Site‐specific conditions, climate, and management decisions all dictate the establishment and composition of desired plant communities within grassland restorations. The uncertainty, complexity, and large size of grassland restorations necessitate monitoring plant communities across spatial and temporal scales. Remote sensing with unmanned aerial vehicles (UAVs) may provide a tool to monitor restored plant communities at various scales, but many potential applications are still unknown. In a tallgrass prairie restoration located in Franklin Grove, IL, we used UAV‐based multispectral imagery to assess the ability of spectral indices to predict ecological characteristics (plant community, plant traits, soil properties) in the summer of 2017. Using 19 sites, we calculated the moments of 26 vegetation indices and four spectral bands (green, red, red edge, near infrared). Models based on each moment and a model with all moments were estimated using ridge regression with model training based on a subset of 15 sites. Each tested for significant error reduction against a null model. We predicted mean graminoid cover, mean dead aboveground biomass, mean dry mass, and mean soil K with significant reductions in cross‐validated root mean square error. Averaged coefficients determined from cross‐validation of ridge regression models were used to develop a final predictive model of the four successfully predicted ecological characteristics. Graminoid cover and soil potassium were successfully predicted in one of the sites while the other two were not successfully predicted in any site. This study provides a path toward a new level of ease and precision in monitoring community dynamics of restored grasslands.

中文翻译：

---

使用无人飞行器监测草丛草原的生态特征

特定地点的条件，气候和管理决策都决定了草原恢复区内所需植物群落的建立和组成。草地恢复的不确定性，复杂性和大面积性使得有必要跨时空尺度监测植物群落。无人飞行器（UAV）的遥感技术可以提供一种工具来监测各种规模的恢复的植物群落，但是许多潜在的应用仍是未知的。在伊利诺伊州富兰克林格罗夫的高草草原修复区，我们使用基于无人机的多光谱图像评估了光谱指标预测2017年夏季生态特征（植物群落，植物性状，土壤特性）的能力。使用19个站点，我们计算了26个植被指数和4个光谱带（绿色，红色，红色边缘，近红外）。在基于15个站点的子集的模型训练中，使用岭回归对基于每个时刻的模型和所有时刻的模型进行了估计。每个测试都针对空模型显着减少了错误。我们通过交叉验证的均方根误差预测了平均的类动物覆盖度，平均的死于地上生物量，平均的干重和平均的土壤K。根据岭回归模型的交叉验证确定的平均系数被用于开发四个成功预测的生态特征的最终预测模型。在其中一个地点成功预测了类固醇覆盖和土壤钾，而在任何地点均未成功预测出另外两个地点。