The "Great Green Wall" (GGW) initiative emphasizes tree-based restoration to combat desertification and improve livelihoods across sub-Saharan Africa (Sahel). However, the effectiveness of current GGW plans and the implications of not distinguishing between tree and non-tree vegetation restoration efforts remain uncertain under a changing climate. To address this gap, we propose a multi-task prediction framework that integrates Coupled Model Intercomparison Project Phase 6 model outputs, habitat similarity theory, and deep learning to forecast sub-pixel-level vegetation coverage from 2015 to 2100. Each pixel is represented as a proportional mixture of different ground cover types, such as percent tree cover, non-tree vegetation, and bare ground, rather than being classified as a single discrete land cover type. Our results indicate that the potential for non-tree vegetation restoration in the Sahel reaches a maximum increase of 7.51%, which is at least five times greater than that of tree-based vegetation under low, medium, and high emission scenarios. Spatial analysis reveals that prime areas for non-tree vegetation recovery consistently cluster between 13 degrees N and 15 degrees N in the central Sahel, covering about 4 to 4.7 million square kilometers under SSP2-4.5. While these identified areas largely align with current GGW planning zones, discrepancies emerge in the middle and western coastal regions. For instance, southern Chad is projected to have significant recovery potential, yet it is not currently included in GGW initiatives. Our findings underscore the need for more nuanced, localized, and climate-responsive strategies in restoring both tree and non-tree vegetation to bolster resilience in the Sahel.