1.2 Data preparation
Given that the soil was collected from different depth in the original studies, we treated soil depth as a continuous variable when performing statistical analysis and processed according to the following rules. At the same location (same latitude and longitude), if the amount of SMB came from different depths (such as 0-10, 10-20, 20-30 cm), we averaged them and the result was used as represent 0-30cm soil layer data. In one location, data of soil samples from the same soil layer would be averaged. If the soil depth was less than 30cm, we assumed that the measurements represent the top 0-30 cm soil profile.
For each selected study site, we collected the geographic location, ecosystem type, parameters of soil (soil organic carbon(SOC), soil pH, and soil sand fraction), SMBC and/or SMBN, if reported. Other data, such as AI, were collected in open access databases based on their geographic location (http://www.csi.cgiar.org). To explore the pattern, we classified AI into different categories. According to FAO guidelines, sites with AI < 0.65 were considered as drylands and further divided into drought subtypes of hyper-arid class (AI < 0.05), arid class (0.05 ≤ AI < 0.2), semi-arid class (0.2 ≤ AI < 0.5), and dry sub-humid class (0.5 ≤ AI < 0.65) (UNEP, 1992). In our study, the values of AI ranged from 0.0868 to 0.6477. Ecosystem types are mainly divided into four categories: desert, grassland, forest, and cropland. The classification standard is to look for keywords in the method of the references. If those data were not given in the references, we used the Harmonized World Soil Database V1.2 to extract such information based on the coordinates.