高分辨率且非接触式的水深监测对钙华湖泊景观的管理与保护至关重要。卫星遥感测深无法捕获钙华湖泊细微的的水下沉积特征。近年来，轻小型无人机遥感技术逐渐应用于浅水区超高分辨率的水深探测。然而，水深反演中经典的对数模型难以适应钙华湖泊内广泛存在的瑞利散射现象。因此，本文利用机器学习模型开展基于无人机影像的钙华湖泊水深反演研究。以中国四川九寨沟火花海为实验区，对基于随机森林（RF, Random Forest）、支持向量机（SVM, Support Vector Machine）与多层感知机（MLP, Multi-Layer Perceptron）的水深反演模型进行训练与验证，其均方根误差依次为0.816 m、0.945 m、0.832 m。实验结果表明，机器学习模型相较于传统的对数模型具有更高的水深反演精度。其中，随机森林模型与多层感知机模型比支持向量机模型更适合基于无人机影像的钙华湖泊水深反演。

High-resolution and non-contact water depth monitoring is crucial for the management and protection of tufa lake landscapes. Satellite-derived bathymetry cannot capture the subtle underwater sedimentary characteristics of tufa lakes. In recent years, the remote sensing technology of light and small unmanned aerial vehicles (UAV) has gradually been applied to ultra-high-resolution bathymetric mapping in shallow water areas. However, the classic logarithmic model in water depth inversion is difficult to adapt to the widespread Rayleigh scattering phenomenon in tufa lakes. Therefore, in this article machine learning methods are used to construct bathymetric inversion models of tufa lakes based on UAV imagery. Taking Spark Lake in Jiuzhaigou National Nature Reserve, Sichuan Province, China as the experimental area, aerial image data for bathymetric model construction were extracted from UAV platforms. Based on the pre- and post-earthquake UAV images, the pre-earthquake orthophoto with water and the post-earthquake surface model without water are generated by the Structure-from-Motion algorithm, respectively. After excluding anomalous areas, sample points for the bathymetric inversion were randomly selected. Each sample data has both the red, green, and blue band (RGB) digital number (DN) values of the pre-earthquake orthophoto and the relative depth values of the post-earthquake exposed terrain relative to the pre-earthquake water surface. Based on this dataset, machine learning regression models based on random forest (RF), support vector machine (SVM), and multi-layer perceptron (MLP) are constructed respectively. The above machine learning models are trained repeatedly to determine their respective optimal parameters. Finally, the accuracy of the estimated bathymetry was verified using the exposed lake terrain after the earthquake. The results indicate that the water depth distribution of the three models has small differences in shallow water areas, and the areas with significant differences are mainly deep-water regions. The bathymetric map simulated by the RF model is susceptible to high-frequency signals, while the bathymetric maps simulated by SVM and MLP models suffer from localized overestimation of water depth distribution. In terms of accuracy assessment, the RF, SVM, and MLP models have root mean square errors (RMSE) of 0.816 m, 0.945 m, and 0.832 m, and coefficients of determination (R2) of 0.948, 0.930, and 0.946. The RF and MLP models have relatively good consistency across the entire depth range, while the SVM model has an overestimation of depth in general within the interval of 6-9 m. To sum up, machine learning models have higher accuracy in water depth retrieval compared to traditional logarithmic models. Among them, the RF and MLP models are more suitable than the SVM model for water depth retrieval of tufa lakes based on UAV imagery. Unlike the models only utilizing blue and green bands, the introduction of red bands into machine learning models both improves the accuracy of the shallow-water bathymetry and at the same time increases the local bathymetry uncertainty. In the future, further research using UAV multispectral imagery with the coastal blue band is necessary. Given an adequate dataset, it is proposed to construct a deep convolutional neural network-based bathymetric inversion model for tufa lakes.