Global forest main disturbance types between 2000 and 2020

This dataset describes the main types of disturbances that forests worldwide have experienced over the past 20 years. There may be some forest patches that are subject to various types of disturbances, but this dataset only records the most significant types of disturbances. Mainly using machine learning classification methods, 18 forest disturbance characteristic indicators including global forest disturbance overall characteristics, pre disturbance characteristics, post disturbance characteristics, disturbance potential characteristics, surface cover characteristics, and spatial location are collected and calculated to determine 11 forest disturbance types: other disturbance (0), shifting cultivation (11), forestry replanting (12), plantation renewal (13), deforestation (14), forest fire (15), built-up expansion (18), cropland expansion (19), flood disturbance (20), oil palm cultivation (21), and artificial forest expansion (22). The Spatial resolution of this dataset is 30m. The main types of global forest disturbance are forestry replanting (26.52%), shifting cultivation (26.07%), and forest fires (12.58%), with newly forests accounting for 15.47% of the global forest disturbance area. The large-scale deforestation of natural forests also accounts for 7.94%. The disturbance caused by plantations, oil palm expansion, cropland expansion, and built-up expansion to forests also accounted for 4.63%, 1.54%, 3.88%, and 0.25%, respectively. The validation results based on 16000 sample points show that the accuracy of this dataset is as high as 95%.

本数据集涵盖了过去二十年间全球森林所经历的主要干扰类型。部分森林斑块可能受到多种类型的干扰，但本数据集仅记录其中影响最为显著的干扰类型。本数据集主要采用机器学习分类方法，采集并计算了包括全球森林干扰整体特征、干扰前特征、干扰后特征、干扰潜在特征、地表覆盖特征及空间位置在内的18项森林干扰特征指标，以此界定出11类森林干扰类型：其他干扰（0）、刀耕火种（11）、林业更新造林（12）、人工林更新（13）、森林砍伐（14）、森林火灾（15）、建成区扩张（18）、农田扩张（19）、洪涝干扰（20）、油棕种植（21）以及人工林扩张（22）。本数据集的空间分辨率为30米。全球森林干扰的主要类型包括林业更新造林（26.52%）、刀耕火种（26.07%）与森林火灾（12.58%），新增森林面积占全球森林干扰总面积的15.47%。天然林大规模砍伐占比也达7.94%。人工林、油棕种植扩张、农田扩张及建成区扩张对森林造成的干扰占比分别为4.63%、1.54%、3.88%及0.25%。基于16000个样本点的验证结果显示，本数据集的准确率高达95%。

搜集汇总

背景与挑战

背景概述

该数据集记录了2000至2020年间全球森林主要干扰类型，通过机器学习方法分类出11种干扰类型（如林业补植、轮耕、森林火灾等），空间分辨率为30米，验证精度高达95%。数据集揭示了林业补植(26.52%)、轮耕(26.07%)和森林火灾(12.58%)是主要的干扰类型。