Predicting Deep Sea Polymetallic Nodule Abundance Based on Geophysical Data with Uncertainty Quantification

Polymetallic nodules constitute a strategically important deep sea mineral resource enriched in nickel, cobalt, copper, and other critical metals. Accurate prediction of their spatial distribution is crucial for resource assessment and sustainable exploitation. However, sparse sampling, high survey costs, and complex nonlinear interactions among environmental factors make this task difficult. To address these challenges, this study applies a quantile regression forest to analyze the data from 233 box-core stations across a 23,000 km2 area in the eastern Pacific, integrating bathymetry and backscatter intensity. Cross-validation verifies model performance. The trained model predicts nodule abundance across an adjacent 3000 km2 study area and validates it against 12 independent stations that were not involved in the training. This external validation gives a root mean square error of 4.11 kg/m2. The quantile regression forest model also estimates uncertainty to quantify prediction reliability. The resulting uncertainty map distinguishes high-confidence zones from areas with elevated uncertainty. In 46% of the study area, normalized uncertainty falls below 0.50, indicating higher reliability. The remaining 54% shows higher uncertainty and requires additional sampling. This method combining independent spatial validation and uncertainty visualization provides a transparent tool for deep sea mineral resource assessment where data are sparse.