Zero-inflated alpha unit regression with spatially varying coefficients

Real-world proportion data often exhibit an excess of fixed boundary values, such as zeros in mineral prospectivity mapping, which standard continuous unit-interval models cannot accommodate. This work proposes a new likelihood-based regression framework for proportional responses on [0,1] with excess probability mass and spatial information. We first introduce the zero/one-inflated Alpha Unit (iAU) distribution, derive its main stochastic properties, and obtain closed-form maximum likelihood estimators for the inflation parameter ν and the shape parameter α. We then extend this distribution to a geographically weighted regression specification, yielding a zero-inflated spatially varying model (GWR-ziAU) in which both the inflation probability and the continuous component depend on covariates and may vary smoothly over space. Simulation studies indicate low bias, decreasing MSE, and empirical coverage close to nominal levels, even under strong zero inflation. An application to cobalt (Co) concentration data from mineral deposits in northern Chile shows that the proposed model captures spatial heterogeneity in both the continuous and inflated components and attains competitive predictive performance. The proposed framework supports decision-making on cobalt enrichment potential by combining zero-inflation probabilities, positive concentration estimates, and spatially structured geochemical information.