Soil erosion control and water resource protection can closely interact during restoration of terrestrial ecosystems. In semi‐arid ecosystems, an urgent issue is how vegetation restoration can achieve the goal of soil erosion mitigation and water conservation, which in turn, feeds back to ecosystem functioning.

We reviewed 78 articles from 22 countries in semi‐arid areas to evaluate the effects of vegetation type (i.e. forest, grassland and scrubland) on runoff and sediment yields across different environmental conditions (i.e. vegetation coverage, rainfall intensity, slope gradient and soil texture).

Our meta‐analysis shows that runoff and sediment reduction both increased as the vegetation coverage increased, and tended to be stable when vegetation coverage exceeded 60%. Vegetation provided a greater benefit for sediment reduction than for runoff control under intense rainfall. Grasslands were generally more effective in reducing sediment than other vegetation types. Forests, grasslands and scrublands were most efficient in soil erosion control on 20°–30°, 0°–25° and 10°–25° slopes respectively. Grasslands and scrublands generally performed better with respect to soil erosion control on moderately coarse soils, whereas forests were most effective on medium‐textured and moderately fine soils.

Synthesis and applications. Effective restoration and soil erosion control in semi‐arid ecosystems strongly depends on the selection of vegetation type. Our study further indicates that, for land managers, it is critical to consider local slope, and soil texture, and maintain appropriate vegetation coverage to achieve ecosystem sustainability. Grasslands might be particularly suitable to optimize the trade‐off between soil erosion control and surface water resource in semi‐arid regions.

Data collection and compilation

To compare the efficiency of vegetation in controlling soil erosion, we used the Web of Science and Google Scholar search engines to identify peer‐reviewed journal articles on these topics between 1965 and 2017. The search included either “runoff”, “sediment” or “soil erosion”, and either “semi‐arid” or “semiarid”, resulting in 4,625 publications. To avoid publication bias, the primary criteria for the data compilation were as follows: (a) the aridity index: precipitation (mm)/potential evapotranspiration (mm) at the study site must be <0.5; (b) at least one of the relevant vegetation types (i.e. forest, grassland and scrubland) and a control (bare land or farmland) representing soil erosion without vegetation; (c) the same variables (runoff or sediment) were measured in both the control and vegetated plots; and (d) the mean values, number of replicates and variances for each data pair were reported. In total, 78 publications fulfilled these criteria, and a list of data sources used in the study is provided in the Data sources section (see Appendix Table S1; Wu et al., 2020).

From each study, the mean values of the soil erosion response variables and their corresponding variance and the number of replicates were extracted directly from the text or tables, or from figures using GetData Graph Digitizer 2.24 software or estimated from the median, range and sample size (Hozo, Djulbegovic, & Hozo,2005 ). Other data, such as mean annual rainfall, mean annual temperature, mean annual evapotranspiration, vegetation coverage, rainfall intensity, study scale, slope and soil texture, were also extracted from the publications. Other factors affecting soil erosion processes, such as the amount of rainfall, were not used in this meta‐analysis due to an insufficient number of papers reporting these variables.

Soil erosion and runoff were measured primarily using natural and simulated rainfall, but four studies involved model estimation and fallout radionuclide techniques. The study scale ranged from 0.0625 m2 to 16.8 ha, with most studies focussing on areas smaller than 100 m2. The mean annual temperatures of the study sites included in the meta‐analysis ranged from 6.3 to 29.7°C. The mean annual rainfall ranged from 180 to 970 mm, and the mean annual evapotranspiration from 290 to 2,500 mm. Most studies were conducted in China and Spain; other areas represented in this meta‐analysis included the USA and Australia (Figure 1 ). Little information was available for large areas of Africa and South America. However, we believe that our results are applicable to most semi‐arid regions because of the inclusion of data from varied sources.

The vegetation types included forests, grasslands and scrublands as reported in the original study (Table 1). Forest refers to natural vegetation or plantations with a dominance of trees (canopy density > 20%). Grassland refers to artificial grassland or natural grassland. Scrubland refers to a plant community characterized by vegetation dominated by shrubs, often also including grasses and forbs. Soil texture was classified into five categories (coarse, moderately coarse, medium, moderately fine and fine) according to the Soil Science Division Staff (2017) and Elias et al. (2018). The slope gradients were divided into the following eight categories: 0°–5°, 5°–10°, 10°–15°, 15°–20°, 20°–25°, 25°–30°, 30°–35°, and >35°.