1. Ecological restoration of tropical open ecosystems remains challenging for both science and practice. Over the last decade, innovative techniques have been developed, but whether they have been successful or not remains to be demonstrated. Assessing the outcomes of these initiatives is crucial to drive the following steps to improve tropical grasslands and savanna restoration.

2. Analyzing 82 data sets from the literature and primary data collection, we assessed the effectiveness of passive and active restoration techniques applied in Cerrado open ecosystems. We used plant diversity variables (species and growth forms) as indicators, considering ruderals and exotics as non-target species. Specifically, we aimed to answer: (i) How does the diversity of target species change through time in areas subject to passive restoration? (ii) Are active and passive restoration techniques effective in restoring the proportion of target species found in old-growth reference ecosystems? (iii) Have the current techniques been successful in recovering the proportions of growth forms of reference ecosystems?

3. We found that target species proportions do not increase with time, suggesting limitations of typical species to colonize degraded sites. Hence, passive restoration will promote the conservation of a limited and constant number of target species. This number will depend on the magnitude of degradation and previous land use.

4. The restoration techniques currently applied to restore the biodiversity of Cerrado open ecosystems are not reaching the reference standards, with distinct techniques driving plant communities to different sets of growth forms. Active restoration based on propagules obtained from pristine donor sites (topsoil translocation, plant material transplant, and seeding) performed better than passive restoration for most of the growth forms analyzed.

5. Synthesis and Applications: Different growth forms have different roles in determining the structure and functioning of Cerrado vegetation. A mix of techniques can better approximate plant diversity and the proportionality of target species of pristine ecosystems. Singular restoration approaches are insufficient for restoring Cerrado open ecosystem biodiversity. Mixed efforts encompassing various techniques are required instead. Furthermore, it is likely restoration success can be improved with greater investment in improving our understanding of and developing existing restoration techniques.

Literature review: We surveyed published articles and theses addressing different restoration techniques for open Cerrado vegetation in the databases available (Scopus, Web of Science, ScienceDirect, PubMed, Google Scholar). To be incorporated into our database, the following criteria were applied: (i) restoration was performed in areas that were open ecosystems (cerrado grasslands or open savannas) before land conversion or degradation occurred, (ii) the study presents at least a list of species introduced, and (iii) describes the site history and the restoration technique in sufficient detail. Techniques were excluded from the analysis if they were not applied in at least three separate study areas. Sites originally occupied by rupestrian and wet grasslands were also not included. Several studies collected data in the same restoration area, therefore, we did not consider more than one publication about the same restoration initiative. Details for each data set considered here are presented in Table S1 in the original paper. A lack of standardized sampling procedures among studies turned impossible to compare changes in plant community structure (e.g. ground cover, biomass or even species richness). Instead, we analyzed community composition in terms of species and growth forms, which could be obtained from a majority of the studies and is a good surrogate for vegetation structure and diversity. 

Data collection in the field: Studies presenting data from reference ecosystems or control plots were rare. Consequently, to assess restoration success relative to a reference ecosystem for these studies we sampled several pristine Cerrado open ecosystems representing the regions where the restoration interventions were implemented (see Figure S1). We, therefore, assumed that the proportions of target species and growth forms follow standards for conserved open ecosystems in the Cerrado and would provide a reliable comparison. Data collection for the 13 conserved areas was performed in four Brazilian States (São Paulo, Paraná, Mato Grosso do Sul and Goiás) across the Cerrado ecoregion  (Figure S1 and Table S1). We selected natural remnants free from exotic species and other sources of degradation; all areas were characterized as cerrado grassland (campo cerrado) – dominated by a rich herbaceous ground layer with few scattered shrubs and trees, with < 20% canopy cover. We focused on areas with < 20% canopy cover because these ecosystems are the most threatened and the hardest to recover (Durigan & Ratter 2006; Pilon et al., 2018, 2019).

For the restoration techniques to be sampled in the field, we selected: (i) two areas subject to direct seeding in 2016 and 2018 in the National Park of Chapada dos Veadeiros (PNCV, at Goías state); (ii) one area restored by grass tussock transplanting at Santa Ecological Station (EEcSB, at São Paulo state); and (iii) two areas where passive restoration has been conducted after eradication of former pine plantation at EEcSB and Ecological Station of Itirapina (also in São Paulo state) (Table S1). In each area, restoration sites and in undisturbed open ecosystems in the Cerrado, we performed a floristic characterization following the “quick survey” method. This method consists in carrying out at least three walks in a straight line across the vegetation 500 m apart from each other, sampling during regular time intervals the floristic composition (see details Walter & Guarino 2006). The walks were performed in homogeneous vegetation patches, and in our sites, up to three straight-line walks were necessary. In each straight line, sampling stopped when less than five species were added in two consecutive intervals of 15 minutes. The species lists for areas in natural regeneration after eucalypt plantation were obtained from 19 former eucalypt stands, differing in the time lag since eucalypt clearcutting (unpublished data) (Table S1 – from the paper). In October 2019, permanent plots were established in each area (221 plots in total, of 25 m x 4 m), resulting in 19 lists of plant species (all growth forms included). In addition, three unpublished lists from abandoned pastures of different ages were also added to our data set, collected in 15 permanent plots of 25 m x 4 m (details Table S1).