关键词: 叶片氮浓度, 叶片水分影响, 光谱变换技术, 高光谱分辨率, 湿地植被

Abstract:

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.

Key words: foliar nitrogen concentration, foliar water effect, spectral transformation techniques, hyperspectral, wetland vegetation