Fire season and drought influence fire effects on invasive grasses: A meta-analysis

Fire can shape plant communities when species respond differently to fire characteristics like season and intensity. If invasive plants are more vulnerable than native species to specific aspects of fire, managers could use prescribed fires to control non-native species. We conducted a meta-analysis of fire effects on six closely related Afro-Eurasian and Australian grasses (Bothriochloa bladhii, B. ischaemum, B. pertusa, Dichanthium annulatum, D. aristatum, and D. sericeum, collectively “invasive bluestems”) that have invaded grasslands worldwide. Using data from 31 studies (only 3 from their native range), we evaluated whether these grasses (275 effect sizes) responded differently than native grasses (184 effect sizes) to fire season, fuel load, and drought. Original data included 15 different response variables (e.g., biomass, survival) that were converted to standardized mean difference for analysis. Fires in summer, fall, and early winter had negative effects on invasive bluestems; no fire season had significant positive effects. Most data were for B. ischaemum, but the other bluestems may also be vulnerable to summer fire. Native grasses did not show significant negative responses in any month. Drought (Keetch-Byram Drought Index) in the month before fire increased the negative effects of fire on invasive bluestems but not native grasses. Drought after fire led to similar negative effects on both groups. Unexpectedly, fuel load (which influences fire intensity) did not significantly influence fire effects in any analysis. At the fuel loads examined (~600 – 10,000 kg/ha dried herbaceous biomass), fire intensity may have been too low to cause meristem mortality. Between-study heterogeneity was large in all analyses (I2>80%), suggesting that additional factors beyond those reported in the studies influence fire effects. These factors could include plant phenology, fire behavior, weather conditions during the fire, and soil characteristics.

Synthesis and applications:

Fires during summer and fall, especially during dry conditions, could harm invasive bluestems relative to native grasses, likely due to subtle differences in heat sensitivity, phenology, and drought resistance. Other invasive species may have similar vulnerabilities to specific fire seasons and rainfall conditions that allow the use of fire as a control method.

Methods
We searched Web of Science, Agricola, Proquest, and GoogleScholar on July 15, 2023 (Table S1). A separate search was conducted in each database for each of the six focal species (Bothriochloa bladhii, Bothriochloa ischaemum, Bothriochloa pertusa, Dichanthium annulatum, Dichanthium aristatum, and Dichanthium sericeum). The search term included scientific and common names as well as “fire OR burn*” (Table S2); search terms were the same for each database. After de-duplication, these initial searches identified 272 unique documents (including theses, dissertations, and unpublished reports).

Table S1: Details of databases searched

Database

Details

Web of Science

Searched ‘topic’. Web of Science Core Collection included Science Citation Index Expanded (SCI-EXPANDED—1900-present); Social Sciences Citation Index (SSCI)—1900-present; Arts & Humanities Citation Index (AHCI)—1975-present; Conference Proceedings Citation Index—Science (CPCI-S)—1900-present; Conference Proceedings Citation Index—Social Science & Humanities (CPCI-SSH)—1990-present; Book Citation Index—Science (BKCI-S)—2005-present; Book Citation Index—Social Sciences & Humanities (BKCI-SSH)—2005-present; Emerging Sources Citation Index (SCI)—2005-present; Current Chemical Reactions (CCR-EXPANDED)—1985-present; Index Chemicus (IC)—1993-present.

Agricola

1967 to present; searched ‘TX All Text Fields’

ProQuest Dissertations and Theses Global

1974 to present; searched ‘anywhere but full text-NOFT’

GoogleScholar

No date restrictions; used search strings with only scientific names (otherwise the search term was too long). Stopping point was chosen when there were 10 non-relevant search results on a page.

Table S2: Exact search terms used when searching the databases listed in Table 1.

(“Bothriochloa bladhii” OR “Bothriochloa caucasica” OR “Andropogon bladhii” OR “Bothriochloa intermedia” OR “Andropogon intermedius” OR “Andropogon caucasicus” OR “Caucasian bluestem” OR “Australian beard grass” OR “forest-bluegrass” OR “plains bluestem” OR “purple plume grass”) AND (fire OR burn*)

(“Bothriochloa ischaemum” OR “Amphilophis ischaemum” OR “Andropogon ischaemum” OR “Dichanthium ischaemum” OR “King Ranch bluestem” OR “KR bluestem” OR “yellow bluestem” OR “plains bluestem” OR “bearded finger grass” OR “dogstooth grass” OR “Turkestan bluestem”) AND (fire OR burn*)

(“Bothriochloa pertusa” OR “Andropogon pertusus” OR “Holcus pertusus” OR “Indian-bluegrass” OR “pitted beardgrass” OR “hurricane grass” OR “Indian couch grass” OR “Seymour grass” OR “Barbados sourgrass” OR “Antigua hay” OR “sweet pitted grass” OR “silver grass”) AND (fire OR burn*)

(“Dichanthium annulatum” OR “Andropogon annulatus” OR “Andropogon nodosus” OR “Kleberg bluestem” OR “marvel grass” OR “Diaz bluestem” OR “Hindi grass” OR “ringed dichanthium” OR “sheda grass” OR “medio bluestem” OR “jargu grass” OR “Delhi grass” OR “vuda bluegrass” OR “two-flowered golden-beard” OR “Santa Barbara grass”) AND (fire OR burn*)

(“Dichanthium aristatum” OR “Andropogon aristatus” OR “Angleton bluestem” OR “Angelton bluestem” OR “wildergrass”) AND (fire OR burn*)

(“Dichanthium sericeum” OR “Andropogon sericeus” OR “Silky bluestem” OR “Queensland blue grass” OR “silky bluegrass” OR “slender bluegrass” OR “tassel bluegrass”) AND (fire OR burn*)

Search results were stored in Rayyan (Ouzzani et al., 2016) and were screened manually by one of the authors. Papers were retained for further screening if they included prescribed fire or wildfire, at least one of the focal grass species (in the native or introduced range), and a measure of fire effects on the focal grass. Title and abstract screening eliminated 168 records.

Remaining records underwent full-text screening with additional criteria: studies included a comparison between burned and unburned treatments; no additional treatments (e.g., fertilizer, mowing) were applied to the measured populations; studies provided information needed for a quantitative meta-analysis (response means, sample size, and measures of variation); and studies provided the month or exact dates of the fires. Where the necessary data were not available in the papers, we attempted to contact the authors. Full-text screening eliminated 76 records.

We then conducted a forward/backward citation search based on the papers included after full-text screening, as well as articles that were themselves eliminated because data were combined across treatments (e.g., averaged across fertilizer levels, n = 20), no variance was reported (n = 8), or fire history was not available (n = 2). These searches were conducted in August 2023 and added three additional papers for a final total of 31 papers that met all search criteria.

We extracted data from tables, figures (using WebPlotDigitizer 4.6, Rohatgi 2022), published supplemental datasets, or data provided by the authors. We extracted means for burn and control (unburned) treatments along with sample sizes and standard deviation or standard error. Original data types included basal area, biomass, change in cover, change in frequency, cover, crown area, dead crown density, density, frequency, number of plants, number of seed heads, number of tillers, survival, and stem count. We extracted data for the focal grass species as well as any native grass species presented in the same papers. When multiple papers were published about the same study, we used the data from the most recent publication, but included additional data from earlier papers if they presented different information (e.g., cover vs biomass; more detailed treatment groupings), taking care to not duplicate data.

We compiled additional information to serve as moderators (defined in Table S3) including site, latitude, species name, species range of bluestems (native vs introduced), photosynthesis type of native grasses (C3 and C4 species according to Cerros-Tlatilpa et al. [2011] and Osborne et al. [2014]), seeding with native species (excluding focal bluestems), current grazing, study type (experimental, observational), fire type (prescribed fire, wildfire, burn box/burn barrel), time since fire, date/month of fire, and response type (e.g., frequency, cover). Because we found more sites in the northern hemisphere, fire months from the southern hemisphere were adjusted by adding 6 months, making them seasonally equivalent to northern hemisphere months (e.g., “July” is always summer). We also recorded fuel load (dried herbaceous biomass) and soil depth when available.

To examine the influence of drought on fire effects, we calculated the Keetch-Byram Drought Index (KBDI, Keetch, and Byram 1968; Alexander 1990) before and after each fire. This drought index represents the amount of rainfall needed to return the soil to saturation and changes daily based on temperature and rainfall. Values range from 0 (no moisture deficit) to 800. We used this drought index instead of rainfall because the index takes mean annual rainfall into account, allowing comparisons among regions with different climates. To calculate the index, we downloaded temperature and precipitation data from the closest weather station(s) to each study site from National Oceanic and Atmospheric Administration’s National Centers for Environmental Information (for the United States and Guam, National Centers for Environmental Information, 2023) or the Australian Bureau of Meteorology (for Australia, Australian Bureau of Meteorology 2023). For studies where the exact fire date was reported, we calculated the average drought index for 2, 6, and 10 weeks before and after the fire (as 6 separate variables). For studies where only the month of fire was reported, we calculated the average drought index for the month in which the fire happened, as well as for 1 and 2 months before and after the fire (creating 5 drought variables).

Continuous moderators (fuel load, soil depth, and the absolute value of latitude) were centered on their mean and scaled by their standard deviation across the whole dataset. The drought variables were centered on 400 (the mid-point of the Keetch-Byram Drought Index) and scaled from -1 to 1, keeping all drought variables on the same scale.

The adjusted standardized mean difference (SMD, Hedges’ g) was calculated for each pair of burned and unburned data as the effect size for meta-analysis (Hedges & Olkin, 1985). Negative SMD values indicate a negative effect of fire (e.g., lower cover in burned vs unburned plots). SMD was calculated using the “esc” package as SMD = (mean1 – mean2)/√(((N1-1)*sd12 + (N2-1)*sd12)/(N1+N2-2)), where N = sample size and sd = standard deviation (Ludecke, 2019). In a few cases, study data were proportions or counts so we first calculated odds ratio or binary proportion effect sizes and then converted those into SMD using the “esc” package.