Abstract Between 2013 and 2015, during a time of severe drought and elevated bark beetle (Dendroctonus spp.) activity in California, the amount of conifer mortality in the Southern Sierra Nevada increased greatly. Remote sensing is a critical means of providing up-to-date information on the location, magnitude, and extent of mortality across a broad geographic area. We used eleven Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) flight lines, resampled to 30 by 30 m pixel size and acquired on six separate dates as part of the HyspIRI Preparatory Campaign to simulate spaceborne imaging spectroscopy. We also spectrally degraded the AVIRIS images to simulate Landsat-8 data. We tested the ability of single-date and multi-temporal remote sensing to identify red stage conifer mortality, healthy conifer, and non-conifer dominated pixels using a random forest algorithm. Accuracy was assessed with an independent validation dataset acquired via WorldView imagery in areas spatially separate from where training data were collected, as well as through comparison with aerial detection survey and canopy water loss data, with generally good agreement. We found that classifications based on imaging spectroscopy significantly outperformed broadband multispectral feature sets (with a highest overall accuracy of 85.1% obtained by imaging spectroscopy and 80.2% obtained by the simulated multispectral images). We also found that classifications based on multi-temporal imaging spectroscopy were more accurate than single-date imaging spectroscopy (the highest overall accuracy obtained for single-date imaging spectroscopy was 83.4%). Imaging spectroscopy that included interseasonal data from the end of the drought outperformed all other datasets, including interannual data that included only images collected in the summer. Multi-date analysis also improved accuracy using broad band systems. Although current spaceborne assets are adequate for monitoring bark beetle mortality in a heterogeneous ecosystem, a spaceborne imaging spectrometer would further improve operational accuracy.

中文翻译：

---

使用高空间分辨率星载成像光谱检测整个生态区针叶树死亡率的框架

摘要 2013 年至 2015 年，在加利福尼亚严重干旱和树皮甲虫 (Dendroctonus spp.) 活动增加的时期，内华达山脉南部的针叶树死亡率大幅增加。遥感是提供有关广泛地理区域内死亡率的位置、数量和范围的最新信息的重要手段。我们使用了 11 条机载可见光/红外成像光谱仪 (AVIRIS) 飞行路线，重新采样到 30 x 30 m 像素大小，并在六个不同的日期采集，作为 HyspIRI 准备活动的一部分，以模拟星载成像光谱。我们还对 AVIRIS 图像进行了光谱降级以模拟 Landsat-8 数据。我们测试了单日期和多时相遥感识别红期针叶树死亡率、健康针叶树、和非针叶树主导的像素使用随机森林算法。准确性是通过 WorldView 图像在空间上与收集训练数据的区域分离的区域中获取的独立验证数据集进行评估的，并通过与航空探测调查和冠层失水数据进行比较，总体上具有良好的一致性。我们发现基于成像光谱的分类显着优于宽带多光谱特征集（成像光谱获得的最高总体准确率为 85.1%，模拟多光谱图像获得的总体准确率为 80.2%）。我们还发现基于多时相成像光谱的分类比单日期成像光谱更准确（单日期成像光谱获得的最高总体准确率为 83.4%）。包含干旱末期季节间数据的成像光谱优于所有其他数据集，包括仅包含夏季收集的图像的年际数据。多数据分析还使用宽带系统提高了准确性。尽管目前的星载资产足以监测异质生态系统中的树皮甲虫死亡率，但星载成像光谱仪将进一步提高操作精度。