ABSTRACT Amazonian tropical rainforest is being converted to other land cover types including crops and pasture. In deforested areas, secondary forest grows after pastures are abandoned, and ‘dirty pasture’ that has trees and shrubs but is actively used for grazing are also regionally important land cover types following forest conversion. This study describes a multistage process land cover classification method to map primary forest, secondary forest, pasture, pasture with trees, built and water in the Brazilian state of Rondônia. A recently developed Spectral Variability Vegetation Index (SVVI) is tested to discriminate land cover types with differing tree cover amounts. Random Forest classifier (RF) is applied to inputs from a) spectral mixture analysis (SMA), and b) tasselled cap (TC) transformation, both with and without SSVI as an additional input feature. SVVI improved the classification accuracy from 73% (TC) to 85% (TC-SVVI), and TC-SVVI yielded a land cover map with higher accuracy than that from SMA-SVVI (82%). Pasture-with-trees, secondary forest and primary forest were all distinguishable with the SVVI. Pasture-with-trees accounted for 67% of all pastures, demonstrating its importance for regional land cover. This land cover classification workflow with the SVVI index improves the accuracy of mapping heterogeneous tropical land cover types.

中文翻译：

---

使用光谱植被变异指数、波段变换和随机森林分类绘制巴西亚马逊地区的异质森林-牧场镶嵌图

摘要亚马逊热带雨林正在转变为其他土地覆盖类型，包括作物和牧场。在森林砍伐地区，牧场被废弃后会产生次生林，而有乔木和灌木但积极用于放牧的“脏牧场”也是森林转变后区域重要的土地覆盖类型。本研究描述了一种多阶段过程土地覆盖分类方法，用于绘制巴西朗多尼亚州的原生林、次生林、牧场、有树木的牧场、建筑和水域图。测试最近开发的光谱变异性植被指数 (SVVI) 以区分具有不同树木覆盖量的土地覆盖类型。随机森林分类器 (RF) 应用于来自 a) 光谱混合分析 (SMA) 和 b) 流苏帽 (TC) 变换的输入，有和没有 SSVI 作为附加输入功能。SVVI 将分类精度从 73% (TC) 提高到 85% (TC-SVVI)，TC-SVVI 生成的土地覆盖图精度高于 SMA-SVVI (82%)。有树木的牧场、次生林和原始林都可以通过 SVVI 区分。有树木的牧场占所有牧场的 67%，表明其对区域土地覆盖的重要性。这种带有 SVVI 索引的土地覆盖分类工作流程提高了绘制异质热带土地覆盖类型的准确性。有树木的牧场占所有牧场的 67%，表明其对区域土地覆盖的重要性。这种带有 SVVI 索引的土地覆盖分类工作流程提高了绘制异质热带土地覆盖类型的准确性。有树木的牧场占所有牧场的 67%，表明其对区域土地覆盖的重要性。这种带有 SVVI 索引的土地覆盖分类工作流程提高了绘制异质热带土地覆盖类型的准确性。