least 86 % for 3 categories and 76 % for 4 categories for four years from 2017 to 2020. Accuracy thus improved by 5-10 % or more compared to the previous version. Figure 1 shows global mosaic images and forest/non-forest maps, and Figure 2 shows an example of Borneo island. Please refer to the "Global 25m Resolution PALSAR-2 Forest/Non-forest Map Dataset Description" document for details on the classification method.
JAXA plans to update the FNF classification algorithm to improve its accuracy. Any comments or requests for the datasets are welcome.

For details, please see the following dataset descriptions;

Please register your information from the following URL to download the dataset. Your email address is required.