Plant trees for the planet: the potential of forests for climate change mitigation and the major drivers of national forest area

Abstract

Forests are one of the most cost-effective ways to sequester carbon today. Here, I estimate the world’s land share under forests required to prevent dangerous climate change. For this, I combine newest longitudinal data of FLUXNET on forests’ net ecosystem exchange of carbon (NEE) from 78 forest sites (N = 607) with countries’ mean temperature and forest area. This straightforward approach indicates that the world’s forests sequester 8.3 GtCO₂year⁻¹. For the 2 °C climate target, the current forest land share has to be doubled to 60.0% to sequester an additional 7.8 GtCO₂year⁻¹, which demands less red meat consumption. This afforestation/reforestation (AR) challenge is achievable, as the estimated global biophysical potential of AR is 8.0 GtCO₂year⁻¹ safeguarding food supply for 10 billion people. Climate-responsible countries have the highest AR potential. For effective climate policies, knowledge on the major drivers of forest area is crucial. Enhancing information here, I analyze forest land share data of 98 countries from 1990 to 2015 applying causal inference (N = 2494). The results highlight that population growth, industrialization, and increasing temperature reduce forest land share, while more protected forest and economic growth generally increase it. In all, this study confirms the potential of AR for climate change mitigation with a straightforward approach based on the direct measurement of NEE. This might provide a more valid picture given the shortcomings of indirect carbon stock-based inventories. The analysis identifies future regional hotspots for the AR potential and informs the need for fast and forceful action to prevent dangerous climate change.

Appendix

1.1 Methods and materials

1.1.1 Global and regional forest carbon sink

To assess the net ecosystem exchange of carbon (NEE) of countries’ forests, I use the newest available direct measurements of NEE of 78 micrometeorological measurement towers located in forests of 16 countries from 2000 to 2014 provided by FLUXNET (NASA 2015). See Table S1 in Supplementary Figures and Tables for an overview of the analyzed tower sites. FLUXNET sites collect data on the exchanges of CO₂ between forests and the atmosphere, precipitation, and air temperature at least in a 30-min interval. Table S3 provides a summary of the descriptive statistics. The tower sites use eddy covariance methods to measure forests’ NEE. The unique dataset utilized here, “FLUXNET2015”, provides standardized values for these characteristics and underwent several quality control tests and gap-filling (Pastorello et al. 2017).

To infer the NEE of countries’ forests from these 78 FLUXNET sites, I apply a straightforward approach consisting of three steps: Firstly, I regress their annual NEE on several site characteristics (average temperature, average temperature squared, precipitation, latitude, and elevation) controlling for overall time-trends by including dummy variables of the years observed. While primarily interested in the variation between the forest sites, the inclusion of the 607 site-years available for this model minimize the influence of a specific observation period stemming from annual variation in climatic and other conditions. Therefore, all standard errors are clustered by tower site to ensure robustness with respect to heteroscedasticity and autocorrelation. The results of this linear ordinary least squares (OLS) regression model (Table S2) indicate that only average temperature substantially relates to NEE. As Fig. S1 shows, the temperature-NEE relationship of forests follows a u-shaped pattern. Forests with an annual mean temperature of − 5 to 0 °C are net emitters of carbon, whereas the carbon sequestration of forests is highest in climatic domains with an average of about 15 °C. Even higher temperatures are associated with lower sequestration. Note that uncertainty between 15 and 26 °C is relatively high, because of a rather limited number of tower sites in this climatic forest domain. The reported regression results of Table S2 were tested for robustness: First, the model was rerun excluding one measurement tower each time from the regression. Second, all parameters were tested for linearity including a penalized splines fixed-effects (FE) regression model (Ruppert et al. 2003). Furthermore, the robustness of standard errors was investigated via non-parametric bootstrapping. None of these checks had any substantial influence on the estimates. In addition, the robustness of all estimates with respect to model specification was assessed using the procedure suggested by Young and Holsteen (2017). The potential influence of omitted variables was examined using the method suggested by Frank (2000). Also, these checks detected no fundamental deviations from the reported results. The analyses were conducted using the statistical software package STATA 15.1.

Secondly, I predict the mean annual sequestration between the years 2000 and 2014 (t) of country i’s forests in tons CO₂ per hectare (y_i) from model 1 of Table S2 according to the following formula:

$$ {y}_i=\frac{1}{T}{\sum}_{t=1}^T\left({\beta}_0+{\beta}_1{a}_{it}+{\beta}_2{a}_{it}^2+{\beta}_3{b}_i+{\gamma}_t\right) $$

(1)

β₀ represents the model intercept. a_it stands for the average air temperature of country i in year t, β₁ for the regression coefficient of the sites’ average temperature, and β₂ for the coefficient of its square. b_i denotes country i’s centroid’s latitude, and β₃ the regression coefficient for the forest sites’ latitude. γ_t represents the regression coefficient for year t. With β₀=5.60, β₁ = − 2.20, β₂ = 0.07, β₃ = − 0.06 from model 1 of Table S2 follows:

$$ {y}_i=\frac{1}{T}{\sum}_{t=1}^T\left(5.60-2.20{a}_{it}+0.07{a}_{it}^2-0.06{b}_i+{\beta}_t\right) $$

(2)

Data for a_it is taken from the Climate Change Knowledge Portal of the World Bank (2018); Table S6), and from the Country Geography Database of Portland State University (2018) for b_i. Computation of Eq. 2 yields a global average of − 8.8 tCO₂ha⁻¹ year⁻¹ (median = − 9.2, sd. = 4.8, min = − 15.1, max = 16.3) sequestered by forests in 2015. With roughly 2.7 trillion trees (Crowther et al. 2015) in the 40.0 Mkm² (FAO 2018) of forests worldwide, this translates into a mean of − 8.8 kgCO₂year⁻¹ per tree (tropical forests (latitude 0° to < 25° North (N) or South (S)) − 8.4, temperate forests (25° to < 50° N or S) − 17.3, boreal forests (≥ 50° N): − 1.9) as weighted by the share of trees by forest type (tropical 0.48, temperate 0.24, boreal 0.27; Crowther et al. 2015).

Thirdly, simply multiplying countries’ average NEE per hectare by their forest area gathered from the FAO (2018) gives countries’ forest carbon sink. Summing up yields an estimate for the global forest carbon sequestration of − 8.3 GtCO₂year⁻¹ or − 1.1 tCO₂year⁻¹ per capita (p.c.; UNPD 2017).

1.1.2 Afforestation/reforestation (AR) scenarios

To prevent dangerous climate change, the required and achievable forest land share of three different AR scenarios is developed. The basis for these scenarios is the 2 °C target and the associated remaining carbon budget until 2100. With the Paris Climate Agreement, the world community has agreed upon the limitation of global warming to well below 2 °C relative to preindustrial levels (UNFCCC 2015). A maximum of 2 °C of warming until 2100 may provide a relatively safe operating space for humanity and prevent dangerous climate change alongside a lock-in of a “Hothouse Earth” pathway with potentially hazardous consequences for ecosystems and human socio-economic systems (IPCC 2014; Steffen et al. 2018; Fischer et al. 2018; Rockström et al. 2009). Yet, humanity allegedly has already committed to 1.3 °C of warming (Mauritsen and Pincus 2017). Hence, limiting global warming to 1.5 °C and presumably providing an even safer operating space (IPCC 2018) seems out of reach (Raftery et al. 2017). Global CO₂ emissions of fossil fuel use and industrial processes have risen to 35.8 GtCO₂ or 4.8 tCO₂ per capita (p.c.) in 2016 (Janssens-Maenhout et al. 2017). This surpasses the global annual gross carbon budget (an estimated 30 GtCO₂) to fulfill the 2 °C target with a probability of at least 66% (IPCC 2014; Friedlingstein et al. 2014; Meinshausen et al. 2009). Assuming an average annual world population of 9.8 billion people until 2100 (UNPD 2017), this goal translates into ~ 3 t of gross CO₂ emissions p.c. and year.

Scenario 1 is the baseline assuming business-as-usual production and consumption patterns, constant other carbon sinks, further required emission reductions of 1.0 tCO₂year⁻¹ p.c. after accounting for the overall forest carbon sequestration of 0.8 tCO₂year⁻¹ p.c. with an expected average population of 9.8 billion people per year until 2100. Hence, the required additional absorption by forests for the 2 °C respectively the 3 t p.c. target is 9.8 GtCO₂year⁻¹. Assuming similar carbon sequestration of established forests and afforested/reforested land, simple solution of the rule of three and addition to the existing forest area (40.0 Mkm²) delivers a required forest area of 88.0 Mkm². With a global land area of 129.7 Mkm² (FAO 2018), this corresponds to a forest land share of 67.8% necessary to reach the 2 °C target with AR activities alone. This implicitly assumes similar tree density, species, species richness, and forest health of afforested/reforested land and established forests. To quantify the land area suitable for AR, land unsuitable for near-term and cost-efficient AR was excluded. These land cover types are artificial surfaces (including urban and associated areas), permanent snow and glaciers, terrestrial barren land, and sparsely natural vegetated areas as quantified by the FAO (2018). One hundred percent AR of all shrub-covered areas and herbaceous vegetation (18.1 Mkm²) result in an achievable 44.8% of forest land share in this scenario.

Scenario 2 further assumes current diets and an associated demand of agricultural land of 2100 m² p.c. (Hallström et al. 2015). As there are 3770 m² p.c. of agricultural land currently available, 44% of permanent grassland and cropland (FAO 2018) can be additionally afforested/reforested (16.4 Mkm²) for feeding an expected 9.8 billion people per year. Hence, a forest land share of 57.5% can be realized in scenario 2 (77.6 Mkm²). This accounts for more than two thirds of the AR climate target outlined in scenario 1.

To achieve the AR target fully, further reduction in the demand for agricultural land is required. In scenario 3 healthier diets with reduced red and ruminant meat consumption further decrease agricultural land demand by 28.0% to 1510 m² p.c. while dietary-related emissions decrease by 0.2 tCO₂yr⁻¹ p.c. (Table 1 in Hallström et al. 2015). Hence, this reduction in carbon emissions implies a global reduction of the required carbon uptake by forests of 2.0 GtCO₂year⁻¹ to 7.8 GtCO₂year⁻¹. This resembles a required forest land share of 60.0% or 77.9 Mkm² of forest area. Via a further 28.0% AR of permanent grassland and cropland, a forest land share of 62.0% or 80.4 Mkm² of forest area can be achieved to additionally sequester 8.0 GtCO₂year⁻¹.

1.1.3 Predictors of national forest land share

Compared to cross-sectional regression models, the FE panel model has the advantage of exploiting the longitudinal structure of the data as it only includes within-country variation. Hence, the FE model is not biased by cross-sectional unobserved heterogeneity (Brüderl and Ludwig 2015; Wooldridge 2010). If the strict exogeneity assumption (r (x_it,ε_it) = 0) holds, FE models adequately estimate unbiased causal effects (Vaisey and Miles 2017). The model can be written as

$$ {y}_{it}-{\overline{y}}_i=\left({\boldsymbol{x}}_{it}-{\overline{\boldsymbol{x}}}_i\right)\boldsymbol{\beta} +{\boldsymbol{Z}}_t\boldsymbol{\gamma} +{\varepsilon}_{it}-{\overline{\varepsilon}}_i $$

(3)

Here, y_it denotes the forest land share of country i in year t. ${\overline{y}}_i$ represents country i’s mean of the whole observation period. x_it stands for the vector of all exogenous variables for country i at time t, and ${\overline{\boldsymbol{x}}}_i$ for the average of the time observed. The model further comprises a vector of dummy variables (Z) for every year to control period effects for all countries (time FE). A country’s time-varying stochastic error term is represented by ε_it. All metric variables are included by taking their natural logarithm, which allows the estimation of elasticities. All standard errors are clustered by country and year and are therefore robust with respect to heteroscedasticity and autocorrelation. The reported regression results of Fig. 4 were tested for robustness analogous to the results of the analysis for the FLUXNET data as already explained above. Furthermore, all six models were recalculated using the total forest land area as dependent variable instead of forest land share. None of these checks detected any substantial deviations from the results reported in Fig. 4.