On-time and accurate estimation of the soil nitrogen mineralization rate (SNMR) is critical for nitrogen (N) management and protecting the environment. This study evaluated the performance of a visible-to-near-infrared reflectance (vis-NIR) spectroscopy for predicting SNMR for four texture groups. A total of 62 topsoil samples were collected from 17 management zones distributed over four fields and incubated with seven destructive sampling events. Samples were analysed for total mineral N (NH+NO) content and scanned using a vis-NIR sensor simultaneously at each of the seven-sampling times. Four partial least squares regression models were calibrated and validated for four textural groups (groups- 1– 4) identified over the United State Department of Agriculture (USDA) texture triangle. Prediction accuracies indicated that vis-NIR sensor was moderately to highly accurate for predicting SNMR, while observing variable accuracies across texture groups. The highest accuracy was obtained for group 1 (sandy-loam; coefficient of determination, R = 0.90; root mean square error, RMSE = 0.04 mg N kg soil day), successively followed by group 2 (mostly loam; R = 0.80, RMSE = 0.05 mg N kg soil day) group 4 (mostly silt; R = 0.66, RMSE = 0.08 mg N kg soil day), and group 3 (silt-loam; R = 0.44, RMSE = 0.08 mg N kg soil day). Variable importance in projection score revealed that the key spectral bands to predict SNMR were in 2150 – 2260 nm and 2470 – 2480 nm, resembling the key bands associated with soil organic compounds and clay minerals. In-advance texture information required for soil stratification is regarded a limitation of the proposed approach. In conclusion, vis-NIR holds potential for a rapid estimation of SNMR when samples are stratified into similar texture groups in advance, however, confirmatory research will be needed to validate the current findings for soils from different origin and under different management.

中文翻译：

---

通过结构均匀性标准预测样品分层后土壤氮矿化的可见光到近红外光谱


及时、准确地估算土壤氮矿化率 (SNMR) 对于氮 (N) 管理和保护环境至关重要。本研究评估了可见光到近红外反射 (vis-NIR) 光谱在预测四个纹理组的 SNMR 方面的性能。从分布在四个田地的 17 个管理区采集了总共 62 个表土样本，并进行了 7 个破坏性采样事件的培养。对样品进行总矿物氮 (NH+NO) 含量分析，并使用可见近红外传感器在七个采样时间的每个时间同时进行扫描。针对在美国农业部 (USDA) 纹理三角形上确定的四个纹理组（组 1-4），对四个偏最小二乘回归模型进行了校准和验证。预测精度表明，vis-NIR 传感器在预测 SNMR 方面具有中等至高度的准确度，同时观察不同纹理组的不同准确度。第 1 组（沙壤土；确定系数，R = 0.90；均方根误差，RMSE = 0.04 mg N kg 土壤日）获得最高准确度，其次是第 2 组（主要是壤土；R = 0.80，RMSE） = 0.05 mg N kg 土壤日）第 4 组（主要是淤泥；R = 0.66，RMSE = 0.08 mg N kg 土壤日）和第 3 组（淤泥壤土；R = 0.44，RMSE = 0.08 mg N kg 土壤日）。投影得分中的变量重要性表明，预测 SNMR 的关键光谱带位于 2150 – 2260 nm 和 2470 – 2480 nm，类似于与土壤有机化合物和粘土矿物相关的关键光谱带。土壤分层所需的预先纹理信息被认为是该方法的局限性。 总之，当样品提前分层到相似的质地组时，vis-NIR 具有快速估计 SNMR 的潜力，但是，需要验证性研究来验证不同来源和不同管理下的土壤的当前发现。