Abstract Eastern redcedar (Juniperus virginiana L.) encroachment on the Great Plains has led to decreases in biodiversity, water availability, and grazing land while increasing the risk of catastrophic wildfires. Quantifying the current spatial distribution of eastern redcedar can aid land management efforts to combat encroachment, and extending this information over time can improve our understanding of the patterns and drivers of encroachment. We compared several remote sensing methods for mapping percent conifer cover to develop an approach that would be applicable for monitoring eastern redcedar encroachment across the Great Plains to support management goals. Leaf-off lidar was filtered through a novel approach using normalized return intensity and local canopy density to remove residual points pertaining to deciduous trees, which enabled us to calculate percent conifer cover. A sample of the conifer cover derived from leaf-off lidar was then used to test passive imagery-based methods that could be applied over a greater spatiotemporal extent. These imagery-based methods included Spatial Wavelet Analysis applied to very high-resolution imagery and random forest regression modeling with Landsat 8 and Sentinel-2 based predictive layers generated in Google Earth Engine from a single image, seasonal composites, and harmonic regression coefficients produced from an annual time series. Spatial Wavelet Analysis provided high accuracy (~5% RMSE) in areas where conifer cover was less than 10%, but accuracy rapidly decreased with increases in observed cover such that this method had a very low overall accuracy (42.5% RMSE). Landsat 8 and Sentinel-2 based predictors yielded similar performance to each other in models of conifer cover (9.7 to 13.3% RMSE), but seasonal composites from either sensor provided higher predictive power than either use of a single winter image or harmonic regression coefficients. According to recent estimates from the Forest Inventory and Analysis Program, eastern redcedar comprises more than 90% of conifer basal area in 277 counties of the central Great Plains, and thus mapping conifer cover can be assumed to reflect eastern redcedar cover. By applying the LandTrendr algorithm to Landsat seasonal composites we produced stable estimates of eastern redcedar cover from 1984 to 2018 and quantified encroachment as a 2.3% per year increase in eastern redcedar forest, defined as areas with ≥10% redcedar cover. The comparison of these methods and their application through time lays the groundwork for monitoring of redcedar encroachment across the central Great Plains and provides an approach for mapping fractional cover of conifer trees more generally.

中文翻译：

---

监控针叶树覆盖：内布拉斯加州中部东部红杉入侵的叶落激光雷达和基于图像的跟踪

摘要 东部红杉 (Juniperus virginiana L.) 对大平原的侵占导致生物多样性、水资源可用性和牧场减少，同时增加了发生灾难性野火的风险。量化东部红杉当前的空间分布可以帮助土地管理努力打击侵占，随着时间的推移扩展这些信息可以提高我们对侵占模式和驱动因素的理解。我们比较了几种用于绘制针叶树覆盖率地图的遥感方法，以开发一种适用于监测大平原东部红杉入侵以支持管理目标的方法。通过使用归一化返回强度和局部冠层密度的新方法过滤掉叶激光雷达，以去除与落叶树有关的残留点，这使我们能够计算针叶树覆盖率。然后使用从叶状激光雷达获得的针叶树覆盖样本来测试基于被动图像的方法，这些方法可以应用于更大的时空范围。这些基于影像的方法包括应用于超高分辨率影像的空间小波分析和使用基于 Landsat 8 和 Sentinel-2 的预测层的随机森林回归建模，这些预测层在 Google 地球引擎中从单个影像、季节性合成和谐波回归系数产生年度时间序列。空间小波分析在针叶树覆盖率低于 10% 的区域提供了高精度 (~5% RMSE)，但随着观测覆盖率的增加，精度迅速下降，因此该方法的整体精度非常低 (42.5% RMSE)。基于 Landsat 8 和 Sentinel-2 的预测器在针叶树覆盖模型中产生了彼此相似的性能（9.7% 到 13.3% RMSE），但来自任一传感器的季节性合成提供比使用单个冬季图像或谐波回归系数更高的预测能力。根据森林清查和分析计划最近的估计，东部红杉占大平原中部 277 个县超过 90% 的针叶树基底面积，因此可以假设绘制针叶树覆盖率可以反映东部红杉覆盖率。通过将 LandTrendr 算法应用于 Landsat 季节性复合数据，我们对 1984 年至 2018 年东部红杉覆盖率进行了稳定估计，并将侵占量化为东部红杉林每年增加 2.3%，定义为红杉覆盖率≥10% 的区域。