Bathymetric Inversion Research Articles

Shallow Water Bathymetry Inversion Based on Machine Learning Using ICESat-2 and Sentinel-2 Data

Shallow water bathymetry is essential for maritime navigation, environmental monitoring, and coastal management. While traditional methods such as sonar and airborne LiDAR provide high accuracy, their high cost and time-consuming nature limit their application in remote and sensitive areas. Satellite remote sensing offers a cost-effective and rapid alternative for large-scale bathymetric inversion, but it still relies on significant in situ data to establish a mapping relationship between spectral data and water depth. The ICESat-2 satellite, with its photon-counting LiDAR, presents a promising solution for acquiring bathymetric data in shallow coastal regions. This study proposes a rapid bathymetric inversion method based on ICESat-2 and Sentinel-2 data, integrating spectral information, the Forel-Ule Index (FUI) for water color, and spatial location data (normalized X and Y coordinates and polar coordinates). An automated script for extracting bathymetric photons in shallow water regions is provided, aiming to facilitate the use of ICESat-2 data by researchers. Multiple machine learning models were applied to invert bathymetry in the Dongsha Islands, and their performance was compared. The results show that the XG-CID and RF-CID models achieved the highest inversion accuracies, 93% and 94%, respectively, with the XG-CID model performing best in the range from −10 m to 0 m and the RF-CID model excelling in the range from −15 m to −10 m.

Integration of geographic features and bathymetric inversion in the Yangtze River's Nantong Channel using gradient boosting machine algorithm with ZY-1E satellite and multibeam data

This study investigates the integration of geographic features from ZY-1E satellite data with advanced machine learning techniques to enhance water depth inversion in the Yangtze River's Nantong Channel. Utilizing the Gradient Boosting Machine (GBM) and its geospatially enhanced version, GBM-Lon./Lat., significant improvements in modeling precision were observed, as reflected by lower RMSE and higher R² values compared to traditional depth inversion methods. The research underscores the benefits of incorporating geospatial data, which allows for a more nuanced understanding of the hydrological dynamics and facilitates more accurate predictions in the turbid waters of the channel. Challenges such as atmospheric effects, water turbidity, and data acquisition issues under variable weather conditions were identified. The study proposes further optimization of these models to handle diverse environmental conditions and enhance the accuracy of bathymetric mapping. The integration of machine learning with remote sensing not only supports navigational safety and efficient waterway management but also contributes significantly to environmental monitoring and sustainable riverine infrastructure development.

Pixel-level bathymetry mapping of optically shallow water areas by combining aerial RGB video and photogrammetry

Combining geometric and optical methods based on a low-cost UAV platform can achieve high-resolution bathymetry mapping without ground-truth data in optically shallow water areas. However, with the increasing spatial resolution, water surface fluctuation interferes with the imaging. In this study, we propose a bathymetry mapping approach that combines video and geometric-optical principles. The multi-sampling of the video data allows for a temporal averaging window of each pixel. A motion-based frame registration method was developed to compose an image from a video acquired by UAV push-broom sampling to mitigate the instantaneous changes caused by water surface fluctuations. The composite images were used for bathymetry mapping using data from the photometric point clouds from the UAV images for calibration. Then, the improved effect of video multi-sampling on the optical bathymetric model was evaluated by comparing the results of bathymetric inversion based on single images and video composite images. An evaluation case in the coastal area of Hainan Island demonstrates that results based on composite images processed with three optical bathymetric models of different complexity increased the coefficient of determination from 0.8111, 0.8652, 0.9255 to 0.8652, 0.8750, 0.9363, and reduced the root mean square error from 0.234 m, 0.192 m, 0.143 m to 0.197 m, 0.178 m, 0.133 m, respectively. Qualitatively, using composite images from aerial RGB videos for bathymetry mapping effectively removes radiative anomalies due to wave focusing or reflection and provides a more accurate description of underwater objects' shape than single images.

Satellite-Derived Bathymetry Using a Fast Feature Cascade Learning Model in Turbid Coastal Waters

Obtaining accurate bathymetric maps is very valuable for marine environment monitoring, port planning, and so on. Accurately estimating water depth in turbid coastal waters using satellite remote sensing encounters challenges originating from low water transparency, but it is limited by the quantity, quality, and water quality of samples. This study introduces a fast feature cascade learning model (FFCLM) to enhance the accuracy of bathymetric inversion from multispectral satellite images, particularly when limited field samples are available. FFCLM leverages spectral bands and in situ data to derive effective inversion weights through feature concatenation and cascade fitting. Field experiments conducted at Nanshan Port and Rushikonda Beach gathered water depth, satellite, and in situ data. Comparative analysis with conventional machine learning algorithms, including support vector machine, random forest, and gradient boosting trees, indicates that FFCLM achieves lower errors and demonstrates more robust performance across study areas. This is especially more pronounced when using small training samples ( n < 100). Examination of key parameters and water depth profiles highlights FFCLM’s advantages in generalization and deep-water inversion. This study presents an efficient solution for small-sample bathymetric mapping in turbid coastal waters, utilizing spectral and physical information to overcome sample size limitations and enhancing satellite remote sensing capabilities for shallow water monitoring.

An improved method for water depth mapping in turbid waters based on a machine learning model

Water depth data products are fundamental for the study of near-shore marine science and coastal engineering; therefore, it is essential to identify the optimal technique for coastal water depth mapping. There are reliable mapping techniques for clear open waters. However, near-shore waters is typically turbid, which decreases the accuracy of previous methods for clear waters. To address these challenges, in this study, a machine learning-based topographic inversion method for turbid waters was developed. The log band ratio algorithm presented in previous studies was modified initially to enhance its application in turbid waters. Then, we established a model called AdaBoost-GBDT for bathymetric inversion in turbid waters using the Gradient Boosting Decision Tree (GBDT) algorithm fused with the Adaptive Boosting (AdaBoost) algorithm. Comparison of the results revealed that the model is highly reliable in all water depth intervals. To evaluate the quality of our model derived product in coastal research, a comparison of the data quality among four water depth datasets was conducted in this study. The comparison demonstrated that the inversion results of AdaBoost-GBDT have the best data coverage and accuracy. Furthermore, when applied to the Finite Volume Community Ocean Model (FVCOM), the AdaBoost-GBDT bathymetric product had the best performance. This study provides a new method for the bathymetric inversion of turbid waters, which can contribute to coastal management and engineering.

Georeferencing of UAV imagery for nearshore bathymetry retrieval

In dynamic coastal areas where rapid changes in the seabed can have a significant impact on day–to–day operations, as is the case of Figueira da Foz harbor inlet, it is important to have updated knowledge of the bathymetry. As in energetic wave conditions it is not possible to carry out traditional bathymetric surveys, this work proposes a methodology that allows, using UAVs, to estimate the bathymetry through bathymetric inversion methods. On two different occasions, flights were carried out with a commercial UAV to obtain local bathymetry at Figueira da Foz harbor inlet. Since this location is far from the coast, there is not a minimum amount of land to use ground control points that allow georeferencing the acquired images in the traditional way with the necessary spatial resolution. Therefore, an automatic image georeferencing methodology was developed and validated in this work. A spectral analysis is used to retrieve both wave periods and wavelengths, respectively, by evaluating the positions of the wave crests in both time and spatial domains. From this information, the water depths are estimated using the linear dispersion relation and the results are validated with field survey data. The automatic georeferencing methodology is shown to be capable of placing the image in the correct location by overlaying on a generic basemap. The spectral analysis showed positive results in estimating wavelengths and wave periods. However, the quality of bathymetric inversion in areas where morphological structures exist (sandbars), presents results with different accuracies depending on the wave and tide characteristics observed each day.

Satellite–Derived Bathymetry in Shallow Waters: Evaluation of Gokturk-1 Satellite and a Novel Approach

For more than 50 years, marine and remote sensing researchers have investigated the methods of bathymetry extraction by means of active (altimetry) and passive (optics) satellite sensors. These methods, in general, are referred to as satellite-derived bathymetry (SDB). With the advances in sensor capabilities and computational power and recognition by the International Hydrographic Organization (IHO), SDB has been more popular than ever in the last 10 years. Despite a significant increase in the number of studies on the topic, the performance of the method is still variable, mainly due to environmental factors, the quality of the deliverables by sensors, the use of different algorithms, and the changeability in parameterization. In this study, we investigated the capability of Gokturk-1 satellite in SDB for the very first time at Horseshoe Island, Antarctica, using the random forest- and extreme gradient boosting machine learning-based regressors. All the images are atmospherically corrected by ATCOR, and only the top-performing algorithms are utilized. The bathymetry predictions made by employing Gokturk-1 imagery showed admissible results in accordance with the IHO standards. Furthermore, pixel brightness values calculated from Sentinel-2 MSI and tasseled cap transformation are introduced to the algorithms while being applied to Sentinel-2, Landsat-8, and Gokturk-1 multispectral images at the second stage. The results indicated that the bathymetric inversion performance of the Gokturk-1 satellite is in line with the Landsat-8 and Sentienl-2 satellites with a better spatial resolution. More importantly, the addition of a brightness value parameter significantly improves root mean square error, mean average error, coefficient of determination metrics, and, consequently, the performance of the bathymetry extraction.

Evaluation of nearshore bathymetric inversion algorithms using camera observations and synthetic numerical input of surface waves during storms

Nearshore bathymetry is difficult to measure using survey methods when wave heights approach the breaking limit. Remote sensing using digital cameras offers a way to observe the surf zone and calculate water depths based on phase speed but comes with its challenges of potentially noisy data that can introduce error into estimates of frequency and wavenumber used in phase speed calculation. This study investigates the robustness of a new version of a bathymetric inversion algorithm (cBathy, version 2.0) in moderate to energetic wave conditions by comparing depth estimates from timeseries’ of pixel intensity with depth estimates from synthetic water level data. The synthetic data are generated by the phase-resolving numerical model, SWASH, and optical data were collected during a field experiment in 2015. Model results from SWASH computed with known bathymetry are used as input to cBathy, and depth estimates are compared to nearshore surveys. Argus camera observations are also used as input to cBathy for the same times as the SWASH simulations. The SWASH simulations resolve breaking waves and do not include (optical) changes to the relation between water surface slope and pixel intensity, termed modulation transfer function, that occur during wave breaking, enabling better estimates from bathymetric inversion algorithms near morphologic features like sand bars. The results indicate that improvements result from eliminating of disruptions to the modulation transfer function caused by wave breaking and residual foam. We show that the use of synthetic wave data is a valuable means of isolating errors in bathymetric inversion algorithms.

APPLICATION OF A NUMERICAL MODEL AND BATHYMETRIC INVERSION ALGORITHMS TO ENHANCE UNDERSTANDING OF NEARSHORE CHANGE

During extreme wave events, shoaling occurs farther from shore, extending the surf zone width and intensifying wave breaking (Thornton and Guza, 1983; Mulligan and Hanson, 2016). Remote sensing using digital cameras is an inexpensive way to gather continuous observations of the surf zone and calculate water depths based on phase speed. This study investigates the robustness of the newly released version of cBathy (Holman et al., 2013; Holman and Bergsma, 2021), a bathymetric inversion algorithm. cBathy is employed in moderate to energetic wave conditions using synthetic water level data generated by the non-hydrostatic numerical phase resolving model, SWASH (Simulating WAves till Shore) (Zijlema et al., 2011; Gomes et al., 2016; Fiedler et al., 2019). The overall performance results are described by the variation in root-mean-squared errors (RMSE) with significant wave height, and with depth. The results show that RMSE are dependent on water depth, significant wave height, and domain coverage by breaking waves. cBathy V2.0 consistently performs better than the preceding version, when comparing depth estimates to survey observations.

Multispectral Satellite-Derived Bathymetry Based on Sparse Prior Measured Data

Satellite-derived bathymetry is an economic and effective method of obtaining large-scale, high-resolution bathymetric information. At present, bathymetric inversion models require existing bathymetric data as a necessary condition, but these may be difficult to obtain around many small islands. Under the condition of sparse measurement data, many common methods have low levels of accuracy. This paper proposes a method of transferring a bathymetric model for an island that has in situ data to an island that has little in situ data. First, in situ data are used to establish a high-precision satellite bathymetry model for a given island. Second, this model is transplanted to other islands. The addition of just 2–3 data points to correct the model results can reduce the mean relative error to 6.62% in the range of 0–20 m. This is close to the accuracy of establishing the model using a large number of measurement data (mean relative error of 5.25%).

Satellite-derived bathymetry integrating spatial and spectral information of multispectral images.

As a significant and cost-effective method of obtaining shallow seabed topography, satellite derived bathymetry (SDB) can acquire a wide range of shallow sea depth by integrating a small quantity of in-situ water depth data. This method is a beneficial addition to traditional bathymetric topography. The seafloor's spatial heterogeneity leads to inaccuracies in bathymetric inversion, which reduces bathymetric accuracy. By utilizing multispectral data with multidimensional features, an SDB approach incorporating spectral and spatial information of multispectral images is proposed in this study. In order to effectively increase the accuracy of bathymetry inversion throughout the entire area, first the random forest with spatial coordinates is established to control bathymetry spatial variation on a large scale. Next, the Kriging algorithm is used to interpolate bathymetry residuals, and the interpolation results are used to adjust bathymetry spatial variation on a small scale. The data from three shallow water sites are experimentally processed to validate the method. Compared with other established bathymetric inversion techniques, the experimental results show that the method effectively reduces the error in bathymetry estimation caused by spatial heterogeneity of the seabed, producing high-precision inversion bathymetry with a root mean square error of 0.78 to 1.36 meters.

High-Precision Inversion of Shallow Bathymetry under Complex Hydrographic Conditions Using VGG19—A Case Study of the Taiwan Banks

Shallow bathymetry is important for ocean exploration, and the development of high-precision bathymetry inversion methods, especially for shallow waters with poor quality, is a major research aim. Synthetic aperture radar (SAR) image data benefit from a wide coverage, high measurement density, rapidity, and low consumption but are limited by low accuracy. Alternatively, multibeam data have low coverage and are difficult to obtain but have a high measurement accuracy. In this paper, taking advantage of the complementary properties, we use SAR image data as the content map and multibeam images as the migrated style map, applying the VGG19 neural network (optimizing the loss function formula) for bathymetric inversion. The model was universal and highly accurate for bathymetric inversion of shallow marine areas, such as turbid water in Taiwan. There was a strong correlation between bathymetric inversion data and measured data (R2 = 0.8822; RMSE = 1.86 m). The relative error was refined by 9.22% over those of previous studies. Values for different bathymetric regions were extremely correlated in the region of 20–40 m. The newly developed approach is highly accurate over 20 m in the open ocean, providing an efficient, precise shallow bathymetry inversion method for complex hydrographic conditions.

Research on Bathymetric Inversion Capability of Different Multispectral Remote Sensing Images in Seaports.

In recent years, remote sensing has become an indispensable supplementary method for determining water depth in the seaports. At present, many scholars use multi-spectral satellite data to invert the water depth of the seaports, but how to select the appropriate satellite data in the seaports area is worth exploring. In this article, the differences in the retrieving ability between domestic and foreign multispectral images are compared, through building the random forest model and the band ratio model, which use different multispectral images to conduct retrieving water depth in Nanshan Port in conjunction with the WBMS multi-beam sounding system. The band ratio model and random forest model are chosen for water depth exploration, remote sensing images use GF-6, GF-2, Sentinel-2B, and Landsat 8 OLI data, which are all popular and easily accessible. The final experiment results from the constant adjustment of the model parameter show that the domestic series of GF-6 images performed the best in this experiment. The Root Mean Square Error (RMSE) and Mean Relative Error (MRE) of the random forest model are only 1.202 and 0.187, respectively. Simultaneously, it is discovered that the 'Red Edge' band of GF-6 is also very helpful in improving the accuracy of water depth inversion, which is rarely mentioned in previous studies. To some extent, the preceding studies demonstrate that it is possible to investigate water depth using common multispectral remote sensing images. In the case of some bathymetry inversion models or in some waters, the aforementioned study demonstrates that it is possible to examine the water depth using domestic remote sensing images that are superior to foreign multispectral images in terms of bathymetry inversion ability.

ICESat-2 Bathymetric Signal Reconstruction Method Based on a Deep Learning Model with Active–Passive Data Fusion

When carrying out SDB (satellite-derived bathymetry) in island area based on ICESat-2 (Ice, Cloud, and land Elevation Satellite 2) data, it is often found that the ICESat-2 bathymetric signals are partially missing due to the influence of thick aerosols such as clouds and fog. This not only hinders the accurate extraction of the along-track underwater topography, but also restricts the active–passive fusion bathymetry based on ICESat-2 data and multi/hyperspectral remote sensing images. In this paper, aiming at the partially missing ICESat-2 bathymetric signals, combined with passive optical remote sensing images, and based on an LSTM (long short-term memory) deep recurrent neural network model, an ICESat-2 bathymetric signal reconstruction method based on active–passive data fusion is proposed. It is found that this method can effectively reconstruct the local missing bathymetric signals. When the reconstructed ICESat-2 bathymetric data are applied to carry out active–passive fusion and bathymetric inversion, the accuracy indices are better than those of the inversion results of the data with partial missing signals, and the performance is comparable to that of the original data without missing data, which is of great value for the bathymetric application of ICESat-2 data in island and reef areas.

A Comparison of Machine Learning and Empirical Approaches for Deriving Bathymetry from Multispectral Imagery

Knowledge of the precise water depth in shallow areas of the ocean is of great significance to the safe navigation of ships and hydrographic surveying. Compared with traditional bathymetry, satellite remote sensing for water depth determination makes it possible to cover large areas by dynamic observation. In this paper, we conducted an optically shallow water bathymetric inversion study using a Stumpf empirical model, random forest model, neural network model, and support vector machine model based on Sentinel-2 satellite images and Ganquan Dao measured bathymetry data. We compared and analyzed the inversion results based on the empirical model and different machine learning models. The results show that the Stumpf empirical and machine learning models are capable of inverting optically shallow water depth. Moreover, the machine learning models had better fitting ability than the Stumpf empirical model with a sufficient number of samples, especially when the water depth was greater than 15 m. In addition, the random forest model had the highest overall accuracy among these models, with a root mean square error (RMSE) of 1.41 m and a regression coefficient (R2) of 0.96 for the test data.

Machine-Learning-Method-Based Inversion of Shallow Bathymetric Maps Using ICESat-2 ATL03 Data

The application of empirical methods for satellite-derived bathymetry is limited by the lack of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in situ</i> bathymetric data in remote, inaccessible areas. This challenge has been addressed with the launch of ICESat-2 (Ice, Cloud, and land Elevation Satellite-2). This study provides an accurate bathymetric photon extraction process for ICESat-2 ATL03 data, and the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">R</i> <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of the bathymetric photons obtained using this process and airborne bathymetric LiDAR data is up to 99%. Next, based on two types of remote sensing data, ICESat-2 and Sentinel-2, machine learning (ML) models, including linear regression (LR), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost), were trained to obtain bathymetric maps. The experimental results show that the mean RMSE (root mean square error), mean MAE (mean absolute error), and mean MRE (mean relative error) values of the LR models are less than 3.02 m, 2.38 m, and 86.03%, respectively. The mean RMSE, MAE, and MRE values of the LightGBM and CatBoost models are less than 0.91 m, 0.66 m, and 23.17%, respectively. It is concluded that the proposed denoising process for ICESat-2 ATL03 data is effective, and the results of the bathymetric maps obtained using these data are satisfactory. Thus, the proposed approach is effective, and this strategy can be used to replace conventional bathymetric inversion methods to obtain high-accuracy bathymetric maps.

Evaluation and Improvement of No-Ground-Truth Dual Band Algorithm for Shallow Water Depth Retrieval: A Case Study of a Coastal Island

Conventional bathymetric inversion approaches require bathymetric data as ground truth to obtain shallow water depth from high spatial resolution remote sensing imagery. Thus, bathymetric mapping methods that do not require inputs from in situ measurements are highly desirable. In this paper, we propose a dual-band model improvement method and evaluate the performance of this novel dual-band model approach to obtain the underwater terrain around a coastal island by using four WorldView-2/3 imageries. Then, we validate the results through changing water column properties with the Kd multiple linear regression model simulated by Hydrolight. We multiply the best coefficient and blue–green band value with different substrates on the pixels, which sample along the coastal line and isobath. The results show that the mean bias of inversed depth ranges from 1.73 to 2.96 m in the four imageries. The overall accuracy of root mean square errors (RMSEs) is better for depths shallower than 10 m, and the average relative error is 11.89%. The inversion accuracy of this new model is higher than Lee’s classical Kd model and has a wider range of applications than Chen’s dual-band model. The no-ground-truth dual-band algorithm has higher accuracy than the other log-ratio methods mentioned in this paper.

Atmospheric Correction Model for Water–Land Boundary Adjacency Effects in Landsat-8 Multispectral Images and Its Impact on Bathymetric Remote Sensing

Atmospheric correction (AC) is the basis for quantitative water remote sensing, and adjacency effects form an important part of AC for medium- and high-spatial-resolution optical satellite images. The 6S radiative transfer model is widely used, but its background reflectance function does not take the radiance changes at water–land boundaries into account. If the observed land possesses bright features, the radiance of the adjacent water will be affected, leading to deviations in the AC results and increasing the uncertainty of water depth-based optical quantitative remote sensing. In this paper, we propose a model named WL-AE (a correction model for water–land boundary adjacency effects), which is based on the obvious radiance differences at water–land boundaries. This model overcomes the problem by which the background reflectance calculation is not terminated due to the highlighting pixel. We consider the influences of different Rns (neighborhood space) on the target pixel. The effective calculation of the equivalent background reflectance of the target pixel is realized, and the influence of the land area anomaly highlighting the pixel on the adjacent water is avoided. The results show that WL-AE can effectively improve the entropy and contrast of the input image and that the water–land boundary is greatly affected by adjacency effects, especially in the green and near-infrared bands, where the Mrc (mean rate of change) are as high as 14.2% and 20.1%, respectively. In the visible wavelength, the Sd of Rrc (the relative rate of change) is positively correlated with Rns, and the Sd reaches 16.9%. Although the adjacency effect is affected by ground object types, its influence area remains within 3 km offshore. Based on the WL-AE and 6S results, the comparative test regarding bathymetric inversion shows that the influence is significant in the 0–5 m depth section. In Penang, the MRE of the 0–4 m inversion results is 31.4%, which is 10.5% lower than that of the 6S model.

Video-Based Nearshore Bathymetric Inversion on a Geologically Constrained Mesotidal Beach during Storm Events

Although geologically constrained sandy beaches are ubiquitous along wave-exposed coasts, there is still a limited understanding of their morphological response, particularly under storm conditions, which is mainly due to a critical lack of nearshore bathymetry observations. This paper examines the potential to derive bathymetries from video imagery under challenging wave conditions in order to investigate headland control on morphological beach response. For this purpose, a video-based linear depth inversion algorithm is applied to three consecutive weeks of frames collected during daylight hours from a single fixed camera located at La Petite Chambre d’Amour beach (Anglet, SW France). Video-derived bathymetries are compared against in situ topo-bathymetric surveys carried out at the beginning and end of the field experiment in order to assess the performance of the bathymetric estimates. The results show that the rates of accretion/erosion within the surf zone are strongly influenced by the headland, whereas the beach morphological response can be classified into three main regimes depending on the angle of wave incidence θp: (1) under deflection configuration (θp&gt;0°), the alongshore sediment transport was trapped at the updrift side of the headland, promoting sand accretion. (2) Under shadowed configuration (θp&lt;0°), the interruption of the longshore current drove a deficit of sand supply at the downdrift side of the headland, leading to an overall erosion in the surf zone. (3) Under shore-normal configuration (θp=0°), rip channels developed, and up-state beach transition was observed. A comparison between video-derived bathymetries and surveys shows an overall root mean square error (RMSE) around 0.49 to 0.57 m with a bias ranging between −0.36 and −0.29 m. The results show that video-derived bathymetries can provide new insight into the morphological change driven by storm events. The combination of such inferred bathymetry with video-derived surface current data is discussed, showing great potential to address the coupled morphodynamics system under time-varying wave conditions.

A Combined Approach for Retrieving Bathymetry from Aerial Stereo RGB Imagery

Shallow water bathymetry is critical in understanding and managing marine ecosystems. Bathymetric inversion models using airborne/satellite multispectral data are an efficient way to retrieve shallow bathymetry due to the affordable cost of airborne/satellite images and less field work required. With the increasing availability and popularity of unmanned aerial vehicle (UAV) imagery, this paper explores a new approach to obtain bathymetry using UAV visual-band (RGB) images. A combined approach is therefore proposed for retrieving bathymetry from aerial stereo RGB imagery, which is the combination of a new stereo triangulation method (an improved projection image based two-medium stereo triangulation method) and spectral inversion models. In general, the inversion models require some bathymetry reference points, which are not always feasible in many scenarios, and the proposed approach employs a new stereo triangulation method to obtain reliable bathymetric points, which act as the reference points of the inversion models. Using various numbers of triangulation points as the reference points together with a Geographical Weighted Regression (GWR) model, a series of experiments were conducted using UAV RGB images of a small island, and the results were validated against LiDAR points. The promising results indicate that the proposed approach is an efficient technique for shallow water bathymetry retrieval, and together with UAV platforms, it could be deployed easily to conduct a broad range of applications within marine environments.