中文摘要：

以青海省玛沁县和贵南县高寒草甸作为典型研究区，利用地物光谱仪采集了20块样地的高光谱数据，并测定了对应样地所有样方中牧草的养分含量，分析了牧草中氮磷钾素含量与冠层原始光谱反射率和一阶微分光谱反射率之间的相关关系；采用回归统计方法，基于光谱位置变量、光谱面积变量及植被指数变量构建了高寒草甸氮磷钾素的估测模型，并对模型进行了精度评价。结果表明，1）与原始光谱反射率曲线相比，一阶微分光谱反射率曲线能较好地反映牧草中N、P、K素所对应的敏感波段；2）高寒草甸牧草中N、P、K素含量与冠层高光谱相关性较强的波段大多分布在红光区域（680-760nm）；3）基于光谱位置变量构建的估测模型能更好地反演高寒草甸N、P、K素含量。其中，以光谱位置变量R′708.88为自变量的对数模型对氮素含量估测效果较好，R^2为0．67，估测精度达到83．56％；以光谱位置变量R′704.85为自变量的对数模型对磷素含量估测效果较好，R^2为0．55，估测精度达到92．15％；以光谱位置变量R′697.36为自变量的对数模型对钾素含量估测效果较好，R^2为0．86，估测精度达到82．44％。

英文摘要：

Using typical alpine meadows in Maqin and Guinan Counties in Qinghai province as the research area, hyperspectral data from 20 plots were collected by spectrometer,while the forage nutrient concentrations were measured in the laboratory for each plot.Using regression analysis,the correlations between the nitrogen, phosphorus and potassium contents of the alpine meadow forage and the original reflectance data and the first-order differential of reflectance were analyzed.Inversion models were established for estimating the nitrogen, phosphorus and potassium content of alpine meadow forage based on spectrum location,spectrum area and veg-etation index,and the accuracy of the models was also evaluated.It was found that first-order differential re-flectance curve better predicts nitrogen,phosphorus and potassium content in the forage,than does the original data.In the red band (680-760 nm),the nitrogen,phosphorus and potassium levels in forage show a strong relationship with canopy hyperspectral reflectance curve parameters.The model which included spectrum loca-tion worked well for estimating concentration of nitrogen,phosphorus,and potassium in alpine meadow for-age.A logarithmic model for spectrum location (R′708.88 )can estimated forage nitrogen content with an R 2 of 0.67,and an accuracy of 83.56%,while a logarithmic model for spectrum location (R′704.85 )can estimated for-age phosphorus content,with an R 2 of 0.55,and an accuracy of 92.15%,and a logarithmic model for spectrum location (R′697 .36 )estimated potassium content,with an R 2 of 0.86,and an accuracy of 82.44%.