The association with mycorrhizal fungi, predominantly arbuscular mycorrhizal (AM) fungi or ectomycorrhizal (EM) fungi, is a critical strategy for trees to cope with drought stress, one of the most pervasive stressors affecting forest dynamics. Although experimental evidence suggests varying drought responses among trees with different mycorrhizal associations, a quantitative large-scale synthesis is still lacking. In this study, we compiled global datasets encompassing three types of drought responses (drought-induced mortality, growth reduction by severe droughts, and growth recovery from severe droughts) and tested the differences in these drought responses between the two mycorrhizal types using spatial autoregressive models. To elucidate the significant variations in drought responses, we identified key influential climatic factors, edaphic (soil-related) factors, and species-specific traits using model selection based on Akaike information criterion. Globally, EM trees experienced slightly greater drought-induced mortality compared to AM trees but showed significantly less pronounced growth reduction and recovery following severe droughts. The drought responses of EM trees were more strongly influenced by environmental stresses and size-related traits than those of AM trees, with higher soil pH, greater climate seasonality, and lower plant height significantly increasing the drought responses of EM trees. Despite the above-mentioned apparent differences, drought responses were generally more drastic at warmer and drier sites for species producing heavier seeds regardless of mycorrhizal type.

Drought-response dataset compilation

Global datasets of three aspects of tree drought responses were compiled: drought-induced mortality, growth reduction by severe drought and growth recovery from severe drought.

In the current study, we defined drought from an ecological perspective. That is, droughts are the events in which the water supply from precipitation falls below plant water demand. Therefore, the standardized precipitation evapotranspiration index (SPEI) is an ideal drought metric for our study. SPEI is a site-specific drought indicator that involves both precipitation and plant potential evapotranspiration (Vicente-Serrano et al. 2010; Beguería et al. 2022). It has been widely used in ecological studies and has demonstrated superior capability to capture drought impact at the global scale than the conventional Palmer drought indices (PDIs). Additionally, SPEI has exhibited better performance in capturing ecosystems’ responses to summer droughts than the widely used standardized precipitation index (SPI) (Vicente-Serrano et al. 2012). Because mortality data and growth-response data were all collected based on calendar years (see details below), we defined a drought year as a calendar year with an SPEI smaller than 0, while a severe drought year was a calendar year with an SPEI smaller than -1.5 in accordance with previous literature (Haile et al. 2020; Ma et al.2020; Beguería et al. 2022).

Data on drought-induced mortality were mainly based on the global dataset compiled by Greenwood et al. (2017), which included the drought-induced annual mortality rate for 257 species from 28 studies across the world. Additionally, we included other drought-induced mortality data compiled for global studies (Anderegg et al. 2016; Caudullo & Barredo 2019; Hammond et al. 2022). Because our goal was to explore the differences between AM and EM trees, we determined the mycorrhizal type of each species based on the FungalRoot Database (Soudzilovskaia et al., 2020). For the species that had multiple observation records, all possible mycorrhizal types were recorded and combined. When the mycorrhizal type was undetermined or unknown at the species level, we also referred to the recommended mycorrhizal status for plant genera (Soudzilovskaia et al. 2022). All species that were determined to be associated with neither AM or EM fungi or dual AM-EM species were removed from our dataset. Additionally, because drought intensity strongly covaried with tree mortality (Greenwood et al. 2017), we downloaded the SPEI from 1901 to 2018 at a 12-month timescale from SPEIbase v.2.7 (https://spei.csic.es/spei_database) (Beguería et al. 2022) and extracted the SPEI of the drought years within the period that mortality data were collected (Michna & Woods 2020). We removed the mortality data that were collected during a period without a drought year (i.e., SPEI < 0). As a result, we obtained a dataset of 490 drought-induced mortality entries from 43 studies for 359 species (i.e., 301 AM compared with 58 EM species), which were collected from 46 different locations across the world (Figure 1a,b). See the pipeline for compilation of mortality dataset in Figure S1.

Data on growth reduction by severe drought and growth recovery from severe drought were calculated using ring width data and SPEI. We first determined the mycorrhizal type for each species listed in the International Tree-Ring Data Bank (ITRDB) (https://www.ncdc.noaa.gov/data-access/paleoclimatology-data/datasets/tree-ring) and downloaded the total ring width of all AM and EM species that had been documented by 3 or more studies to ensure a considerable spatial coverage of each species. Similar to the mortality dataset, mycorrhizal type was determined by the FungalRoot Database (Soudzilovskaia et al. 2022). Because tree growth always tends to decline with age (Gower et al. 1996), all ring width series were detrended to remove nonclimatic signals before calculating growth responses to drought (Yang et al. 2017). Among all commonly used detrending methods, the two methods based on biological growth models (i.e., the modified negative exponential curve and the modified Hugershoff curve) were chosen because they do not remove much of the climate signal or require sufficiently large samples from the same study site (Fang et al. 2010). Ring width detrending was performed in R version 4.1.2 using the package “dplR” (Bunn 2008, 2010; Bunn et al. 2020; R Core Team 2020). Both methods produced similar results, but the negative exponential curve produced some invalid, negative ring widths and was thus excluded from further analyses. For each study site, growth responses to drought were calculated using the averaged detrended ring widths of all core samples. To quantify growth responses to severe droughts, we first identified the years when severe droughts occurred at each site, defined as years with a Standardized Precipitation Evapotranspiration Index (SPEI) less than -1.5. Similar to the mortality dataset, the SPEI data were downloaded at a 12-month timescale from SPEIbase v.2.7. Given that tree growth can vary significantly across years and there is often a lag between a drought event and subsequent growth responses (Greenwood et al. 2017), we compared the average ring widths before and after the drought events. Specifically, we compared the average ring widths of one, three, or five year(s) before the drought with those of the one-, three-, or five-year periods afterwards. We determined the time scale that produced the largest number of significant reductions in growth across the entire dataset. Similarly, after identifying the best time scale for growth reduction quantification, we assessed the recovery phase. We compared the averaged ring widths of one, three or five year(s) after the period of growth reduction with averaged ring widths during the period of reduction. Then, wedetermined the time scale that produced the largest number of significant recoveries for the whole dataset. Because of the potentially low frequency of severe droughts at some sites that may result in small sample sizes and nonnormal data distributions, the nonparametric Wilcoxon rank-sum test (also known as the Mann-Whitney U test) was conducted using the R package “rstatix” (Kassambara 2021). To account for different frequencies of severe droughts across sites, the effect sizes of the Wilcoxon test (ranging between 0 and 1) were used. According to our analyses, the best time scale for quantifying growth reduction was three years before drought versus the year of drought, while the best time scale for quantifying growth recovery was five years after drought versus the year of drought (Table S1). In other words, most tree individuals in our dataset exhibited significant growth reduction immediately after a severe drought, whereas most individuals exhibited significant growth recovery after a time lag of four to five years. As a result, we obtained a dataset of 2876 growth-response entries for 119 species (i.e., 34 AM compared with 85 EM species) from 2446 different locations across the world (Figure 1b,c). See the pipeline for compilation of growth-performance dataset in Figure S1.

Environmental data collection

For climate factors, we downloaded historical monthly temperature and precipitation between 1901 and 2022 from Climate Research Union (version 4.07) (https://crudata.uea.ac.uk/cru/data/hrg/index.htm#current) (Harris et al. 2020) and quantified the mean annual temperature (MAT) and mean annual precipitation (MAP) for each geographic location. In addition to mean values, variations in temperature and precipitation may also be important driving factors that function in a disturbance-like manner. Therefore, we further quantified temperature and precipitation seasonality by their annual standard deviations and averaged these standard deviations across years.

For edaphic (soil-related) factors, we downloaded data of soil properties, including soil pH, total carbon content (TC), total nitrogen content (TN), and total phosphorous content (TP), from the Global Soil Dataset for Use in Earth System Models (GSDE) (Shangguan et al. 2014). The dataset included edaphic data of eight vertical layers to the depth of 2.3 m. We used the mean value across all eight layers to represent the reference status of each soil property at each study site. See the pipeline for compilation of climatic and edaphic dataset in Figure S1.

Trait data collection

According to previous studies, specific leaf area (Valladares & Sánchez-Gómez 2006; Greenwood et al. 2017), plant size (Bennett et al., 2015; Mcdowell & Allen, 2015), rooting depth (Irvine et al. 2002), stomatal conductance(Medrano et al. 2002; Zhu et al. 2018) and seed mass (Lazarus et al. 2018) are important predictors of tree drought responses. Thus, data on these traits were collected in addition to mycorrhizal type so that we could determine the key traits that drive the drought response patterns influenced by mycorrhizal type.

For the 453 tree species included in our drought response datasets, we requested public data of specific leaf area (SLA), maximum plant height, rooting depth, stomatal conductance, and seed mass from the TRY trait database (request ID 21286) (Kattge et al. 2020) and obtained the median values for all available traits for each study species. If a species did not have any available data but belonged to a genus containing available data for more than five different species, we extrapolated the genus-median trait value and used it as a reference value for the focal species. As a result, we obtained trait values of SLA, maximum height, rooting depth, stomatal conductance, and seed mass for 397, 349, 345, 237 and 348 species, respectively. This resulted in a total of 214 and 2337 observations with available data for all study traits for the mortality and growth-response datasets, respectively. See the pipeline for trait dataset compilation in Figure S1.