Dynamic Global Vegetation Models (DGVMs) simulate the effect of global climate on biological systems, often focusing on carbon dynamics in terrestrial ecosystems because carbon sources drive climate change while carbon sinks mitigate other emission sources. Disturbances like wildfire are a key driver of carbon exchange between vegetation and climate. Yet this complexity and associated computational demands have limited the number of models that attempt to predict interactions between fire disturbances and species composition under a changing climate. We developed a hybrid model from MC1 (a DGVM with emergent fire prediction) and DISTRIB (a statistical habitat suitability model for tree species) to explore potential feedbacks between climate, fire, and tree species. A fire index based on the carbon in biomass consumed and killed during fire events is calculated from MC1 output and used in combination with a ranking representing each species? adaptation or vulnerability to fire to modify DISTRIB's species composition predictions under a specific climate scenario (historical and A2). Then, fuel depths associated with this fire-influenced species composition are used to influence the fire regimes predicted by MC1. These simulations demonstrate potential shifts in fire regimes in the Eastern U.S. as more fire-adapted southern tree species migrate northward.
The U.S. Geological Survey?s LandCarbon project produced downscaled, qualitative and quantitative scenarios for ecoregions of the United States, with scenario characteristics and assumptions consistent with IPCC Special Report on Emissions Scenarios (SRES) storylines. The model FOREcasting SCEnarios of land-use change (FORE-SCE) used downscaled scenarios to create spatially explicit LULC projections for ecoregions of the United States. FORE-SCE was designed to take advantage of both historical and contemporary LULC research and data from the USGS. Data from the USGS Land Cover Trends project and the National Land Cover Database (NLCD) were used to parameterize a unique, patch-based modeling approach. Individual patches of LULC ?change? were placed on the landscape in an iterative modeling procedure until the scenario-prescribed proportions of LULC change for a given year were met, with patch characteristics determined by historical regional trends and scenario storylines. We also modeled forest cutting and forest stand age, using regional histories and scenario characteristics to set scenario-based cutting rotations for each region. When complete in late 2012, we will have produced 250m resolution, thematically detailed, annual LULC maps from 2001 to 2100 for each of four IPCC SRES storylines for the entire conterminous United States.
The Energy Independence and Security Act of 2007 (EISA) mandated the U.S. Department of Interior to assess carbon storage, carbon sequestration, and fluxes of other greenhouse gases (GHG) for ecosystems of the United States. The U.S. Geological Survey developed a methodology to quantify baseline (current) carbon sequestration and GHG fluxes, and to evaluate carbon sequestration potential and GHG fluxes for multiple future scenarios. The baseline component relied on existing inventory and land-use data to analyze spatial distributions of carbon stocks and GHG fluxes. Potential future conditions were analyzed using IPCC scenarios. Qualitative storylines and quantitative proportions of land use and land cover (LULC) were downscaled to individual ecoregions for each scenario. The FORE-SCE model was then used to produce spatially explicit LULC projections based on each IPCC SRES scenario. A separate disturbance model was used to model fire occurrence and distribution for each scenario. Integrated LULC and disturbance projections were used by the General Ensemble Modeling System (GEMS) to analyze future carbon stocks and GHG fluxes for each scenario. Our scenario-based LULC and disturbance projections have the potential to support analyses of other ecosystem processes, including impacts of LULC change on hydrology, biodiversity, and socially or economically important ecosystem services.
The evaluation of the accuracy of predictive models can require the comparison between a reference Boolean map and a rank map, where higher ranks indicate a higher predicted likelihood of the presence of the Boolean feature. In this article, I propose a new method to compare these maps at multiple thresholds. Each threshold generates a 2 by 2 confusion matrix with four entries called: hits, misses, false alarms, and null successes. This method plots the four entries at each of many thresholds on a figure. The plotted figure shows five concepts: 1) the percentage of the feature in the reference map 2)the four entries in the confusion matrix at the corresponding threshold for the predicted ranking 3) the best possible ranking 4) the worst possible ranking, and 5) the random ranking. I illustrate the procedure with examples from species distribution modeling and land change modeling. The newly proposed figure is easier to interpret and conveys more information compared to the commonly used Receiver Operating Characteristic (ROC) curve. The new method shows information concerning the threshold values in the map and the entries in the confusion matrices, which the ROC method hides. The new method also provides a numerical index that is equal to the Area under the ROC curve (AUC).