Abstract. In this study, high-resolution bathymetric multibeam and optical image data, both obtained within the Belgian manganese (Mn) nodule mining license area by the autonomous underwater vehicle (AUV) Abyss, were combined in order to create a predictive random forests (RF) machine learning model. AUV bathymetry reveals small-scale terrain variations, allowing slope estimations and calculation of bathymetric derivatives such as slope, curvature, and ruggedness. Optical AUV imagery provides quantitative information regarding the distribution (number and median size) of Mn nodules. Within the area considered in this study, Mn nodules show a heterogeneous and spatially clustered pattern, and their number per square meter is negatively correlated with their median size. A prediction of the number of Mn nodules was achieved by combining information derived from the acoustic and optical data using a RF model. This model was tuned by examining the influence of the training set size, the number of growing trees (ntree), and the number of predictor variables to be randomly selected at each node (mtry) on the RF prediction accuracy. The use of larger training data sets with higher ntree and mtry values increases the accuracy. To estimate the Mn-nodule abundance, these predictions were linked to ground-truth data acquired by box coring. Linking optical and hydroacoustic data revealed a nonlinear relationship between the Mn-nodule distribution and topographic characteristics. This highlights the importance of a detailed terrain reconstruction for a predictive modeling of Mn-nodule abundance. In addition, this study underlines the necessity of a sufficient spatial distribution of the optical data to provide reliable modeling input for the RF.

The abyssal seafloor in the Clarion-Clipperton Zone (CCZ) in the NE Pacific hosts the largest abundance of polymetallic nodules in the deep sea and is being targeted as an area for potential deep-sea mining. During nodule mining, seafloor sediment will be brought into suspension by mining equipment, resulting in the formation of sediment plumes, which will affect benthic and pelagic life not naturally adapted to any major sediment transport and deposition events. To improve our understanding of sediment plume dispersion and to support the development of plume dispersion models in this specific deep-sea area, we conducted a small-scale, 12-hour disturbance experiment in the German exploration contract area in the CCZ using a chain dredge. Sediment plume dispersion and deposition was monitored using an array of optical and acoustic turbidity sensors and current meters placed on platforms on the seafloor, and by visual inspection of the seafloor before and after dredge deployment. We found that seafloor imagery could be used to qualitatively visualise the redeposited sediment up to a distance of 100 m from the source, and that sensors recording optical and acoustic backscatter are sensitive and adequate tools to monitor the horizontal and vertical dispersion of the generated sediment plume. Optical backscatter signals could be converted into absolute mass concentration of suspended sediment to provide quantitative data on sediment dispersion. Vertical profiles of acoustic backscatter recorded by current profilers provided qualitative insight into the vertical extent of the sediment plume. Our monitoring setup proved to be very useful for the monitoring of this small-scale experiment and can be seen as an exemplary strategy for monitoring studies of future, upscaled mining trials. We recommend that such larger trials include the use of AUVs for repeated seafloor imaging and water column plume mapping (optical and acoustical), as well as the use of in-situ particle size sensors and/or particle cameras to better constrain the effect of suspended particle aggregation on optical and acoustic backscatter signals.

No abstract

Machine learning spatial modeling is used for mapping the distribution of deep-sea polymetallic nodules (PMN). However, the presence and influence of spatial autocorrelation (SAC) have not been extensively studied. SAC can provide information regarding the variable selection before modeling, and it results in erroneous validation performance when ignored. ML models are also problematic when applied in areas far away from the initial training locations, especially if the (new) area to be predicted covers another feature space. Here, we study the spatial distribution of PMN in a geomorphologically heterogeneous area of the Peru Basin, where SAC of PMN exists. The local Moran’s I analysis showed that there are areas with a significantly higher or lower number of PMN, associated with different backscatter values, aspect orientation, and seafloor geomorphological characteristics. A quantile regression forests (QRF) model is used using three cross-validation (CV) techniques (random-, spatial-, and cluster-blocking). We used the recently proposed “Area of Applicability” method to quantify the geographical areas where feature space extrapolation occurs. The results show that QRF predicts well in morphologically similar areas, with spatial block cross-validation being the least unbiased method. Conversely, random-CV overestimates the prediction performance. Under new conditions, the model transferability is reduced even on local scales, highlighting the need for spatial model-based dissimilarity analysis and transferability assessment in new areas.