• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Asian conference on remote sensing;ACRS >Machine learning approaches to estimate chlorophyll-a concentration using GOCI satellite data
【24h】

Machine learning approaches to estimate chlorophyll-a concentration using GOCI satellite data

机译：使用GOCI卫星数据估算叶绿素a浓度的机器学习方法

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Water quality has been an important issue in Korea since most industrial facilities and residential areas located in the coastal region and monitoring the coastal water quality has been considered important. In this study, we attempted to estimate the concentration of chlorophyll-a (chl-α) over the southern coast of the Korean peninsula using Geostationary Ocean Color Imager (GOCI) satellite data. Recently, the ocean color remote sensing technologies have been widely used in the field of water quality monitoring due to its continuous spatial distribution through a wide area. Although Korea Ocean Satellite Center (KOSC) provide some algorithms to retrieve water quality indicators such as chl-a, total suspended solids (TSS), and colored dissolved organic matter (CDOM), the products show low consistency with in situ data. To improve the accuracy of estimating chl-a concentration, the machine learning method of applied in this study. We used GOCI remote sensing reflectance (Rrs) data processed by the GOCI Data Processing System (GDPS v2.0.0) of 8 spectral bands and their ratio as the input variables of the machine learning algorithm. A total of 36 variables were initially used, and we applied the Boruta algorithm as the feature selection method to decrease the dimension of the input variables. The variables which confirmed through the feature selection were used as the final variables. In situ chl-α data was collected from the FerryBox program, which is the automatic water quality monitoring systems on ships provided by Korea Marine Enviromnent Management Corporation (KOEM). The estimated chl-a concentration from GOCI data was compared with the in situ data from 2013 to 2016. Four machine learning approaches including Random Forest (RF), Extreme Gradient Boost (XGB), Gradient Boosting Machine (GBM), and Artificial Neural Network (ANN) were attempted for chl-a estimation and the results show that RF outperformed the other three models. The coefficient of detennination (R~2) and root-mean-square-error (RMSE) between the estimated and in situ chl-a was about 0.93 and 0.4572 μg/L for train dataset, 0.47 and 0.9119 μg/L for test dataset, respectively. It seems to be a quite meaningful result for estimating chl-α concentration compare to the performance (R~2 = 0.23) of the OC3G algorithm provided from KOSC for the same test dataset.
机译：自从大多数工业设施和居住区位于沿海地区以来，水质一直是韩国的重要问题，对沿海水质的监控也被认为是重要的。在这项研究中，我们尝试使用地球静止海洋彩色成像仪（GOCI）卫星数据估算朝鲜半岛南部海岸上的叶绿素a（chl-α）浓度。近来，海洋色彩遥感技术由于其在广阔区域中的连续空间分布而已广泛用于水质监测领域。尽管韩国海洋卫星中心（KOSC）提供了一些算法来检索chl-a，总悬浮固体（TSS）和有色溶解有机物（CDOM）等水质指标，但这些产品与原位数据的一致性较低。为了提高估计chl-a浓度的准确性，在本研究中应用了机器学习方法。我们使用由GOCI数据处理系统（GDPS v2.0.0）处理的8个光谱带的GOCI遥感反射率（Rrs）数据及其比率作为机器学习算法的输入变量。最初总共使用了36个变量，我们将Boruta算法用作特征选择方法以减小输入变量的维数。通过特征选择确认的变量用作最终变量。原位chl-α数据是从FerryBox程序中收集的，该程序是由韩国海洋环境管理公司（KOEM）提供的船上自动水质监测系统。将GOCI数据中估计的chl-a浓度与2013年至2016年的原位数据进行了比较。四种机器学习方法，包括随机森林（RF），极端梯度增强（XGB），梯度增强机（GBM）和人工神经网络（ANN）被尝试用于chl-a估计，结果表明RF优于其他三个模型。列车数据集的估计值与原位chl-a之间的确定系数（R〜2）和均方根误差（RMSE）分别为0.93和0.4572μg/ L，对于测试数据集为0.47和0.9119μg/ L ， 分别。与相同测试数据集的KOSC提供的OC3G算法的性能（R〜2 = 0.23）相比，估计chl-α浓度似乎是一个非常有意义的结果。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号