基于Landsat8数据的水体指数模型研究

Study on Water Index Construction Based on Landsat8 Data

摘要: 为寻找一种环境适应能力强、提取效果稳定的水体指数模型，本文根据可见光到近红外波段水体与常见噪声的光谱差异，构建水体指数GRN-WI。为验证模型的有效性，选取3幅不同环境下的landsat8数据中4个区域进行实验，实验采用K均值聚类法代替手动调节阈值提取水体，将实验结果与5种不同的水体指数进行对比分析。结果表明，GRN-WI能有效提取水体，总体精度优于96.21%，kappa系数优于92.35%，总体误差小于8.15%，总体优于其它指数；模型在含有大量山体阴影、大量薄云和冰雪的环境下较其它指数精度明显提高。

Abstract: In order to find a water index model with strong environmental adaptability and stable extraction effect, based on the spectral difference between water and common noise in visible to near-infrared bands, this paper constructs water index GRN-WI. In order to verify the validity of the model, four regions in three Landsat8 data under different environments were selected for experiments. K-means clustering method was used in the experiment to extract water instead of manually ad-justing the threshold value, and the experimental results were compared with five different water indexes. The results show that: GRN-WI can effectively extract water; the overall accuracy is better than 96.21%; the Kappa coefficient is better than 92.35%; the overall error is less than 8.15%; the overall is better than other indexes. The model is more accurate than other indexes in the envi-ronment with a lot of mountain shadow, a lot of thin cloud and ice and snow.