基于机器学习和动力过程的南极冰架崩解特征分析
Analysis of the Features of Antarctic Ice Shelf Calving Based on Machine Learning and Dynamic Processes
龙清云 1张通 1王彻 2车涛 3效存德1
作者信息
- 1. 北京师范大学地表过程与资源生态国家重点实验室,北京 100875
- 2. 首都师范大学水资源安全北京实验室,北京 100048
- 3. 中国科学院西北生态环境资源研究院,甘肃 兰州 730000
- 折叠
摘要
冰架崩解对南极质量损失和动力过程有着直接影响,因此研究其变化的空间特征、环境条件和受控因子尤为重要.基于机器学习算法和冰盖动力模式,利用2005-2020年南极冰架崩解遥感数据、冰架表面裂隙数据、冰架支撑值、南极冰架损伤空间分布数据以及表面融化数据,结合机器学习二元分类算法,分析了 18种影响冰架动力过程的特征要素的重要性,并测算7种不同机器学习算法的准确性.结果表明,随机森林算法在冰架崩解事件的二元分类中具备最高准确率,其中,冰架表面流速和冰架表面融水对冰架崩解具有较高的影响,表明利用冰架自身动力性质和外部环境影响因子进行冰架崩解的预测具有一定的可行性.后续需进一步耦合动力模式和机器学习算法,并构建相应的数值模式体系,来刻画更高时空分辨率的冰架崩解事件强度和范围.
Abstract
Ice-shelf calving has a direct impact on Antarctic mass loss and dynamic processes,and it is particularly important to study its spatial characteristics,environmental conditions,and controlling factors.Based on the machine learning algorithms and ice sheet dynamic models,utilizing remote sensing data on Antarctic ice shelf calving from 2005 to 2020,ice shelf surface fracture data,ice shelf buttressing value,spatial distribution data of Antarctic ice shelf damage,and basal melting data,combined with machine learning binary classification,the importance of 18 characteristic elements influencing ice shelf dynamic processes was analyzed,and the accuracy of seven different machine learning algorithms was calculated.The results indicate that the random forest algorithm achieves the highest accuracy in the binary classification of ice shelf calving and that surface meltwater has a significant impact on ice shelf collapse,indicating the feasibility of using both the intrinsic dynamics of the ice shelf and external environmental factors for prediction.Subsequent efforts should further couple dynamic models with machine learning algorithms and establish corresponding numerical modeling systems to depict ice-shelf calving events with higher spatiotemporal resolutions in terms of intensity and extent.
关键词
南极冰盖/冰架崩解/冰架损伤/冰盖动力模式/机器学习Key words
Antarctic ice sheet/Ice shelf calving/Ice shelf damage/Ice sheet dynamic model/Machine learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据