High economic costs of reduced carbon sinks and declining biome stability in Central American forests

Tropical forests represent important supporting pillars for society, supplying global ecosystem services (ES), e.g., as carbon sinks for climate regulation and as crucial habitats for unique biodiversity. However, climate change impacts including implications for the economic value of these services have been rarely explored before. Here, we derive monetary estimates for the effect of climate change on climate regulation and habitat services for the forests of Central America. Our results projected ES declines in 24–62% of the study region with associated economic costs of $51–314 billion/year until 2100. These declines particularly affected montane and dry forests and had strong economic implications for Central America’s lower-middle income countries (losses of up to 335% gross domestic product). In addition, economic losses were mostly higher for habitat services than for climate regulation. This highlights the need to expand the focus from mere maximization of CO2 sequestration and avoid false incentives from carbon markets.

(4) C3 grasses (C3), which cover parts of the higher mountain ranges as parámo grasslands; and (5) C4 grasses, which are adapted to high tempratures and thrive in savanna-like landscapes (e.g. Mexico, parts of the Pacific coast). PFT-specific settings are summarized in Table S3.1, general settings in Table S3.2.
To account for the effect of topography within the coarse 0.5° resolution climate inputs, we adapted the landforms approach first introduced by Werner et al. 2 . The main idea of this method is to adjust climate inputs based on fundamental geophysical relationships of topography with temperature and insolation. As temperature decreases with elevation and insolation depends on aspect, slope and time of the year, these causalities can be used to adapt climate inputs at the fine resolution of digital elevation models (DEMs). While a high spatial resolution allows for a precise representation of gradients in mountain ranges, the computational effort for modelling is high and potentially not well-invested for large homogenous landscapes. Therefore, each grid cell is classified into so-called With these adjusted settings we ran our simulations for the above described PFTs for each stand (landform) with 15 replicate patches each.

Model evaluation and adaptation
For a basic evaluation of model performance we first tested LPJ-GUESS with default settings and compared the outputs to satellite-derived data and maps. Biome types were derived following the biome classification scheme of Snell et al. 6 and compared to the biome map by Olson et al. 7 . In terms of biome distribution, the default settings with activated fire module (GlobFIRM) 8  A comparison between simulated and satellite-derived NPP (MODIS-NPP) 11 further revealed an underestimation, particularly in montane regions. In relation to this, Atkin et al. 12 pointed out, that the thermal acclimation of respiration can play a significant role for plant growth, yet is rarely considered in dynamic vegetation models. To better account for this fact, we adapted the respiration function in our model code based on adjustments by Dantas de Paula et al. 13 and Thum et al. 14 .
Finally, we compared carbon mass outputs to estimations of above-and belowground carbon in vegetation by Spawn and Gibbs 15 . For most areas, our model simulations strongly exceeded the satellite-based predictions (RMSE > 16). Yet, the largest discrepancies also fell together with the areas with the highest uncertainty in the compared map. In addition to this, the standard LPJ-GUESS disturbance interval of 200 years allowed for a long accumulation period of carbon.
Opposed to this, the region is quite regularly affected by disturbances like tropical storms, floods, land slides and droughts 16 -also in combination with El Niño/La Niña dynamics -and may be more so in the future 17 . Therefore, we decided to reduce the disturbance interval to 100 years. With these adjusted settings, our model showed moderate to good agreement with satellite-derived products and biome maps and was able to reproduce major spatial patterns (see Figures S8-9).   Leaf phenology evergreen evergreen raingreen evergreen any any