Platform Selection, Sensor Trade-offs, and Architecture Performance: A Decision Framework for Operational Wildlife Monitoring

Fragile Andean moorlands lack of pipeline methodologies for aerial wildlife monitoring. This situation enables invasive anthropocentric pressures, which negatively impact the ecosystem. Operational deployment of cattle monitoring requires navigating through critical trade-offs among pipeline decisions into the system design. This comprehensive review synthesizes relevant studies in the last decade, following PRISMA methodology, from platform capabilities to architectural trade-offs focused in their application to moorland ecosystems in order to establish operational computer vision pipelines. Aerial platform characteristics determine spatial resolution as well as pre-processing and architectural needs. UAVs cover medium scales with trade-off in ground sampling distance. This spatial resolution makes them optimal for surveys under 50 km2. Manned aircraft and VHR satellites become cost-effective only beyond 100 km2, but resolution degrades. Sensor decision depends on habitat type characteristics. Sensor fusion quantification demonstrates that thermal-RGB integration improves species detection, in order to justify cost in heterogeneous habitats versus RGB sufficient for open landscapes that achieves higher precision. Pre-processing impact assessment shows ESRGAN super-resolution provides detection gain for objects at altitudes up 250 meters, whereas CLAHE enhancement deliver high classification accuracy at negligible computational cost. Architecture performance taxonomy establishes YOLO variants dominate real-time embedded systems, transformer models justify computational overhead exclusively in multi-species scenarios, and density regression achieves MAE below 10 for ultra-dense aggregations. Critical deployment barriers include performance degradation across altitude variations, domain generalization failures that require annotation effort reduction through point-based weak supervision, and absence in transfer learning protocols for data-sparse ecosystems. This review establishes methodological foundations for high-altitude Andean moorland conservation where operational datasets remain absent, in order to provide quantitative decision matrices applicable to fragile mountainous ecosystems globally that requires evidence-based monitoring system design.