The ecosystems characterized by arid and semi-arid climate conditions are threatened heavily by processes of degradation, especially due to vegetation loss and soil salinization. However, despite the revolution in environmental monitoring that has been caused by the development of Remote Sensing (RS) and Machine Learning (ML) approaches, it is hard to synthesize rapidly evolving developments in this field. In the current paper, we offer an overview of the application of RS and ML approaches for vegetation dynamics and soil salinization monitoring. Thus, our review proposes a new, change-oriented, and comprehensive conceptual framework that goes beyond linear processing with the use of multi-temporal features extraction, hybrid predictive modeling, and adaptive decision-making (MCDM inclusion). In our work, we emphasize several crucial methodological bottlenecks, such as data scarcity, spatial autocorrelation, and the black box of advanced algorithms. With the increasing use of ML frameworks in environmental decisions, issues related to ethics, reproducibility, and open science should be considered carefully. The combination of high-resolution remote sensing techniques, including those using Unmanned Aerial Vehicles (UAVs), can pose privacy problems and requires strict compliance with local data governance. In addition, this review tackles institutional barriers along with the pressing need for Open Science, which involves the development of benchmark datasets and reproducible methodologies. In essence, this paper offers a clear strategy to guide researchers and policy makers towards developing ML-enabled monitoring systems for managing arid areas sustainably.