This study's objective was to develop a method by which smallholder forest plantations can be mapped accurately in Andhra Pradesh, India, using multitemporal visible and near-infrared (VNIR) bands from the sentinel-2 multispectral instruments (MSIs). Conversion to cropland, coupled with secondary dependencies on and scarcity of wood products, has driven the deforestation and degradation of natural forests in Southeast Asia. Concomitantly, forest plantations have been established both within and outside of forests, with the latter (as contiguous blocks) and are the focus of this study. Accurately mapping smallholder forest plantations in South and Southeast Asia is difficult using remotely sensed data due to the plantations' small size (average of 2 hectares), short rotation ages (4–7 years for timber species), and spectral similarities to croplands and natural forests. Cloud-free Harmonized landsat sentinel-2 (HLS) S10 data were acquired over six dates, from different seasons, over four years (2015–2018). Available in situ data on forest plantations was supplemented with additional training data resulting in 2230 high-quality samples aggregated into three land cover classes: nonforest, natural forest, and forest plantations. Image classification used random forests on a thirty-band stack consisting of the VNIR bands and NDVI images for all six dates. The median classification accuracy from the 5-fold cross-validation was 94.3%. Our results, predicated on high-quality training data, demonstrate that (mostly smallholder) forest plantations can be separated from natural forests even using only the sentinel-2 VNIR bands when multitemporal data (across both years and seasons) are used.

中文翻译：

---

使用多时相协调 landsat sentinel-2 S10 数据绘制印度安得拉邦小农林场的地图

本研究的目标是开发一种方法，使用来自哨兵 2 多光谱仪器 (MSI) 的多时相可见光和近红外 (VNIR) 波段，可以准确地绘制印度安得拉邦的小农林场。转为农田，加上对木材产品的次要依赖和稀缺，导致东南亚天然林的森林砍伐和退化。与此同时，森林内外都建立了人工林，后者（作为连续块）是本研究的重点。由于人工林面积小（平均 2 公顷）、轮伐期短（木材品种为 4-7 年），因此难以使用遥感数据准确绘制南亚和东南亚的小农林种植园图。以及与农田和天然森林的光谱相似性。无云协调 landsat sentinel-2 (HLS) S10 数据是在四年（2015-2018 年）的六个日期、不同季节中获得的。现有的人工林原位数据补充了额外的训练数据，从而将 2230 个高质量样本汇总到三个土地覆盖类别中：非森林、天然林和人工林。图像分类在 30 波段堆栈上使用随机森林，该堆栈由所有六个日期的 VNIR 波段和 NDVI 图像组成。5 折交叉验证的中位数分类准确率为 94.3%。我们的结果基于高质量的训练数据，