These files are associated with,
Hansen, W.D., N.B Schwartz, A.P. Williams, K. Albrich, L.M. Kueppers, A. Rammig, C.P.O. Reyer, A.C. Staver, and R. Seidl. 2022. Global forests are influenced by legacies of past inter-annual temperature variability. Environmental Research:Ecology 1, 011001.
Inter-annual climate variability (hereafter climate variability) is increasing in many forested regions with climate change. This variability could have larger near-term impacts on forests than decadal shifts in mean climate, but how forests will respond remains poorly resolved, particularly at broad scales. Individual trees and even forest communities often have traits and ecological strategies – the legacies of exposure to past variable conditions – that confer tolerance to subsequent climate variability. However, whether local legacies also shape global forest responses is unknown. Our objective was to assess how past and current climate variability influences global forest productivity. We hypothesized forests exposed to large climate variability in the past would better tolerate current climate variability than forests where past climate was relatively stable. We used historical (1950-1969) and contemporary (2000-2019) temperature, precipitation, and vapor pressure deficit and the remotely-sensed Enhanced Vegetation Index to quantify how historical and contemporary climate variability relate to patterns of contemporary forest productivity. Consistent with our hypothesis, forests exposed to large temperature variability in the past were more tolerant of contemporary temperature variability than forests where past temperatures were less variable. Forests were 19-fold less sensitive to contemporary temperature variability where historical inter-annual temperature variability was 0.66°C (2 standard deviations) greater than the global average historical temperature variability. We also found that larger increases in temperature variability between the two study periods often eroded the tolerance conferred by legacy effects of historical temperature variability. However, the hypothesis was not supported in the case of precipitation and vapor pressure deficit variability, potentially due to physiological tradeoffs inherent in how trees cope with dry conditions. We conclude that sensitivity of forest productivity to imminent increases in temperature variability may be partially predictable based on legacies of past conditions.
The Cary Institute of Ecosystem Studies furnishes data under the following conditions: The data have received quality assurance scrutiny, and, although we are confident of the accuracy of these data, Cary Institute will not be held liable for errors in these data. Data are subject to change resulting from updates in data screening or models used. To cite these data, click on the Cite button on this page.
Time periodBegin date: 06/01/2019
End date: 01/01/2022
Monthly mean, minimum, and maximum temperature (°C), mean vapor pressure (hPa), and total precipitation (mm) (record length 1950-2019; product CRU TS v. 4.04) gridded at a 0.5° spatial resolution were retrieved from the Climate Research Unit (CRU), University of East Anglia. The CRU TS 4.04 dataset is interpolated based on a global network of weather stations. We used monthly minimum and maximum temperature to calculate vapor pressure at saturation and then subtracted actual vapor pressure to derive monthly mean vapor pressure deficit (VPD). CRU TS is the only interpolated weather-station based product that provides global measurements of temperature, precipitation, and vapor pressure at a relatively fine spatial resolution for the last several decades. This was essential for our analysis, as we were interested in the concurrent influence of multiple climate variables on forest productivity.
We used climate records for two study periods; a historical period (1950-1969) and a contemporary period (2000-2019). The contemporary period was selected to correspond with the MODIS satellite record. Two criteria determined selection of the historical period. Because trees are long lived, we wanted to maximize time between the two study periods to ensure that potential legacies of historical climate variability (e.g., trait plasticity and shifts in community composition) would have time to manifest. However, the global density of weather stations was much lower in the first half of the 20th century than in the second half, which leads to large uncertainty regarding spatial and temporal patterns of inter-annual climate variability prior to 1950 in the CRU data.
We aggregated temperature and VPD climate records from monthly to annual values by selecting the month of each year with the warmest mean temperature and with the highest mean VPD to capture growing conditions without prescribing changes in growing season across latitude. For precipitation, which can be stored in snow and the subsurface for months, we quantified each year’s annual total. We then expressed each annual precipitation total as a relatively anomaly (a percentage) of the respective 20-year mean annual total for the two study periods. This was done because precipitation variance tends to scale with mean precipitation, where wetter places also inherently have more inter-annual variability in precipitation totals. As we were interested in effects of climate variability separate from long-term mean trends, climate records were linearly detrended for each grid cell and each time period. We then quantified inter-annual variability of temperature, VPD, and relative precipitation anomalies during each period using the standard deviation (SD). We also calculated the change in inter-annual temperature, precipitation, and VPD variability between the two periods (hereafter referred to as the late 20th-century change in climate variability).
Monthly mean EVI (record length 2000-2019; product MOD13C2 V006) and the University of Maryland annual gridded land cover classification for the year 2019 (Product MOD12C1 V006) were retrieved from the United States Geological Survey Land Processes Distributed Active Archive Center (LPDAAC). EVI and land cover were gridded at a 0.05° spatial resolution, derived from 1-km2 MODIS pixels. Off-nadir viewing angles have been shown to bias patterns of seasonal and inter-annual EVI. The MOD13C2 EVI product minimizes effects of viewing angle in the compositing algorithm by selecting the highest-quality constituent 1-km2 MODIS pixels with the lowest viewing angle for each 0.05° grid cell. To ensure results were not an artifact of sensor viewing angle, we took a conservative approach by including only the 0.05° grid cells where most (≥ 50%) of the constituent 1-km2 pixels had nadir viewing angles (< 30°). Grid cells are also assigned the historical mean EVI when the satellite observations are of insufficient quality (missing data, cloudy, etc.). Thus, we only retained 0.05° grid cells assigned the top quality flag of “good: use with confidence”.
We masked monthly EVI to include only forested grids cells, which we defined as any cell where the forest types in the land cover classification (evergreen needleleaf, evergreen broadleaf, deciduous needleleaf, deciduous broadleaf, mixed forest) summed to 80%. In the landcover product, forests were delineated based on a canopy height > 2m and > 60% tree cover in 2019. We chose 2019 to ensure we did not include grid cells that experienced a severe natural disturbance or deforestation during the contemporary period. We categorized forest grid cells as boreal, temperate, Mediterranean, or tropical using The World Wildlife Federation’s Terrestrial Ecoregions of the World.
For each grid cell that had six or more months of EVI observations in ten or more years, we aggregated from monthly to annual values by selecting the month with the highest mean EVI, following the treatment of the climate data. All grid cells that did not meet the six months of observations in ten or more years were excluded from analysis. We linearly detrended annual EVI in each grid cell and quantified inter-annual variability in EVI as the SD. We aggregated the EVI grid from 0.05° to the 0.5° resolution of the climate data. This final dataset of inter-annual variability in EVI included 7,477 0.5° grid cells. We also calculated the mean percent cover of evergreen, deciduous, broadleaf, and needleleaf in each 0.5° grid cell and converted these to categorical variables of leaf shape (one when % needleleaf exceeded % broadleaf, else zero) and leaf habit (one when % evergreen exceeded % deciduous, else zero) to avoid zero inflation of using the continuous percent vegetation-type variables in regressions.
We first assessed how contemporary climate variability influenced forest productivity globally and within biomes. We fit a linear regression that predicted the SD(EVI2000-2019) as a function of SD(temperature2000-2019), SD(precipitation2000-2019), and SD(VPD2000-2019), leaf shape, leaf habit, and biome. Based on visual inspection of univariate scatter plots, we included a quadratic form of the climate variables to allow for nonlinear relationships. To ensure normality and homoscedasticity, the response variable was transformed using a box-cox transformation. We also evaluated collinearity among predictor variables using a variance inflation factor cutoff of seven. Response and non-categorical predictor variables were centered and scaled (converted to z-scores). We used exhaustive model selection with a maximum of four terms and selected the best fitting model based on AIC. Regression residuals were spatially autocorrelated (Moran’s I < 0.05). Thus, we also ran spatial error and spatial auto-regressive linear models to confirm the direction and statistical significance of relationships when spatial autocorrelation was accounted for.
We also wanted to assess how historical climate variability contributed to patterns of contemporary forest productivity, but historical and contemporary climate variability could not be included in the same model due to collinearity. However, model residuals (differences between observed values of the dependent variable and the values predicted by regression) can contain ecologically relevant information, and thus, we used the model residuals from the contemporary climate variability regression described previously as a metric of the remaining sensitivity of forest productivity for subsequent analysis (hereafter forest sensitivity). This metric identifies forest grid cells where forest productivity was unusually stable (large negative residuals) and unusually variable (larger positive residuals) relative to grid cells of similar forest types that experienced similar contemporary climate variability.
We fit a linear regression to evaluate whether and how this remaining forest sensitivity was explained by historical climate variability; SD(temperature1950-1969), SD(precipitation1950-1969), SD(VPD1950-1969), the late 20th-century change in climate variability; Δ SD(temperature), Δ SD(precipitation), Δ SD(VPD), and interactions between historical climate variability and the late 20th-century change in climate variability; SD(temperature; 1950-1969)* Δ SD(temperature), SD(precipitation; 1950-1969)* Δ SD(precipitation), SD(VPD; 1950-1969)* Δ SD(VPD). Again, normality, homoscedasticity, and collinearity were assessed, continuous variables were centered and scaled, quadratic terms were included, model selection was conducted, and the most parsimonious model was selected based on AIC. Spatial error and autoregressive models were also run.
We repeated analyses where inter-annual variability of EVI was quantified as the coefficient of variation (CV), instead of the SD. The CV is a relative rather than absolute measure of variability, allowing us to explore whether our results were sensitive to the property that cells with high mean EVI may inherently have more variable EVI. We also repeated analyses with the normalized difference vegetation index (NDVI) instead of EVI. However, EVI is preferable over NDVI because it is less likely to saturate in forests with high biomass, and our analyses confirmed saturation of NDVI in tropical forests. Finally, because the historical period is relatively short for characterizing climate variability, we repeated analyses with a longer historical window of 1950-1989. Analyses were conducted in R version 4.0.5, using the packages ncdf4, gdalUtils, raster, tidyverse, rgdal, sf, broom, spdep, spatialreg, MASS, car, MuMIn, and RStoolbox.
Data provenanceHarris, I., Osborn, T.J., Jones, P. et al. Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset. Sci Data 7, 109 (2020). https://doi.org/10.1038/s41597-020-0453-3
Friedl, M., D. Sulla-Menashe. MCD12C1 MODIS/Terra+Aqua Land Cover Type Yearly L3 Global 0.05Deg CMG V006. 2015, distributed by NASA EOSDIS Land Processes DAAC, https://doi.org/10.5067/MODIS/MCD12C1.006. Accessed 2022-04-06
Didan, K.. MOD13C2 MODIS/Terra Vegetation Indices Monthly L3 Global 0.05Deg CMG V006. 2015, distributed by NASA EOSDIS Land Processes DAAC, https://doi.org/10.5067/MODIS/MOD13C2.006. Accessed 2022-04-06.
Investigator 1Winslow Hansen