基于减小叶片水分影响的湿地芦苇氮浓度高光谱反演研究

以盘锦双台河口湿地国家级自然保护区作为研究区,采用基于bootstrap的偏最小二乘回归模型（PLSR）,分别构建不同光谱变换技术（包括光谱水分影响减小技术WR、包络线去除CR、光谱一阶微分FD、光谱倒数的对数LR）和原光谱数据（R）的芦苇叶片氮浓度预测模型。使用变量投影重要性指标VIP,计算了各光谱波段在估算芦苇叶片氮浓度时的重要性。研究结果表明,WR光谱变换技术的芦苇叶片氮浓度估算精度最高（R²=0.87,均方根误差=0.57）,该方法可以有效减小叶片水分的影响,增强鲜叶片光谱中细微的氮吸收特征。

Estimating Nitrogen Concentrations in Wetland Reeds Based on Reducing Foliar Water Effect by Hyperspectral Data

The high-accuracy quantitative inversions of foliar nitrogen concentrations in wetland plants from hyperspectral data contribute to make the advanced comprehensions of wetland ecosystem functioning, biochemical processes and nitrogen circles. The study area in this paper was located in the Panjin Shuangtaihekou Wetland National Nature Reserve Administration. The wetland was mainly dominated by phragmites australis (Cav.) Trin. ex Steud with an area of 900 km². Foliar spectral reflectance data were measured using an ASD spectroradiometer (FieldSpec Pro FR) with a 25° field-of-view (FOV) and a spectral range of 350 to 2 500 nm. Subsequently, sampling reeds at each plot were cut at ground level and sent immediately to the laboratory. The nitrogen concentration of the vegetation was measured using a standard method of the Kjeldahl technique. Based on the different spectral transformation techniques and original spectral data (R), partial least-squares regressions (PLSR) integrating with the bootstrapping approaches were used to develop the prediction models of nitrogen concentrations in reedleaves. The set of spectral transformation techniques used in this study included the water removal (WR), continuum removal (CR), first derivative (FD), and reciprocal-logarithm transformation (LR). The variable importance of projection (VIP) was used as a metric to quantify the important degree of a spectral band for estimating foliar nitrogen concentrations with the various spectral transformation techniques. The results indicated that WR spectral transformation in combination with PLSR estimated foliar nitrogen concentration with the highest accuracy (R²=0.87,RMSE=0.57). The WR approachcan effectively reduce the water absorption effects on the subtle nitrogen absorption features across the fresh leaf spectrum. This study demonstrated the performance of the WR technique in increasing the accuracy of foliar nitrogen estimation for wetland plants. The estimation accuracies achieved from the applications of all spectral transformation techniques were higher compared to the original spectral data. The selected feature bands for estimating foliar nitrogen concentrations from the WR technique were all centered at the shortwave infrared (SWIR) spectral region, and forty percent (40%) of the selected bands from WR were associated with known nitrogen estimation bands. While, the selected feature bands from other spectral transformation techniques were located in both the SWIR and visible near-infrared (VNIR) spectral region. Phenology plays an important role in foliar nitrogen estimation using spectroscopy data. Therefore, there is a need for the future studies for understanding how phenology influences the performance of the WR technique in estimating foliar nitrogen concentrations in wetland plants.