Random Forest and SVM-RBF target gold at Wadi Umm Eish El-Zarqa, Egypt using EnMAP hyperspectral data, with RF achieving perfect precision and 37% more high-confidence targets

A reproducible, precision-first framework was established for targeting both bedrock and placer gold mineralization at Wadi Umm Eish El-Zarqa, the first such hyperspectral investigation in this region. Prior hyperspectral mineral exploration studies in the Eastern Desert did not cover this area, and no prior study integrated machine learning with EnMAP hyperspectral data to target placer gold deposits in this region. The abstract states it is 'the first such investigation in this region' and 'the first study to integrate machine learning with EnMAP hyperspectral data to target placer gold deposits in this region,' which is the authors' own originality positioning; detailed comparative evidence requires the full text.

RF and SVM-RBF performed differently in delineating mineralized targets: SVM-RBF achieved slightly higher overall accuracy (91.8% vs 89.9%), while RF delivered perfect precision in identifying mineralized targets and delineated 37% more high-confidence (>80% probability) potentially mineralized targets than SVM-RBF. The study directly compares the two models on the same dataset and introduces a 0.05 spectral probability tolerance to expand the training set, quantifying the trade-off between accuracy and target count. Based on ground-truth validation of delineated targets, with specific numerical comparisons of overall accuracy, precision, and high-confidence target counts.

Drainage networks extracted from ALOS-PALSAR DEM link upstream bedrock sources to downstream placers, identifying 3rd to 5th-order streams as optimal zones for placer deposition. Combines geomorphometric analysis with hyperspectral machine learning to link bedrock source areas to placer deposition zones, rather than only delineating bedrock alteration. Based on DEM drainage extraction and stream-order analysis; the abstract gives the conclusion that 3rd to 5th-order streams are optimal zones.

Fieldwork confirmed active artisanal excavations in the study area, with stream sediments reportedly yielding 25-30 grams of gold per truckload (~85 tons). Provides field corroboration for the remotely sensed targets, linking remote sensing predictions to ongoing gold mining activity. The abstract states 'reportedly yielding,' indicating field observation and reported data rather than controlled sampling and assay.

The framework is intended for gold exploration in arid environments, applicable to analogous areas with known mineral occurrences available for training and where EnMAP hyperspectral and ALOS-PALSAR DEM data can be obtained; for exploration teams, it offers a reproducible, precision-first workflow for allocating follow-up validation resources between bedrock and placer targets. The authors explicitly position it as a methodology transferable to analogous arid terrains worldwide.

Only the abstract and references were loaded, without the main text, figures, or tables, so training sample sizes, number of validation points, spectral preprocessing details, model hyperparameters, and the sampling and assay method behind the 25-30 grams per truckload figure cannot be verified; these details affect interpretation of the accuracy numbers and the placer yield statement. Additionally, the accuracy and target-count differences in the abstract come from a single regional case, and transferability still needs replication in more arid regions.