Forests contain ~80% of above ground carbon and sequester ~30% of annual fossil fuel emissions, and thus have a prominent role in the carbon balance [1
]. The distribution of terrestrial ecosystems is strongly influenced by climate [3
], so how ecosystems reorganize from climate change presents an important research area in regards to terrestrial carbon. Paleoecological records predicted forest migration rates during the last glacial period greater than considered possible [7
]. Two theories for how this happened are rapid migration and refugial populations, and is known as Reid’s Paradox [12
]. For large domain studies that predict the potential redistribution of vegetation by plant migration due to expected future climate change, Dynamic Global Vegetation Models (DGVMs) or Earth System Models (ESMs) are used. Research at these domain sizes often implement scaling strategies at the cost of some fine scale processes, such as individual-based plant migration, to reduce computational requirements. Therefore, improvements to their underlying vegetation demographics are continued and are important research topics, especially when predicting the redistribution of vegetation from plant migration due to climate change.
Most DGVMs are cohort, not individual, based, and given the complexities of dispersal between grid cells, they approximate the transient response of plant migration due to climate change through other methods [15
]. TRIFFID [17
] leaves a fraction of its seed bank in all cells, so if climate changes, better adapted species may alter the species composition. Sheffield-DGVM (SHE) [18
], ORCHIDEE [20
], and Lund-Postdam-Jena (LPJ) [21
] have establishment of climatically favored plant functional types (PFTs). However, there is no between grid cell dispersal due to the complexity of this interaction, and partially because they are not individually based. Two DGVMs are individually based, and have attempted to simulate the transient response of vegetation between grid cells. SEIB-DGVM [22
] is individually based and simulated migration between cells in Africa. For each half-degree cell, a 30 m × 30 m forest gap model was run and unlimited vs. no-migration simulated. The difference in the size of the cell and the spatial extent of the gap model did not allow for simulations at a specific dispersal distance. LPJ-GUESS [24
] had dispersal between patches (smaller areas within a grid cell), and calculated the probability of spread to a neighboring cell based on a dispersal kernel, but only in an idealized landscape. The difficulties of simulating this fine scale process in a larger domain means most dispersal and migration studies continue to occur in Forest Landscape Models such as TreeMig [26
], which has simulated regions of European Forest, and LANDIS-PRO [27
], which has simulated multiple regions of the US. Moving beyond regional scales of simulating the transient response of plant migration with explicit dispersal remains a challenge.
Another DGVM, the Ecosystem Demography (ED) [29
], is individually based. One of the reasons it is able to simulate large domains is that it is pseudo-spatial. Within every grid cell, the number of individuals of a PFT and their size and age, and the area they occupy (patch size) are known, but the explicit location in the grid cell is not. Recently, a pseudo-spatial dispersal sub-model was developed and implemented in ED [31
]. It simulates the spatially explicit process of plant migration from dispersal in the pseudo-spatial framework of ED. Here, previous research where the PFT distribution in ED for northern North America was validated with remote sensing data, and then a climate change scenario run that showed the equilibrium response of vegetation and carbon was used with the new dispersal sub-model, and the transient response from individual-based dispersal in large domains explored with multiple scenarios. The migration sub-model is run with a model experimental designed to investigate the impact of (1) dispersal distance, (2) dispersal mode, and (3) disturbance rate on the potential transient redistribution of terrestrial vegetation and carbon from climate change in northern North America over a range of time scales (years–millennia).
This was the first application of the ED migration sub-model that was developed to continue to advance new ways of simulating the transient response of vegetation from individual based dispersal at large domains. Landscape models are adapting the PFT and cohort structure [28
] used in many DGVMs to increase the size of the domains they simulate, and DGVMs implement various strategies to simulate the transient response [18
], but are often not individually based or contained between grid cell dispersal. ED, with its pseudo-spatial structure, was then positioned to continue to help advance research in this field, as it is constructed for large domain simulation but is individually based, so when the spatially explicit process of migration was adapted to its pseudo-spatial framework, it offered another model to simulate the transient response of plant migration. Though all scenarios ran at the half-degree resolution that other DGVMs often run at, this occurred from data limitation, not computational limitation. Future climatologies that contain the necessary specific humidity input are often not readily available, so only North America data from the NARCCAP was used. Additionally, only a portion of that data was used, as transition zones were the focus and ED had previously been validated with remote sensing data on the distribution of deciduous and evergreen PFTs in North America [33
]. ED is readily adaptable in both resolution and number of PFTs. For NASA CMS, ED is being run at 90 m resolution for parts of the US [39
], and NASA GEDI plans to run the contiguous US at 1 km [42
], which is on the larger side of most landscape models, but within reason. As ED is a DGVM, specific species are lost, but additional PFTs can be added. The version used here also has three tropical species and two grass species outside of the domain. ED2, a modification of the original ED that is often used in research in smaller domains, used five temperate PFTs for trees in study at Harvard Forest [49
] and seven tropical tree PFTs at a Costa Rican site [50
], so it can be modified for whatever the research requires. Our findings on the transient response of plant migration in northern North America are consistent with previous studies and present another method for studying the transient response in large domains.
Here, the transient response of plant migration on vegetation and carbon redistribution over a domain where the initial PFT distribution under current climate was verified with remote sensing data [33
] was assessed. The transient responses PFT and carbon distribution never matched the equilibrium response for a variety of reasons. The evergreen PFT was outcompeted by the deciduous PFT at many locations as it moved north, but until the deciduous PFT migrated there, the evergreen PFT stored slightly more carbon (Figure 9
A,B). Migration also increased the northern forest extent (Figure 5
) as sites that were not classified as forest in the equilibrium response exceeded the forest threshold definition when new individuals migrated to those sites (Figure 5
). Migration to the predicted equilibrium response only covered the entire domain when the dispersal distance was 10 km per year (Figure 6
). So, with new forested areas and the evergreen PFT storing more carbon than the deciduous PFT before it is outcompeted, carbon sequestration potential was almost always higher than the predicted equilibrium response, and the PFT distribution lower from the time it took to migrate and the increased extent of the forest. This study, to our knowledge, is one of the first to examine the transient response of individual-based plant migration with an advanced mechanistic model at continental scales with multiple dispersal rates, dispersal modes, and disturbance rates. Though novel in approach, our results are comparable to previous studies on vegetation and carbon redistribution from climate change.
Modest net changes in total carbon with larger underlying grid changes, presented here, were also found by Schaphoff et al. [51
]. Using the LPJ-DGVM with five different general circulation models (GCMs) for a climate change scenario produced an average increase of 7.1% in vegetation carbon across the globe. However, they had boreal forests as a carbon source, whereas we found it to be a temporary sink. This could be a result of the climate change scenario they used, the IS92a. The atmospheric CO2
value used for our research was 575 ppm while they used 703 ppm. Bachelet et al. [52
] used an equilibrium model, MAPSS, and a dynamic model, MC1, to simulate changes in potential equilibrium vegetation and carbon distribution in the US, and found that moderate temperature increases produced an increase in carbon with limited redistribution, but higher temperature changes produced widespread redistribution and carbon loss. Solomon and Kirilenko [53
] used three climate scenarios to predict future equilibrium distribution carbon with and without migration, and found modest total gains in carbon were the product of larger underlying redistribution of ecosystems. Sitch et al. [16
] ran five DGVMs with the A1 scenario and all had the tundra becoming a sink, while their temperate results varied, but overall were also a sink. At the regional scale, Brandt et al. [54
] and Jin et al. [27
] ran three models, Climate Change Tree Atlas, LANDIS-PRO, and LINKAGES in the central hardwood ecosystem and projected significant changes in species composition with moderate carbon changes. Wang et al. [28
] used LANDIS-PRO in the northeastern United States with four climate change scenarios and all found an increase in AGB, with hardwoods replacing conifer species. While not reporting on carbon, Morin [55
] looked at 16 North American tree species and their suitable zones, and Iverson [56
] used various models to predict range shift under climate change, all of which are consistent with our findings. Migration’s greatest influence will occur at transition zones; in North America that means evergreen forests are expected to migrate from the taiga into the tundra [58
], and deciduous forests are expected to move northward [59
]. Northward migration of boreal species into regions previously classified as tundra is already occurring [60
] as remote sensing supports tree line advance [61
]. Both of these trends were represented in these previous studies and our research.
As for disturbance, it can both accelerate and impede migration. Disturbance rates control the probability of new species establishment, as some disturbance is needed for new species to enter an ecosystem, but too much prevents establishment [62
]. The MIGRATE model investigates how available habitat impacts migration rates and shows that increased suitable habitat increases migration rates [64
]. FORSKA, a gap model, also showed increased disturbance lead to faster redistribution in the mixed conifer/northern hardwoods zone of northern Europe [65
]. Representative of this, with increased disturbance the deciduous PFT migrated and established faster (Figure 7
), as it benefited from less competition. However, species are only so resilient to disturbance, so increased disturbance in low biomass areas can impede migration [56
]. Our results showed that the northern extent of the evergreen PFT was reduced (Figure 1
) as the low growth rates there prevented forest establishment with a higher disturbance rate.
This study has made important advances in using an individual-based mechanistic model to predict the potential transient response of vegetation and carbon to climate change over large domains. Future work should prioritize expansion of the scenarios used and incorporate additional metrics. There are many other studies [17
] that simulate the transient response in some capacity, but often lack between grid cell dispersal, or are not individually based. The results are supported by previous studies, and offer another method to potentially examine Reid’s Paradox of rapid plant migration over large domains, but this was still a simplification of a complex process. Additional PFTs can be added [49
] depending on the research question. A static dispersal distance was used, but long distance dispersal is governed by the tail of dispersal kernels [10
], and can be implemented. Only one climate change scenario was used, with a static value of CO2
that is high, but not the highest presented in the SRES. The NARCCAP is producing numerous current and future, and as they are all forced with the A2 scenario, a sensitivity analysis can be performed. The disturbance rate can be PFT specific rather than equal for all types, as climate change is causing increased insect outbreaks that are damaging boreal forests [67
], so the disturbance rate could be increased in at-risk areas or for specific PFTs. Fire is also increasing and altering species distribution [68
], so ED’s fire sub-model could be parameterized and explored. The climatologies were at half-degree but could be downscaled, and if future climatologies for other regions are generated they could be explored. The MsTMIP climate data used for the current climatology was also used in a model intercomparison and demonstrated a wide range in potential changes based on the model [15
], so another intercomparison could be performed. This research presents a novel method to simulate the transient response of vegetation and carbon to climate change in large domains, and future research should replicate many of the studies that have been conducted at smaller scales on disturbance, dispersal, competition, and landscape characteristics, and be implemented at scales up to global in model intercomparison projects and sensitivity analyses.