Abstract Estimating soil organic carbon (SOC) in topsoil can help improve soil quality and food production. This study aimed to explore the potential of airborne hyperspectral image to estimate the SOC of bare topsoil at an agricultural site located in the southeast part of Iowa State, United States. To magnify the subtle spectral signals concerning SOC, and accelerate calibration and improve predictive ability, we developed a framework to combine two advanced spectral algorithms, namely, fractional-order derivative (FOD) and optimal band combination algorithm for SOC predicting. Our case was based on 49 soil samples and a scattered airborne hyperspectral image. Random forest (RF) was utilized to establish SOC estimation models by incorporating the optimal spectral indices processed by different FOD transformations on the basis of the optimal band combination algorithm. Results indicated that when the fractional order increased, overlapping peaks and baseline drifts were gradually removed. However, the magnitude of spectral strength decreased concurrently. More detailed and abundant spectral variability was captured by FOD as compared with those by original reflectance and first and second derivatives. The estimation accuracies developed from the optimal band combination algorithm (cross-validation R2, 0.36–0.66) were generally better than those from full-spectrum data (cross-validation R2, 0.32–0.54). The RF model based on the combination of 0.75-order reflectance and optimal band combination algorithm obtained the highest estimation accuracy for SOC with cross-validation R2 of 0.66. This research provides guidance for future studies in selecting the most appropriate FOD transformation to preprocess spectral data and in using the optimal band combination algorithm to determine the spectral index. Airborne hyperspectral image-based modeling can be further used to map agricultural topsoil SOC to support local-scale agricultural planning.

中文翻译：

---

探索机载高光谱图像估算表土有机碳的潜力：分数阶导数和最优波段组合算法的影响

摘要 估算表土中的土壤有机碳 (SOC) 有助于改善土壤质量和粮食产量。本研究旨在探索机载高光谱图像在估计位于美国爱荷华州东南部农业场地裸露表土 SOC 的潜力。为了放大与 SOC 相关的细微光谱信号，加速校准和提高预测能力，我们开发了一个框架来结合两种先进的光谱算法，即分数阶导数 (FOD) 和用于 SOC 预测的最佳波段组合算法。我们的案例基于 49 个土壤样本和一个分散的空中高光谱图像。在最优波段组合算法的基础上，通过结合不同FOD变换处理的最优光谱指数，利用随机森林(RF)建立SOC估计模型。结果表明，当分数阶数增加时，重叠峰和基线漂移逐渐消除。然而，频谱强度的幅度同时下降。与原始反射率和一阶和二阶导数相比，FOD 捕获了更详细和丰富的光谱变化。从最佳波段组合算法（交叉验证 R2，0.36-0.66）得出的估计精度通常优于全谱数据（交叉验证 R2，0.32-0.54）。基于0的组合的RF模型。75阶反射率和最优波段组合算法获得了最高的SOC估计精度，交叉验证R2为0.66。本研究为未来研究选择最合适的 FOD 转换来预处理光谱数据和使用最佳波段组合算法确定光谱指数提供了指导。基于机载高光谱图像的建模可进一步用于绘制农业表土 SOC 以支持当地规模的农业规划。