The origin and classification of open and mosaic ecosystems, particularly in the tropics and subtropics, have led to controversy worldwide. This has affected biodiversity conservation and, in some cases, promoted the establishment of afforestation projects based on the assumption that open and mosaic ecosystems are degraded forests. Although this initiative can have benefits in terms of carbon storage and climate mitigation if carefully planned and managed, it can also cause biodiversity loss and degradation when afforestation takes place in areas that were previously open ecosystems, or where unsuitable species are used. Madagascar, a world biodiversity hotspot, is one of the countries targeted for the implementation of afforestation projects. The Central Highlands of Madagascar, dominated by grassland matrix with forest patches, is the main region targeted. The nature and origin of the landscape are hotly debated, however, and it is not clear whether these open ecosystems are ancient or anthropogenically derived. Understanding of landscape history is therefore required to identify and conserve ancient open ecosystems, and to distinguish them from areas that have been deforested by people. This research aims to reconstruct the vegetation history and environmental change in the Central Highlands of Madagascar during the Holocene using palaeoecological methods, in order to inform appropriate conservation and management plans. We provide new records of vegetation, hydrological change, fire and herbivory activities by using a multiproxy approach, which includes fossil pollen, stable carbon isotopes, diatoms, charcoal and coprophilous spores, that allows for a comprehensive investigation into the history and drivers of vegetation change. Sediment cores were collected from two sites, Tampoketsa-Ankazobe wetland and Lake Dangovavy, located in the eastern and western slopes of the highlands, respectively. Results indicated that the surrounding area of both sites was composed of mosaic ecosystems, comprising of forest patches of variable extent in a matrix of open grassland and ericoid shrubland vegetation, at least from the Early and Mid-Holocene to ca. 1000 cal years BP, driven mainly by climate variability and fire occurrence. In Tampoketsa-Ankazobe wetland (eastern slopes), the vegetation was characterised by a mosaic of ericoid shrubland and mid-elevation forest taxa, between ca. 11 200 and 8300 cal years BP, under warm/wet period and low fire occurrence. The vegetation in the area changed to a mosaic of ericoid shrubland with more dominance of high-elevation forest from ca. 8300 to 1000 cal years BP under a drier climate and consistent low fire occurrence. The abundance of shrubs and trees during those two periods were confirmed by the dominance of C3 plants as reflected by the stable carbon isotopes results, and coincided with low herbivory activities in the TampoketsaAnkazobe site from ca. 11 200 to 1000 cal years BP. In parallel, the pollen record from Lake Dangovavy (western slopes), between ca. 6200 and 5400 cal years BP, suggests a mosaic ecosystem, dominated by more C3 montane grass, ericoid shrubland and high elevation forest patches promoted by cool/dry climate with low fire occurrence and herbivory activities. Between ca. 5400 and 4200 cal years BP, vegetation in the area was dominated by a mosaic of ericoid shrubland and mid-elevation forest under a wetter period, moderate fire occurrence, and herbivory activities. This mosaic was controlled by climate, fire refugia and herbivory feedbacks. The vegetation changed into a forest-savanna mosaic with an abundance of grassland and pioneer/fire-resistant trees between ca. 4200 and 3000 cal years BP. The period was characterised by an initial increase of local fire followed by a regional drought event. This suggests that a threshold might have been reached, with a resulting shift in vegetation composition. Between ca. 3000 and 1000 cal years BP, reoccurrence of ericoid shrubland with woodland savanna taxa was recorded in the area. The vegetation was conditioned by variation of climate from wet (until ca. 2000 cal years BP) to dry period (ca. 2000–⁠1000 cal years BP) with moderate fire occurrence and herbivory activities. In addition, stable carbon isotope results show that between ca. 6200 to 1000 cal years BP, the site was characterised by C3 plants. During the last ca.1000 cal years BP, pollen records from both sites in the Central Highlands of Madagascar showed a shift to a more open landscape dominated by grassland. Trees and shrubland in the highlands experienced a massive decrease and this correlated with an abundance of C4 plants associated with reduced diversity. The shift of vegetation during this period was likely a result of a centennial severe drought period at ca. 950 cal years BP, as recorded in the literature and confirmed by the peak in aerophilous taxa in our diatom record. The drought was followed by a dramatic increase of fire occurrence and herbivory activities in the region, as recorded in the charcoal and spore records from both sites, indicating human activities at ca. 700 and 500 cal years BP for Lake Dangovavy and Tampoketsa-Ankazobe wetland, respectively. Though the vegetation at both sites in Central Highlands of Madagascar was very dynamic until ca. 1000 cal years BP, complex interactions between climate and fire allowed the forest and ericoid elements to persist, consistent with a heterogeneous mosaic landscape. This changed from 1000 years ago with the occurrence of a regional severe drought event followed by an increase in human activities leading to an increase of grass, a decline in forest and ericoid elements. Our findings suggest that although, the eastern and western slopes in Madagascar might have different vegetation histories over time as a response to the complex climatic-fire drivers at least until ca. 1000 cal years BP, they both: a) Contained ancient open ecosystems such as grasslands and/or ericoid shrubland, and a mosaic landscape which should be considered typical of the highland region. b) Experienced a loss of forest, woodland and mosaic elements, a trend that is consistent with the anthropogenic conversion of some forests to grasslands since ca. 1000 cal years BP. Such findings have implications in terms of conservation, fire management and afforestation projects in the Central Highlands, and provide additional knowledge that contributes to the understanding of its ecological processes and history prior to human arrival on the island. Indeed: 1) Ancient grasslands and ericoid shrubland need to be identified and conserved because of their antiquity and unique biodiversity. To date, there has been some focus on ancient grasslands, but the presence of ancient heathlands has not been discussed. 2) It is important to distinguish ancient from derived grasslands and to target the latter for reforestation, using species that are typical of the remaining forest patches. 3) Fire management should be conducted at a local scale and should incorporate the landscape fire history, considering, for example, the differences between two slopes in the Central Highlands.