Faculty of Navigation, Maritime University of Szczecin, Waly Chrobrego 1-2, 70-500 Szczecin, Poland.
Marine Technology Ltd., 81-521 Gdynia, Poland.
Sensors (Basel). 2023 Apr 13;23(8):3941. doi: 10.3390/s23083941.
Digital bottom models are commonly used in many fields of human activity, such as navigation, harbor and offshore technologies, or environmental studies. In many cases, they are the basis for further analysis. They are prepared based on bathymetric measurements, which in many cases have the form of large datasets. Therefore, various interpolation methods are used for calculating these models. In this paper, we present the analysis in which we compared selected methods for bottom surface modeling with a particular focus on geostatistical methods. The aim was to compare five variants of Kriging and three deterministic methods. The research was performed with real data acquired with the use of an autonomous surface vehicle. The collected bathymetric data were reduced (from about 5 million points to about 500 points) and analyzed. A ranking approach was proposed to perform a complex and comprehensive analysis integrating typically used error statistics-mean absolute error, standard deviation and root mean square error. This approach allowed the inclusion of various views on methods of assessment while integrating various metrics and factors. The results show that geostatistical methods perform very well. The best results were achieved with the modifications of classical Kriging methods, which are disjunctive Kriging and empirical Bayesian Kriging. For these two methods, good statistics were calculated compared to other methods (for example, the mean absolute error for disjunctive Kriging was 0.23 m, while for universal Kriging and simple Kriging, it was 0.26 m and 0.25 m, respectively). However, it is worth mentioning that interpolation based on radial basis function in some cases is comparable to Kriging in its performance. The proposed ranking approach was proven to be useful and can be utilized in the future for choosing and comparing DBMs, mostly in mapping and analyzing seabed changes, for example in dredging operations. The research will be used during the implementation of the new multidimensional and multitemporal coastal zone monitoring system using autonomous, unmanned floating platforms. The prototype of this system is at the design stage and is expected to be implemented.
数字海底模型广泛应用于人类活动的诸多领域，如航海、港口和近海技术或环境研究。在许多情况下，它们是进一步分析的基础。这些模型是基于水深测量数据构建的，这些数据在许多情况下是大型数据集的形式。因此，各种插值方法被用于计算这些模型。在本文中，我们进行了分析，比较了几种用于海底表面建模的方法，特别关注地质统计学方法。目的是比较克里金法的五个变体和三种确定性方法。该研究是使用自主水面船舶采集的真实数据进行的。所采集的水深数据经过简化（从约 500 万个点减少到约 500 个点）并进行了分析。提出了一种排名方法来进行复杂而全面的分析，该方法集成了通常使用的误差统计量，如平均绝对误差、标准差和均方根误差。这种方法允许纳入对评估方法的各种看法，同时集成各种指标和因素。结果表明地质统计学方法表现非常出色。最好的结果是使用经典克里金法的修正方法获得的，即不连续克里金法和经验贝叶斯克里金法。对于这两种方法，与其他方法相比，计算出了良好的统计数据（例如，不连续克里金法的平均绝对误差为 0.23 米，而通用克里金法和简单克里金法的平均绝对误差分别为 0.26 米和 0.25 米）。然而，值得一提的是，基于径向基函数的插值在某些情况下与克里金法的性能相当。所提出的排名方法被证明是有用的，并可用于未来选择和比较 DBM，主要用于海底变化的测绘和分析，例如在疏浚作业中。该研究将用于实施使用自主、无人浮动平台的新的多维和多时区沿海区监测系统。该系统的原型处于设计阶段，预计将投入实施。
