Landslide-tsunamis are generated by masses, such as landslides or icebergs, impacting into water bodies. Such tsunamis resulted in major catastrophes in the recent past. Generic research into landslide-tsunamis has widely been conducted in idealised water body geometries at uniform water depths. However, varying bathymetries can significantly alter landslide-tsunamis. This article investigates this effect in a 2D flume using selected idealised bathymetries to provide methods to predict the transformed wave characteristics downwave of each feature. The selected bathymetries are: (a) linear beach bathymetries, (b) submerged positive and negative Gaussian bathymetric features and (c) submerged positive and negative step bathymetries. The hydrodynamic model SWASH, based on the non-hydrostatic non-linear shallow water equations, was used to simulate 9 idealised landslide-tsunamis (1 approximate linear, 2 Stokes, 2 cnoidal and 4 solitary waves), for a total of 184 tests. The analysed parameters include the free water surface, wave height and amplitude. Shoaling in (a) is represented by either Green's law or the Boussinesq's adiabatic approximation up to wave breaking with an accuracy of $-7\text{\%}$ to $+10\text{\%}$ for cnoidal and solitary waves, respectively. The results are then analysed with an (i) Artificial Neural Network and (ii) a regression analysis. (i) shows a smaller Mean Square Error (MSE) of 0.0027 than (ii) (MSE $=0.024$) and good generalisation in predicting the transformed wave characteristics and, after defining the best dimensionless parameters, (ii) provides empirical equations to predict transformed waves. In addition, simulations were conducted in a 3D basin to investigate the combined effect of the bathymetry and geometry. The efficient use of the developed prediction methods is demonstrated with the 2014 Lake Askja landslide-tsunami where a good accuracy is achieved compared to available numerical simulations.

滑坡海啸是由诸如滑坡或冰山之类的物质撞击水体产生的。这种海啸在最近造成了重大灾难。滑坡海啸的一般研究已经在均匀水深的理想水体几何结构中进行。但是，不同的水深会大大改变滑坡海啸。本文使用选定的理想水深来研究二维水槽中的这种影响，以提供预测每个特征下波变换波形特征的方法。所选的测深标准是：（a）线性海滩测深标准，（b）高斯正负测深特征和负高深测深特征，以及（c）正负负测深特征。基于非静水非线性浅水方程的流体动力学模型SWASH，用来模拟9个理想化的滑坡海啸（1个近似线性，2个斯托克斯，2个正弦波和4个孤波），总共进行了184次测试。分析的参数包括自由水面，波高和振幅。格林定律或Boussinesq的绝热近似表示（a）中的浅滩直至波浪破裂，其精度为$-7\text{％}$ 至 $+10\text{％}$分别用于正弦波和孤立波。然后使用（i）人工神经网络和（ii）回归分析对结果进行分析。（i）的均方误差（MSE）比（ii）（MSE）小，为0.0027$=0.024$）和预测变换波特性的良好通用性，并且在定义最佳无量纲参数后，（ii）提供经验方程式来预测变换波。此外，在3D盆地中进行了模拟，以研究测深和几何形状的综合影响。2014年的Askja湖滑坡-海啸证明了开发的预测方法的有效利用，与可用的数值模拟相比，该方法具有较高的精度。