Forest resilience under global environmental change: Do we have the information we need? A systematic review

The capacity of forests to recover after disturbance, i.e., their resilience, determines their ability to persist and function over time. Many variables, natural and managerial, affect forest resilience. Thus, understanding their effects is critical for the development of sound forest conservation and management strategies, especially in the context of ongoing global environmental changes. We conducted a representative review, meta-analysis, of the forest literature in this topic (search terms “forest AND resilience”). We aimed to identify natural conditions that promote or jeopardize resilience, assess the efficacy of post-disturbance management practices on forest recovery, and evaluate forest resilience under current environmental changes. We surveyed more than 2,500 articles and selected the 156 studies (724 observations) that compared and quantified forest recovery after disturbance under different contexts. Context of recovery included: resource gradients (moisture and fertility), post-disturbance biomass reduction treatments, species richness gradients, incidence of a second disturbance, and disturbance severity. Metrics of recovery varied from individual tree growth and reproduction, to population abundance, to species richness and cover. Analyses show management practices only favored recovery through increased reproduction (seed production) and abundance of recruitment stages. Higher moisture conditions favored recovery, particularly in dry temperate regions; and in boreal forests, this positive effect increased with regional humidity. Biomass reduction treatments were only effective in increasing resilience after a drought. Early recruiting plant stages benefited from increased severity, while disturbance severity was associated with lower recovery of remaining adult trees. This quantitative review provides insight into the natural conditions and management practices under which forest resilience is enhanced and highlights conditions that could jeopardize future resilience. We also identified important knowledge gaps, such as the role of diversity in determining forest resilience and the lack of data in many regions.

This is also a relatively conservative estimate of ES, values varied between -2 and 2, while calculations of the natural log ratio varied between -5.2 and 6.4. The correlation between our ES estimates and those using the natural log ratio was 0.99 (based on 591 observations with ES values between -1.5 and 1.5). To estimate the mean and variance around the effect size we run simulations (equivalent to bootstrapping) using the reported mean and SD or SE values of the response metrics, we also included sample size in the calculations to give more weight to observations with larger sample sizes (Gurevitch & Hedges 2001).

Data analysis
In order to compare natural and managed systems and to differentiate between conditions that were expected to improve resilience (increases in moisture, fertility or diversity, and biomass reduction treatments) versus those that were expected to reduce it (increase in disturbance severity and occurrence of a second disturbance), we analyzed the calculated values of ES, mean (ES) and SD (s) using the following hierarchical approach. For each observation i, ESi was estimated for each system (natural or managed), context of recovery (moisture, fertility, diversity, biomass reduction, severity, second disturbance) and response metric (abundance, change, diversity, growth, reproduction, resilience). We also included study random effects, SRE. Likelihood: 7~( 7 , 7 ; ) Process model: 7 = 1 =>=?@A (7),BCD?@E? (7),F@=GCD=@(7) + 7 Parameter ES1 was then estimated for each combination of system and context of recovery, 1 =>=?@A,BCD?@E?,F@=GCD=@~J 2 =>=?@A,BCDL7?7CD , =>=?@A,BCDL7?7CD M.
We carried out additional analysis to determine what might have affected the variability we observed in ES. From the six contexts of recovery, we had enough data (more than 100 observations) to analyze three of them: moisture gradients, biomass reduction treatments and disturbance severity.
We analyzed the response to moisture gradients (data were only available for natural systems) as a function of the moisture levels of the study's region. The purpose of this approach was to understand if the effects of moisture gradients on resilience depended on the climatic conditions (moisture regime) of the region. For example, moisture gradients may play a more important role on resilience in dry regions than in wet areas. For that we calculated a variant of the De Martonne humidity-aridity index (DMI) that includes temperature and precipitation, DMI: annual precipitation (cm)/July average temperature (°C) (January for southern hemisphere locations) (De Martonne 1926). We also included study random effects (SRE) and years since disturbance (YSD) as the magnitude in ES may vary over time. Since data exploration indicated that responses might vary by biome, parameters were estimated for each biome represented in the data (boreal, temperate, Mediterranean and tropical). DMI and YSD were standardized for the analysis.
Likelihood: 7~( 7 , 7 ; ) Process model: 7 = L7=?VFOWDB@(7) + 7 Finally, we analyzed whether severity of the disturbance affected vegetation responses. We analyzed the available data (for natural and managed systems) and estimated ES as a function of the treatment size (TS; a measurement of severity strength). TS was estimated using the same approach as ES ( = (?F@W?A@D?YBCD?FCZ)

WO=(W[@FW\@)
). Since severity seemed to affect vegetation strata differently (known from our exploratory data analysis), we estimated the effect of TS on ES as a function of vegetation stratum (only 'all strata', 'adult trees' and 'seedlings' categories had enough data, others had much fewer data points, < 6). Study random effects were also added and centered to aid with the convergence of parameters.