The Loess Plateau in China is prone to widespread land degradation (soil erosion, deforestation, and water loss), and therefore, ecological restoration programmes aiming to re-establish the ecosystem by revegetation have been implemented during recent decades. Consequently, a widespread increase in vegetation cover has been reported, but the state and dynamics of forests remain largely unknown. Here, we used field and satellite data to produce annual forest probability scores at 250 × 250 m between 2001 and 2016. We classified the forest dynamics into three types: net gain, net loss, and fluctuation. Assuming that an increasing probability score reflects forest growth, we categorised the annual forest areas into different growing stages indicating the transition between planting and mature forest. In total, the area of forest increase was 48,786 km², and the percentage of forested areas of the region changed from 8.19% to 15.82% within 16 years. In areas of forest net gain, there was a clear transition from low- to high-probability forest areas, providing evidence that a general forest transition is occurring on the Loess Plateau. The method developed for continuous monitoring of forest probability scores offers a benchmarking for further decision making on ecological restoration of the region and for mapping of afforestation projects elsewhere. The findings on major forest increases demonstrate that land degradation is in fact reversible, with implications for many ecosystem services (e.g., carbon storage and provisioning of forest products).