Anantha M. Prasad

Newer Classification and Regression Tree Techniques: Bagging and Random Forests for Ecological Prediction

The task of modeling the distribution of a large number of tree species under future climate scenarios presents unique challenges. First, the model must be robust enough to handle climate data outside the current range without producing unacceptable instability in the output. In addition, the technique should have automatic search mechanisms built in to select the most appropriate values for input model parameters for each species so that minimal effort is required when these parameters are fine-tuned for individual tree species. We evaluated four statistical models—Regression Tree Analysis (RTA), Bagging Trees (BT), Random Forests (RF), and Multivariate Adaptive Regression Splines (MARS)—for predictive vegetation mapping under current and future climate scenarios according to the Canadian Climate Centre global circulation model. To test, we applied these techniques to four tree species common in the eastern United States: loblolly pine (Pinus taeda), sugar maple (Acer saccharum), American beech (Fagus grandifolia), and white oak (Quercus alba). When the four techniques were assessed with Kappa and fuzzy Kappa statistics, RF and BT were superior in reproducing current importance value (a measure of basal area in addition to abundance) distributions for the four tree species, as derived from approximately 100,000 USDA Forest Service’s Forest Inventory and Analysis plots. Future estimates of suitable habitat after climate change were visually more reasonable with BT and RF, with slightly better performance by RF as assessed by Kappa statistics, correlation estimates, and spatial distribution of importance values. Although RTA did not perform as well as BT and RF, it provided interpretive models for species whose distributions were captured well by our current set of predictors. MARS was adequate for predicting current distributions but unacceptable for future climate. We consider RTA, BT, and RF modeling approaches, especially when used together to take advantage of their individual strengths, to be robust for predictive mapping and recommend their inclusion in the ecological toolbox.

A GIS-derived integrated moisture index to predict forest composition and productivity of Ohio forests (U.S.A.)

A geographic information system (GIS) approach was used in conjunction with forest-plot data to develop an integrated moisture index (IMI), which was then used to predict forest productivity (site index) and species composition for forests in Ohio. In this region, typical of eastern hardwoods across the Midwest and southern Appalachians, topographic aspect and position (rather than elevation) change drastically at the fine scale and strongly influence many ecological functions. Elevational contours, soil series mapping units, and plot locations were digitized for the Vinton Furnace Experimental Forest in southeastern Ohio and gridded to 7.5-m cells for GIS modeling. Several landscape features (a slope-aspect shading index, cumulative flow of water downslope, curvature of the landscape, and water-holding capacity of the soil) were used to create the IMI, which was then statistically analyzed with site-index values and composition data for plots. On the basis of IMI values for forest land harvested in the past 30 years, we estimated oak site index and the percentage composition of two major species groups in the region: oak (Quercus spp.), and yellow poplar (Liriodendron tulipifera) plus black cherry (Prunus serotina). The derived statistical relationships were then applied in the GIS to create maps of site index and composition, and verified with independent data. The maps show the oaks will dominate on dry, ridge top positions (i.e., low site index), while the yellow poplar and black cherry will predominate on mesic sites. Digital elevation models with coarser resolution (1:24K, 1:100K, 1:250K) also were tested in the same manner. We had generally good success for 1:24K, moderate success for 1:100K, but no success for 1:250K data. This simple and portable approach has the advantage of using readily available GIS information which is time-invariant and requires no fieldwork. The IMI can be used to better manage forest resources where moisture is limiting and to predict how the resource will change under various forms of ecosystem management.

Potential Changes in Tree Species Richness and Forest Community Types following Climate Change

Potential changes in tree species richness and forest community types were evaluated for the eastern United States according to five scenarios of future climate change resulting from a doubling of atmospheric carbon dioxide (CO2). DISTRIB, an empirical model that uses a regression tree analysis approach, was used to generate suitable habitat, or potential future distributions, of 80 common tree species for each scenario. The model assumes that the vegetation and climate are in equilibrium with no barriers to species migration. Combinations of the individual species model outcomes allowed estimates of species richness (from among the 80 species) and forest type (from simple rules) for each of 2100 counties in the eastern United States. Average species richness across all counties may increase slightly with climatic change. This increase tends to be larger as the average temperature of the climate change scenario increases. Dramatic changes in the distribution of potential forest types were modeled. All five scenarios project the extirpation of the spruce–fir forest types from New England. Outputs from only the two least severe scenarios retain aspen–birch, and they are largely reduced. Maple–beech–birch also shows a large reduction in area under all scenarios. By contrast, oak–hickory and oak–pine types were modeled to increase by 34% and 290%, respectively, averaged over the five scenarios. Although many assumptions are made, these modeled outcomes substantially agree with a limited number of predictions from researchers using paleoecological data or other models.

PREDICTING EXTINCTIONS AS A RESULT OF CLIMATE CHANGE

Widespread extinction is a predicted ecological consequence of global warming. Extinction risk under climate change scenarios is a function of distribution breadth. Focusing on trees and birds of the eastern United States, we used joint climate and environment models to examine fit and climate change vulnerability as a function of distribution breadth. We found that extinction vulnerability increases with decreasing distribution size. We also found that model fit decreases with decreasing distribution size, resulting in high prediction uncertainty among narrowly distributed species. High prediction uncertainty creates a conservation dilemma in that excluding these species under-predicts extinction risk and favors mistaken inaction on global warming. By contrast, including narrow endemics results in over-predicting extinction risk and promotes mistaken inaction on behalf of individual species prematurely considered doomed to extinction.

Potential redistribution of tree species habitat under five climate change scenarios in the eastern US

Global climate change could have profound effects on the Earth’s biota, including large redistributions of tree species and forest types. We used DISTRIB, a deterministic regression tree analysis model, to examine environmental drivers related to current forest-species distributions and then model potential suitable habitat under five climate change scenarios associated with a doubling of atmospheric CO2. Potential shifts in suitable habitat for 76 common tree species in the eastern US were evaluated based on more than 100,000 plots and 33 environmental variables related to climate, soils, land use, and elevation. Regression tree analysis was used to devise prediction rules from current species‐environment relationships. These rules were used to replicate the current distribution and predict the potential suitable habitat for more than 2100 counties east of the 100th meridian. The calculation of an importance value-weighted area score, averaged across the five climate scenarios, allowed comparison among species for their overall potential to be affected by climate change. When this score was averaged across all five climate scenarios, 34 tree species were projected to expand by at least 10%, while 31 species could decrease by at least 10%. Several species (Populus tremuloides, P. grandidentata, Acer saccharum, Betula papyrifera, Thuja occidentalis) could have their suitable habitat extirpated from US. Depending on the scenario, the optimum latitude of suitable habitat moved north more than 20 km for 38‐47 species, including 8‐27 species more than 200 km or into Canada. Although the five scenarios were in general agreement with respect to the overall tendencies in potential future suitable habitat, significant variations occurred in the amount of potential movement in many of the species. The five scenarios were ranked for their severity on potential tree habitat changes. Actual species redistributions, within the suitable habitat modeled here, will be controlled by migration rates through fragmented landscapes, as well as human manipulations. # 2002 Elsevier Science B.V. All rights reserved.

Potential colonization of newly available tree-species habitat under climate change: an analysis for five eastern US species

We used a combination of two models, DISTRIB and SHIFT, to estimate potential migration of five tree species into suitable habitat due to climate change over the next 100 years. These species, currently confined to the eastern half of the United States and not extending into Canada, are Diospyros virginiana (persimmon), Liquidambar styraciflua (sweetgum), Oxydendrum arboreum (sourwood), Pinus taeda (loblolly pine), and Quercus falcata var. falcata (southern red oak). DISTRIB uses a statistical approach to assess potential suitable habitat under equilibrium of 2 × CO2. SHIFT uses a cellular automata approach to estimate migration and is driven primarily by the abundance of the species near the boundary, forest density inside and outside of the boundary, and distance between cells. For each cell outside the current boundary, SHIFT creates an estimate of the probability that each unoccupied target cell will become colonized over 100 years. By evaluating the probability of colonization within the potential ‘new’ suitable habitat, we can estimate the proportion of new habitat that might be colonized within a century. This proportion is low (<15%) for all five species, suggesting that there is a serious lag between the potential movement of suitable habitat and the potential for the species to migrate into the new habitat. However, humans could hasten the migration of certain species by physically moving the propagules, especially for certain rare species that are unable to move sufficiently through fragmented landscapes, or even more common species, e.g., beech, that have lost many of their animal dispersers.

Predicting the Potential Future Distribution of Four Tree Species in Ohio Using Current Habitat Availability and Climatic Forcing

Weinvestigated the effect of habitat loss on the ability of trees to shift in distribution across a landscape dominated by agriculture. The potential distribution shifts of four tree species (Diospyros virginiana, Oxydendron arboreum, Pinus virginiana, Quercus falcata var. falcata) whose northern distribution limits fall in the southern third of Ohio were used to assess possible distribution shift scenarios as a result of global warming. Our predictions derive from the results of simulations using (a) forest inventory based estimates of current distribution and abundance of target species; (b) a satellite-based estimate of forest habitat availability; and (c) a tree migration model (SHIFT). The current distribution and abundance of trees was estimated using USDA Forest Services Forest Inventory Analysis data and distribution maps from the late 1960s; pre-European settlement forest–nonforest maps were used to represent the fully forested condition for calibration and comparison. Habitat-availability estimates in Ohio were estimated using classified Landsat Thematic Mapper (TM) data from 1994. Tree abundance, forest availability and migration were modeled using a 1-km2 pixel size. Forest availability was estimated as the proportion of forested TM pixels within each cell. The probability of a migrating species colonizing an unoccupied cell is modeled as a function of forest availability and distance to occupied cells. The results of the migration models suggest that the species studied are capable of colonizing virtually any forested location within Ohio over the next 100 years if climatic controls over the current distribution that may currently inhibit northward movement are relaxed. The contiguous distribution of these species, however, is not likely to shift more than 10 km during the next century regardless of the magnitude of the climate change. Examining the sensitivity of our simulations by varying critical model attributes, we found that whereas the variables controlling the amount of long-distance dispersal have strong effects on migration rates in the fully forested 1800 situation, they have significantly lesser effects on projections of future migration into highly fragmented forests. The low forest availability that characterizes much of the current Ohio landscape, along with the low likelihood of long distance dispersal, result in potential distribution shifts that are concentrated within the principally forested corridors in southeastern Ohio. We propose that in contrast to the past, future tree migrations are likely to be spatially and temporally correlated as a result of large climatic forcing and channelization through limited regions of available habitat. With respect to the management of biodiversity, this result suggests that it may be very difficult to discern plant migrations of native forest species owing to exceedingly slow rates of movement.

Using landscape analysis to assess and model tsunami damage in Aceh province, Sumatra

The nearly unprecedented loss of life resulting from the earthquake and tsunami of December 26, 2004, was greatest in the province of Aceh, Sumatra (Indonesia). We evaluated tsunami damage and built empirical vulnerability models of damage/no damage based on elevation, distance from shore, vegetation, and exposure. We found that highly predictive models are possible and that developed areas were far more likely to be damaged than forested zones. Modeling exercises such as this one, conducted in other vulnerable zones across the planet, would enable managers to create better warning and protection defenses, e.g., tree belts, against these destructive forces.

Ecosystem vulnerability assessment and synthesis: a report from the Climate Change Response Framework Project in northern Wisconsin

The forests of northern Wisconsin will likely experience dramatic changes over the next 100 years as a result of climate change. This assessment evaluates key forest ecosystem vulnerabilities to climate change across northern Wisconsin under a range of future climate scenarios. Warmer temperatures and shifting precipitation patterns are expected to influence ecosystem drivers and increase stressors, including more frequent disturbances and increased amount or severity of pests and diseases. Forest ecosystems will continue to adapt to changing conditions. Identifying vulnerable species and forests can help landowners, managers, regulators, and policymakers establish priorities for management and monitoring.

Regional scale patterns of fine root lifespan and turnover under current and future climate

Fine root dynamics control a dominant flux of carbon from plants and into soils and mediate potential uptake and cycling of nutrients and water in terrestrial ecosystems. Understanding of these patterns is needed to accurately describe critical processes like productivity and carbon storage from ecosystem to global scales. However, limited observations of root dynamics make it difficult to define and predict patterns of root dynamics across broad spatial scales. Here, we combine species-specific estimates of fine root dynamics with a model that predicts current distribution and future suitable habitat of temperate tree species across the eastern United States (US). Estimates of fine root lifespan and turnover are based on empirical observations and relationships with fine root and whole-plant traits and apply explicitly to the fine root pool that is relatively short-lived and most active in nutrient and water uptake. Results from the combined model identified patterns of faster root turnover rates in the North Central US and slower turnover rates in the Southeastern US. Portions of Minnesota, Ohio, and Pennsylvania were also predicted to experience >10% increases in root turnover rates given potential shifts in tree species composition under future climate scenarios while root turnover rates in other portions of the eastern US were predicted to decrease. Despite potential regional changes, the average estimates of root lifespan and turnover for the entire study area remained relatively stable between the current and future climate scenarios. Our combined model provides the first empirically based, spatially explicit, and spatially extensive estimates of fine root lifespan and turnover and is a potentially powerful tool allowing researchers to identify reasonable approximations of forest fine root turnover in areas where no direct observations are available. Future efforts should focus on reducing uncertainty in estimates of root dynamics by better understanding how climate and soil factors drive variability in root dynamics of different species.