131 Case 2. Alemão: Multivariate TMI + Radiometric Alemão was selected as the representative multivariate case because it provides a direct comparison between single-variable rankings and a fused workflow. In this case, sPCA was first applied separately to each variable, the retained score vectors were concatenated and standardized, and similarity was then computed in the fused space using deposit-based weights. The resulting top-ranked windows from the combined TMI + radiometric U analysis are shown in Figure 3. Using Alemão as the training deposit (Deposit 3) and Deposits 1, 2, 4, and 5 as test deposits, sPCA Multi achieved the strongest validation performance, recovering 77.7% of the total test-deposit area by rank 250 (Error! Reference source not found.). This exceeded sPCA Univariate Radiometric U, which recovered 52.8%, sPCA Univariate TMI, which recovered 31.0%, and Raw Multi, which recovered only 7.9%. The fused multivariate ranking was also the only approach to produce hits for all four test deposits, with the first threshold-reaching hit occurring at rank 11. 5. DISCUSSION This study aimed to develop a quantitative, reproducible, and interpretable framework for geophysical prospecting based on spatial similarity to a known deposit. The results of both case studies support that objective. Rather than comparing windows only in the original pixel space, the sPCA approach decomposes the reference anomaly into spatial modes and ranks candidate windows according to how strongly they reproduce those deposit-specific patterns. An important outcome is that the method remains physically interpretable: the loading maps and reconstruction progression show that the highest-weighted modes are also the modes that rebuild the observed deposit signal. In this sense, the deposit-based weights are not only ranking coefficients, but also measures of which spatial patterns are most diagnostic of the training deposit. This is useful in brownfield exploration, where targets are more valuable if they can be related back to recognizable geophysical signatures rather than treated as black-box outputs. The first case, based on Deposit 6 and univariate TMI, provides the clearest proof of concept. sPCA substantially outperformed the raw TMI ranking, increasing total recovered test-deposit area from 10.2% to 54.1%, hit count from one to five, and cumulative recovery AUC from 14.04 to 83.15. These gains indicate that sPCA performs especially well when the reference deposit has a strong and internally coherent geophysical signature. In such settings, the decomposition isolates the dominant spatial geometry of the
RkJQdWJsaXNoZXIy MTM0Mzk2