Understanding the process of dryland degradation is crucial to combating desertification and maintaining land sustainability. To identify important indicators of dryland degradation, we conducted a field survey of 46 drylands spanning an aridity gradient across the Tibetan Plateau. Our random forest model suggested that soil sand content was the most important predictor of variation in soil fertility (carbon, nitrogen, and phosphorus). Most soil fertility showed a nonlinear and abrupt decline with increasing soil sand content, and displayed three sequential thresholds. First, soil organic carbon sharply decreased beyond a soil sand content threshold of 47%. Second, exceeding a subsequent threshold of 70%, the total nitrogen and light fraction carbon declined dramatically with increasing soil sand content. Finally, soil heavy fraction carbon, inorganic carbon, and total carbon were drastically reduced when the soil sand content exceeded 80%. The three soil fertility thresholds along the gradient of the soil sand content indicated a sequence of dryland degradation phases. Our findings provide an approach for detecting the severity of dryland degradation and suggest vulnerable areas of degradation on the Tibetan Plateau.