Journal of Geophysical Research: Biogeosciences

Climate-induced changes in biome distribution, NPP, and hydrology in the Upper Midwest U.S.: A case study for potential vegetation

Authors

  • Melissa M. Motew,

    Corresponding author
    • Center for Sustainability and the Global Environment (SAGE), The Nelson Institute, University of Wisconsin-Madison, Madison, Wisconsin, USA
    Search for more papers by this author
  • Christopher J. Kucharik

    1. Center for Sustainability and the Global Environment (SAGE), The Nelson Institute, University of Wisconsin-Madison, Madison, Wisconsin, USA
    2. Department of Agronomy, University of Wisconsin-Madison, Madison, Wisconsin, USA
    Search for more papers by this author

Corresponding author: M. M. Motew, Center for Sustainability and the Global Environment (SAGE), The Nelson Institute, University of Wisconsin-Madison, 1710 University Ave., Madison, WI 53726, USA. (motew@wisc.edu)

Abstract

[1] We investigated the impacts of recent climate change on potential vegetation distributions and carbon and water cycling across the Upper Midwest from 1948 to 2007. We used the Agro-IBIS dynamic vegetation model driven by a newly developed gridded daily climate data set at 5 min × 5 min spatial resolution. Trends in climate variables were spatially heterogeneous over the study period and were associated with an overall increase in net primary productivity (NPP). We observed an average regional change in total NPP of 41 ± 30 g C m−2 (8%). Increased summer relative humidity and increased annual precipitation were key variables contributing to the positive trends. Mechanisms for increased productivity included a reduction in soil moisture stress as well as increased stomatal conductance resulting from an increase in summertime humidity. Model simulations also showed an average total increase in annual groundwater recharge throughout the region of 39 ± 35 mm (45%) driven by increases in annual precipitation. Evapotranspiration had a highly variable spatial trend over the 60 year period, with an average total change of 5 ± 21 mm (1%) across all grid cells. The location of the Tension Zone, a broad ecotone dividing northern mixed forests and southern hardwood forests and prairies, was not observed to migrate using analysis of meteorological variables.

1 Introduction

[2] Climate change is expected to severely impact ecosystems [Heyder et al., 2011] and is already causing changes in ecosystem structure and function throughout the world [Walther et al., 2002]. Global studies reveal that recent climate change has caused shifts in species distributions [Hickling et al., 2006] and changes in fluxes of carbon [Boisvenue and Running, 2006] and water [Allen et al., 2010; Breshears et al., 2005]. At local and regional scales, less is known about how ecosystems are responding to climate change, including how changes in vegetation are impacting hydrological cycles and vice versa [Troch et al., 2009]. This knowledge gap is compounded by spatial variability in topography and climate [Karl et al., 2009], as well as uncertainty in fine-scale climate change projections [Lean and Rind, 2009]. Carbon and water cycles are fundamentally linked within ecosystems through photosynthesis and plant water uptake. Shifts in vegetation distributions, which are expected with climate change [Soja et al., 2007], can affect both carbon and water cycling by altering primary productivity [Chiang et al., 2008] as well as hydrologic partitioning [Brooks et al., 2011]. Because of the couplings between vegetation, carbon, and water, climate change has the potential to impact ecosystems through numerous pathways.

[3] In the Upper Midwest U.S., home to a variety of biomes, carbon and water cycles are associated with multiple ecosystem services including runoff regulation, protection of soil resources, and provision of habitat, food, timber, and opportunities for recreation. Over the second half of the 20th century, the Great Lakes region saw changes in hydrology that were consistent with global projections [Hayhoe et al., 2010]. There has been an increase in precipitation and heavy precipitation events, a decrease in snow cover and lake ice duration, earlier spring melt, peak stream flow and lake levels, as well as increased hydrologic flooding [Changnon and Westcott, 2002; Changnon et al., 2001; Dyer and Mote, 2006; Jensen et al., 2007; Kucharik et al., 2010]. At finer spatial scales, however, there was evidence for spatial variation in hydrologic trends. In Wisconsin, for instance, some northern areas of the state underwent annual average precipitation deficits of four inches between 1950 and 2006, whereas some southern areas of the state had surpluses of seven inches [WICCI, 2011]. Region-wide averages therefore do not necessarily represent actual climate change at local levels.

[4] In terms of carbon cycling, the greater Midwest region has likely seen an increase in net primary productivity (NPP) over the second half of the 20th century. Remote-sensing studies show that vegetation underwent increases in NPP toward the end of the century; however, many of these trends were likely attributable to factors other than climate, such as intensive agricultural production and forest regrowth after natural disturbance [Hicke et al., 2002; Xiao and Moody, 2005]. At least one modeling study examined the relationship between recent climate change and NPP for ecosystems of the central and eastern U.S. [Twine and Kucharik, 2009]. Between 1950 and 2002, they found increases in NPP for crops and deciduous broadleaf forests, likely attributable to increases in summertime precipitation, and negligible changes in NPP for evergreen conifer forests located in the Upper Midwest. Our study expands upon the work of Twine and Kucharik [2009] by allowing for shifts in vegetation distributions, using higher spatial and temporal resolution climate driver data, examining hydrological ecosystem indicators in addition to NPP, and considering other potential climatic drivers of change in addition to temperature and precipitation.

[5] In addition to changes in water and carbon cycling, the Upper Midwest has undergone other climatic changes consistent with global projections. These include warming, especially during winter and at higher latitudes, and a lengthened growing season [Kucharik et al., 2010; WICCI, 2011]. Plants and animals have begun to respond to such changes, evidenced in places like Wisconsin with the earlier arrival of migratory birds and blooming of certain plants in spring [Bradley et al., 1999]. Future projections suggest these climate trends will continue, but that changes could be much more profound. In some cases they may occur more rapidly than plants and animals can adapt [WICCI, 2011].

[6] We define the Upper Midwest, our study region, to include the states of Minnesota, Wisconsin, and Michigan as well as portions of Iowa, Illinois, South and North Dakota, and Nebraska. The northern areas of the Upper Midwest are heavily forested, containing both deciduous broadleaf and evergreen conifer species. The southern areas contain a mosaic of agriculture, urban development, southern hardwood forest, savanna, and prairie. A distinguishing feature of the Upper Midwest landscape is the Tension Zone (TZ), an ecologically sensitive region that marks the transition between mixed forests in the north and southern hardwood forests and prairies in the south [Curtis, 1959]. Ecotones are regarded as sensitive regions for detecting climate change impacts [Neilson, 1993; Risser, 1995], especially those based on climatic gradients [Loehle, 2000]. Since the Upper Midwest TZ is associated with a prominent climate gradient [Curtis, 1959; Kucharik et al., 2010], it may serve as an early indicator of climate change impacts within the region. A northward shift of the TZ is expected in the 21st century, marked by a decline of boreal evergreen species and an increase in temperate hardwood species north of the TZ [Jones et al., 1994; Notaro et al., 2012; Scheller and Mladenoff, 2008; WICCI, 2011]. These predictions are supported by a historical precedence for species migrations within the region in response to changing climate [Webb and Bryson, 1972; Webb, 1974, 1987]. One study has already documented a 15–20 km northward shift of isarithms associated with the TZ in Wisconsin in response to recent climate change [Kucharik et al., 2010].

[7] In this study we investigate changes to natural ecosystems of the Upper Midwest resulting from recent climate change by focusing on carbon and water-based ecosystem indicators as well as vegetation distributions. We drive the Agro-IBIS (Integrated BIosphere Simulator) dynamic vegetation model [Foley et al., 1996; Kucharik, 2003; Kucharik et al., 2000] with a climate data set consisting of daily measurements at 5 min × 5 min spatial resolution over the time period 1948–2007. We consider potential vegetation even though much of the landscape of the Upper Midwest, particularly in the southern portion, is highly disturbed by humans. By focusing on potential vegetation, we attempt to isolate the impact of climate on representative natural ecosystems of the region.

[8] Our study explores three main aspects to the structure and functioning of natural vegetation in the Upper Midwest. The first involves the distribution of land biomes, including an analysis of whether or not the TZ has shifted. Second, we investigate impacts to NPP, a carbon-based indicator of ecosystem health. Finally, we consider changes to the hydrologic budget, including changes in evapotranspiration (ET), runoff, groundwater recharge (GWR), water availability to plants, as well as hydrologic links to NPP. For changes in ecosystem variables, we aim to identify the most influential climatic drivers. We consider maximum and minimum temperature (Tmax and Tmin), precipitation (P), relative humidity (RH), solar radiation, as well as other derived climatic quantities such as growing season length and growing degree days (GDD).

2 Materials and Methods

2.1 The Agro-IBIS Model

[9] The Agro-IBIS model is a U.S. regional version of the IBIS dynamic global vegetation model (DGVM) [Kucharik et al., 2000] that simulates a wide range of ecological, biophysical, and physiological processes. These include carbon, water, energy, and nutrient balance for a collection of both natural ecosystems (i.e., trees, grasses, shrubs) and managed agro-ecosystems (i.e., corn, soybean, wheat, Miscanthus, and switchgrass). Agro-IBIS contains all of the original components of the IBIS model [Foley et al., 1996; Kucharik et al., 2000] as well as improved nitrogen cycling, solute transport, and representation of agro-ecosystems [Kucharik, 2003; Vanloocke et al., 2010]. Within a single, hierarchical framework, Agro-IBIS simulates rapid biophysical processes as well as long-term ecosystem dynamics. The model includes representations of land surface and soil physics (energy, water, and momentum exchange between soil, vegetation, and the atmosphere), canopy physiology (canopy photosynthesis, conductance, and respiration), terrestrial carbon balance (net primary productivity, soil respiration, organic matter decomposition), as well as processes relevant to agro-ecosystems such as solute transport and management options [Kucharik, 2003]. Agro-IBIS simulates hydrological quantities including ET, runoff, and GWR. Groundwater recharge is water that infiltrates completely through the soil layers all the way to a hypothetical groundwater reservoir, yet the model does not explicitly simulate groundwater levels or flows.

[10] Vegetation is modeled in the form of plant functional types, or PFTs (Table 1). Agro-IBIS contains a total of 17 PFTs, 12 that are natural vegetation, and five that are crops. Plant functional types are differentiated by ecophysiological characteristics including basic physiognomy (trees, shrubs, and grasses), leaf habit (deciduous and evergreen), photosynthetic pathway (C3 and C4), and leaf form (broadleaf and needle-leaf). In each grid cell, two canopies are simulated: an upper canopy for trees and a lower canopy for shrubs, grasses, and crops (yet crops are omitted in this study). The ecological differences distinguishing PFTs allow them to mechanistically compete for light and water. For example, because trees are taller they have better access to light than shrubs and grasses. However, because shrubs and grasses have shallower roots, they may be able to access water more readily after precipitation events. Biomes are assigned within Agro-IBIS based on the assemblage of PFTs present in a grid cell as well as values of leaf area index (LAI) (Table 2). Leaf area index thresholds for upper and lower canopies and individual PFTs are used to determine if the biome is forest, savanna, shrubland, grassland, tundra, or desert.

Table 1. Plant Functional Types (PFTs) Simulated in Agro-IBIS, Excluding Tropical Forest and Crop PFTsa
Plant Functional TypeTtolTchillTwGDD5Potential Biomes
  1. a

    Ttol is minimum tolerable mean monthly temperature (°C); Tchill is mean monthly temperature chilling requirement (°C); Tw is mean monthly temperature heat requirement or heat stress limit (°C); GDD5 is growing degree days (base 5 °C). “S” and “P” indicate whether the limit was taken from Prentice et al. [1992] (P) or Sitch et al. [2003] (S). One, two, and three asterisks indicate whether the limit appears south, within, or north of the study domain, respectively. Potential biomes include all biomes in which the PFT can be present.

Warm-temperate broadleaf evergreen tree>3 (S*)<18.8 (S*)  Temperate broadleaf evergreen forest, savanna, grassland, dense shrubland, open shrubland
Temperate conifer evergreen tree>−2 (S*)<22 (S**)  Temperate conifer evergreen forest, mixed forest, savanna, grassland, dense shrubland, open shrubland
Temperate broadleaf cold-deciduous tree>−15 (P**)<15.5 (P*) >1200 (P***)Temperate deciduous forest, mixed forest, savanna, grassland, dense shrubland, open shrubland
Boreal conifer evergreen tree>−35 (P***)<−2 (P*)<22 (S**)>600 (S***)Boreal conifer forest, mixed forest, grassland, dense shrubland, open shrubland
Boreal broadleaf cold-deciduous tree <−2 (S*)<22 (S**)>350 (P***)Boreal deciduous forest, mixed forest, grassland, dense shrubland, open shrubland
Boreal conifer cold-deciduous tree <−2 (S*)<22 (S**)>350 (S***)Boreal deciduous forest, mixed forest, grassland, dense shrubland, open shrubland
Evergreen shrub    All biomes
Deciduous shrub    All biomes
Warm (C4) grass  >22 (P**) All biomes
Cool (C3) grass    All biomes
Table 2. Biomes Simulated in Agro-IBIS, Excluding Tropical Biomes and Croplandsa
BiomeBiome NameDominant PFTLAI requirement
  1. a

    Dominant PFT refers to the PFT with the highest LAI within each biome. For mixed forest and savanna, one of several possible PFTs is dominant. Biome is assigned based on the LAI requirement, where total tree LAI is the sum of all values of tree PFT LAI.

1Temperate broadleaf evergreen forestWarm-temperate broadleaf evergreen treeTotal tree LAI > 2.5
2Temperate conifer evergreen forestTemperate conifer evergreen treeTotal tree LAI > 1.5
3Temperate deciduous forestTemperate broadleaf cold-deciduous treeTotal tree LAI > 1.5
4Boreal evergreen forestBoreal conifer evergreen treeTotal tree LAI > 1.0
5Boreal deciduous forestBoreal broadleaf cold-deciduous treeTotal tree LAI > 1.0
6Mixed forestTemperate conifer evergreen tree, temperate broadleaf cold-deciduous tree, boreal conifer evergreen tree, boreal broadleaf cold-deciduous tree, boreal conifer cold-deciduous tree0.45 < fraction of total tree LAI that is deciduous < 0.55
7SavannaWarm-temperate broadleaf evergreen tree, temperate conifer evergreen tree, temperate broadleaf cold-deciduous tree0.5 <total tree LAI < LAI requirement for biomes 1–3
8GrasslandAny PFT(total tree LAI < LAI req. for biomes 1–5) and (grass LAI > shrub LAI)
9Dense shrublandAny PFT(total tree LAI < LAI req. for biomes 1–5) and (shrub LAI > grass LAI)
10Open shrublandAny PFTTotal LAI < 1.0

[11] To calibrate the model over the study region, the bioclimatic limits controlling the suitable climatic domains for PFTs were changed to correspond to mean monthly temperatures rather than annual extreme temperatures (Table 1). This change was necessary for the model to simulate the deciduous and conifer forest biomes representative of the region [Curtis, 1959]. Additionally, a new heat stress limit of 22 °C for boreal plant functional types was needed to limit the southern extent of the boreal evergreen forest biome. Representation of wildfire was also added to the model, following the glob-FIRM model developed by Thonicke et al. [2001].

[12] The ability of Agro-IBIS (and its predecessor IBIS) to simulate exchanges of carbon, energy, and water within natural ecosystems has been tested against flux measurements at several sites, including three temperate forest sites [Kucharik et al., 2006] and three boreal forest sites [El Maayar et al., 2001]. In those studies, IBIS was shown to adequately simulate ET, net ecosystem exchange and production, as well as basic differences between deciduous and coniferous forests, including representative PFTs and Bowen ratios. Validation studies for other natural ecosystems have demonstrated adequate representation of surface fluxes at the site level [Delire and Foley, 1999], as well as vegetation distribution, patterns in runoff, NPP, biomass, LAI, soil carbon, and total CO2 flux at a global scale [Foley et al., 1996; Kucharik et al., 2000]. Groundwater recharge evaluation for IBIS was performed in a study by Dripps and Bradbury [2007] for forests in the Trout Lake region of northern Wisconsin. They compared three independent models of GWR, including IBIS, to observations. IBIS was able to successfully simulate annual GWR over a 5 year period.

[13] Agro-IBIS and IBIS have been used extensively to study terrestrial biosphere-atmosphere interactions [Delire et al., 2004, 2008; Govindasamy et al., 2005; Snyder, 2010; Snyder et al., 2004; Soylu et al., 2011; Twine et al., 2004]. More information on the detailed structure of Agro-IBIS can be found in Kucharik [2003] and Kucharik et al. [2000].

2.2 Model Inputs

[14] Climate inputs required at each 60 minute time step for each grid cell include solar radiation, Tmax, Tmin, P, RH, and wind speed. ZedX Inc. (Bellefonte, PA) developed a daily gridded weather data set at 10 km resolution (0.08333°) for a 60 year period from 1948 to 2007, including all six variables needed as model input. The spatial extent of this data set spanned from 24°N to 52°N latitude and 50°W to 130°W longitude. Three data sets were used to generate the gridded maximum and minimum daily temperatures and precipitation. Input station data for Canada and Mexico were obtained from the Global Historical Daily Climatology (GHCND) database, and the National Climatic Data Center (NCDC) TD3200 and TD3210 station data were used for the United States. For variables of relative humidity and wind speed, the Global Summary of the Day (GSOD) data were used to generate the daily gridded data. A combination of daily average temperature and daily average dew point was used to generate relative humidity. The 10 km gridded data of solar radiation were produced using coarser resolution NCEP/NCAR reanalysis 1 data [Kalnay et al., 1996] and the NCEP/DOE AMIP 2 reanalysis data [Kanamitsu et al., 2002]. Finally, hourly variations in climatic variables were simulated through the use of empirical formulations that relate temperature, specific humidity, precipitation, and radiation variability [Campbell and Norman, 1998].

[15] Land surface inputs at model initialization include soil textural class at each soil layer to a depth of 250 cm. The dominant soil texture of each soil layer in each grid cell was chosen from 11 possible categories based upon percent sand, silt, and clay [Campbell and Norman, 1998]. These data were derived from the USDA State Soil Geographic Database (STATSGO) 1 km resolution data set [White and Miller, 1998]. This data set was previously aggregated to a 5 min × 5 min (0.0833° × 0.0833°) terrestrial grid. The standard thicknesses of the 11 soil layers are 5 cm (layers 1 and 2), 10 cm (layers 3–5), 20 cm (layers 6–8), and 50 cm (layers 9–11). From the assignment of a textural category in each grid cell and each soil layer, the porosity, field capacity, wilting point, saturated air-entry potential, hydraulic conductivity, and moisture release curve “b” coefficient are obtained from a look-up table [Campbell and Norman, 1998].

2.3 Study Region and Model Simulations

[16] The study region included the geographic expanse of the Upper Midwest U.S. between 41.6° and 49.5°N latitude, and −97.3° and −82.2°W longitude. The states of Minnesota, Wisconsin, and Michigan were encompassed in the study region as well as portions of Iowa, Illinois, South and North Dakota, and Nebraska. The Great Lakes and Canada were not part of the study region.

[17] Four model runs were conducted, beginning with a 300 year “CO2” spin-up simulation that was initialized with bare soil (Table 3). The purpose of this run was to allow the model to begin vegetation growth and soil carbon accumulation [Kucharik et al., 2000] as well as have atmospheric CO2 increase according to historic records from 1800 to 1947. An additional 800 year “EQUIL” run was conducted in series in order to allow the model to come to equilibrium in response to the increase in CO2. Equilibrium was judged based on when boreal forests, the slowest growing of all biomes within the region, reached an equilibrium state for NPP and total biomass. The control and experimental runs were initialized using the results of the EQUIL run.

Table 3. Model Runs Performed for This Study
Run NameDescriptionTime PeriodClimate DataCO2
CO2 SPINCO2 is ramped between 1800 and 1947 according to historic observations.1648–1947 (300 years)Random sequence of climate data years chosen between 1948 and 19701648–1799 CO2 = 280 ppm, 1800–1947 CO2 = ramped
EQUIL SPINContinuation of the CO2 SPIN run, with CO2 held fixed at 311 ppm (observed value for 1948). Model reaches equilibrium.(800 years)Random sequence of climate data years chosen between 1948 and 1970All years CO2 = 311 ppm
CONTROLAnthropogenic climate change signal is avoided by using climate years before 1970.1948–2007 (60 years)Random sequence of climate data years chosen between 1948 and 1970All years CO2 = 311 ppm
CLIMATE CHANGEAnthropogenic climate change signal is not avoided.1948–2007 (60 years)Climate driver data run in sequence, 1948 and 2007All years CO2 = 311 ppm

[18] For the experimental control run, we aimed to simulate all ecosystem processes in the absence of any anthropogenic climate change signal. To do this we ran the model using randomly selected years of climate data between the years 1948 and 1970 for a total of 60 years. This procedure was also followed for the spin-up runs. The experimental “climate change” (CC) run used the climate data set in full, sequential form from 1948 to 2007. Atmospheric CO2 concentration was held constant at mid-20th century values (311 ppm) for both control and CC runs. The effect of rising atmospheric CO2 during the second half of the 20th century was thus not considered in this study. Natural fire disturbance was turned on for all simulations.

2.4 Model Validation of ET

[19] We tested the ability of Agro-IBIS to simulate ET within the study area by comparing to observations from the Park Falls US-PFa Fluxnet site, located in northern Wisconsin (45.946°N, −90.272°W). Flux tower data consisted of daily estimates of latent heat flux over the 1996–2000 time period [Davis et al., 2003]. The Park Falls site is characterized as a mixed forest, containing 70% deciduous tree species and 30% coniferous species. Wetlands comprise about 40% of the surrounding landscape. The model was run over one grid cell corresponding to the location of the tower. Given the dominance of deciduous tree species at the site, the model was run with the vegetation type fixed as temperate broadleaf deciduous tree (DEC). To test the sensitivity of the simulation to vegetation type, two additional runs were performed. In the first, the vegetation type was fixed as boreal conifer evergreen tree (EVG). In the second, vegetation was set to include DEC as well as shrubs and grasses. A final run was performed with vegetation type fixed as DEC, and LAI of the upper canopy fixed at 5.0.

2.5 Climate Change Analysis

[20] Annual and seasonal linear trends were calculated for Tmax, Tmin, P, RH, solar radiation as well as number of days with maximum temperature exceeding 32.2 °C (hot days), number of days with minimum temperature below −17.8 °C (cold days), day of last spring freeze, day of first fall freeze, growing season length, growing degree days base 5 °C (GDD5), and date of spring onset, approximated by when the 10 day running mean temperature reached 10 °C [Kucharik et al., 2010]. Seasons referred to meteorological seasons, with winter representing the months December through February, spring representing March through May, summer representing June through August, and fall representing September through November. Annual and seasonal trends were presented as total change in each variable over the 60 year time period. Total change was calculated as the slope of the linear regression multiplied by 60, the number of years in the CC simulation. A t-test was used to compute statistical significance for each grid cell, with significance defined as having a p-value less than 0.1. All reported climate trends include the estimated standard error. All statistics in this study were performed using the Statistics Toolbox in MATLAB [MATLAB, 2010].

2.6 Analysis of Changes in Biome Distribution, NPP, and Hydrology

[21] For each year of the simulation, the model assigned a biome to each grid cell based on the present PFTs (Table 1) and corresponding values of LAI (Table 2). An average biome distribution was generated by selecting the most frequently occurring biome in each grid cell over the 60 year CC run. The most common PFTs to the region included temperate broadleaf deciduous tree (DEC), boreal conifer evergreen tree (EVG), and evergreen shrub (SHRUB) (Table 1). DEC, which grew in the southern part of the study region, was assigned three separate bioclimatic limits within Agro-IBIS: a minimum tolerable limit, a chilling requirement limit, and a GDD requirement [Prentice et al., 1992; Sitch et al., 2003]. EVG was assigned a minimum tolerable limit, a chilling requirement limit, and a GDD limit, plus an additional heat stress limit [Sitch et al., 2003] that is responsible for limiting the southern extent of the boreal PFT in the study region. SHRUB was not assigned bioclimatic limits. Minimum tolerable limits, chilling requirements, and heat stress limits were all applied using a 30 year running average.

[22] To assess changes in simulated NPP and hydrological quantities over time, trends and corresponding p-values were calculated for each grid cell using simple linear regression and the t-test for statistical significance. Trends were calculated for both the control run and the CC run, and a final slope was calculated by subtracting the control run from the CC run. A total change was computed by multiplying the final slope of each grid cell by 60 years, the length of the CC simulation. All region-wide model trends are reported with the estimated standard error.

2.7 Attribution Analysis

[23] We used simple linear regression and partial correlation analysis to assess the influence of individual climate variables on model output. Coefficients of determination (R2) were calculated between all individual climate variables and each model variable. Values of R2 were ranked from greatest to least for each model variable. The three climate variables having the largest averaged R2 were selected as possible drivers for each model variable. Coefficients of partial correlation (Rpartial) were then calculated between each model variable and its corresponding set of three driver variables. A spatially averaged value of R2partial was used to determine the most influential climate driver(s) for each model variable. Spatial averages were calculated for each PFT using only those grid cells in which the PFT grew.

3 Results

3.1 Model Validation of ET

[24] A comparison of Agro-IBIS simulated ET and flux tower estimates of latent heat flux from the US-PFa Fluxnet site is shown in Figure 1. In general, Agro-IBIS captured seasonal variation in ET but consistently underestimated ET, especially during the growing season. Simple linear regression between observed and simulated monthly ET yielded an R2 of 0.90 (Figure 1b). A possible explanation for the systematic underestimation of ET may be the presence of wetlands near the US-PFa site [Baker et al., 2003; Mackay et al., 2002]. Saturated soils and standing water of the nearby wetlands could have contributed additional evaporation throughout the year. Since the model assumed the site to be a forest (and does not consider nearby land), it would lack this additional contribution. Simulated ET was not found to be sensitive to PFT, as separate runs using (1) EVG and (2) a combination of DEC and shrubs and grasses showed very little change from the original validation run using DEC in isolation. A final run with DEC only and LAI fixed to 5.0 (a value greater than what was simulated but consistent with productive forests) showed no substantial change from the original validation run. This implied that the underestimation in ET was not likely due to an underestimation in productivity.

Figure 1.

Simulated and observed monthly ET for the US-PFa Fluxnet site in northern Wisconsin. ET versus time (a); simulated versus observed ET (b).

3.2 Regional Climate Trends

[25] Trends in climate variables showed spatial variation across the Upper Midwest (Figure 2 and Table 4). Winter and spring seasonal trends in Tmax showed a similar spatial pattern of warming. Winter warming was especially pronounced in Minnesota where most trends were statistically significant and ranged from 2 to 3 °C. Summer and fall trends in Tmax showed pervasive cooling trends ranging from 0 to −2 °C. On an annual time scale, trends were spatially variable, with warming the dominant trend in places like central and northern Minnesota, and cooling the dominant trend in eastern Wisconsin and central Michigan.

Figure 2.

Seasonal trends in (a–e) maximum temperature (Tmax, °C), (f–j) minimum temperature (Tmin,°C), (k–o) precipitation (mm), (p–t) relative humidity (%), and (u–y) solar radiation (W m−2), from 1948 to 2007. Dotted grid cells were statistically significant (p < 0.1).

Table 4. Summary of Climatic Changes Occurring Over the Study Region from 1948 to 2007a
Climatic VariableAverage 1948–1977Average 1978–2007Total Change 1948–2007Standard ErrorPercentage Grid Cells p < 0.1
  1. a

    Average, total change, and standard error were spatially averaged over all grid cells in the study region. Total change was calculated as the slope of the linear regression multiplied by 60 years.

Average Annual Temperature (°C)6.86.90.50.434.5
Average Winter Temperature (°C)−8.2−7.31.41.040.9
Average Spring Temperature (°C)6.36.71.00.640.7
Average Summer Temperature (°C)20.220.10.00.54.3
Average Fall Temperature (°C)8.78.4−0.40.64.9
Annual Maximum Temperature (°C)12.712.70.20.417.0
Winter Maximum Temperature (°C)−3.0−2.30.90.923.1
Spring Maximum Temperature (°C)12.412.91.00.825.1
Summer Maximum Temperature (°C)26.726.4−0.50.524.5
Fall Maximum Temperature (°C)14.514.0−0.70.68.2
Annual Minimum Temperature (°C)0.81.20.80.459.7
Winter Minimum Temperature (°C)−13.4−12.31.81.153.5
Spring Minimum Temperature (°C)0.10.51.00.649.3
Summer Minimum Temperature (°C)13.613.90.50.530.5
Fall Minimum Temperature (°C)2.92.7−0.10.55.0
Total Days Above 32.2°C (90°F)8.46.5−4.02.827.3
Total Days Below −17.8°C (0°F)30.226.5−8.24.747.1
Date of Last Spring Freeze (days)128.5125.2−6.14.933.8
Date of First Fall Freeze (days)275.5276.11.74.333.0
Growing Season Length (days)146.9150.97.86.439.6
Growing Degree Days (base 5C)2278.42264.8−9.171.111.9
Spring Onset Date (days)118.0116.2−5.35.122.6
Annual Precipitation (mm)720.8757.874.253.541.8
Winter Precipitation (mm)85.183.30.514.05.6
Spring Precipitation (mm)194.0194.714.825.14.6
Summer Precipitation (mm)272.2284.38.730.113.5
Fall Precipitation (mm)168.3195.249.529.055.3
Annual Solar Radiation (W m−2)209.2207.9−1.61.641.3
Winter Solar Radiation (W m−2)113.1113.30.61.449.3
Spring Solar Radiation (W m−2)256.2256.2−0.53.319.4
Summer Solar Radiation (W m−2)302.9299.4−2.23.710.9
Fall Solar Radiation (W m−2)164.8162.5−4.12.067.3
Annual Relative Humidity (%)69.470.31.51.047.7
Winter Relative Humidity (%)74.475.81.91.346.8
Spring Relative Humidity (%)65.464.8−1.21.627.1
Summer Relative Humidity (%)67.169.03.21.573.2
Fall Relative Humidity (%)70.671.82.21.537.1

[26] Trends in mean annual Tmin exhibited more homogeneity than with maximum temperature, with warming being the dominant direction of change. Warming of winter Tmin was especially pronounced in Minnesota and northern Wisconsin, where temperatures increased by more than 3 °C. Trends in spring and summer Tmin also revealed a warming pattern within the study region, but at magnitudes generally less than those in winter. Trends during fall were more heterogeneous compared to the other seasons.

[27] Annual P trends showed an overall increase during the 60 year period. Seasonal P trends exhibited more spatial heterogeneity than annual trends with the exception of fall. Most trends in winter and spring P were not statistically significant, but some locations had significant trends during summer. Northern Wisconsin and the western Upper Peninsula of Michigan, for example, experienced a significant decrease in summer P by as much as 80 mm. In contrast, some locations in southern Wisconsin and southern Michigan experienced an increase in summer P by as much as 90 mm. Trends in fall P increased in most places, with some areas undergoing a total change in excess of 90 mm.

[28] Trends in RH had smoother spatial patterns compared to temperature and P. An increase in annual RH was observed, with the largest increases around 5%. Most seasonal trends within the study area were positive except for spring, where RH decreased in most places. Summer and fall RH increased nearly everywhere, reaching 6–8% in some locations.

[29] Most of the Upper Midwest region experienced a decrease in annual solar radiation over the study period, with trends ranging from −1 to −6 W m−2. The division between increasing and decreasing solar radiation trends had a roughly northwest-to-southeast alignment during spring, summer, fall, and at the annual time scale.

3.3 Distribution of Biomes

[30] The distribution of potential vegetation biomes remained nearly static for the 60 year simulation (Figure 3). In general, the bioclimatic limit defining the southern edge of the boreal forest did not shift significantly during the study period. However, the bioclimatic limit for DEC shifted north, into the boreal evergreen forest. As simulated by Agro-IBIS, the ecotone between the boreal evergreen forest and temperate deciduous forest did not change significantly, although a handful of grid cells located along the ecotone did change biome designation during the simulation. Additionally, some occasional year-to-year changes in biome were observed between forest, shrubland, and savanna in North and South Dakota and western Minnesota. This region is located near the central US ecotone where deciduous forests transition to the grasslands of the Great Plains. This ecotone is sensitive to precipitation, which varies considerably year to year [Notaro, 2008], and might contribute to fluctuations in biome within that area. The observed year-to-year biome changes between forest, shrubland, and savanna were likely due to competition among PFTs for light and soil moisture, leading to fluctuations in LAI above and below specific biome thresholds (Table 2).

Figure 3.

Agro-IBIS simulated biome distributions for the 1948–2007 study period. Distribution was determined by choosing the most frequently occurring biome for each grid cell over the simulation. Light blue grid cells denote boreal evergreen forest, green grid cells denote temperate deciduous forest, and orange grid cells denote dense shrubland. Green and blue curves show bioclimatic limits for the temperate broad-leaved deciduous tree (DEC) and boreal conifer evergreen tree (EVG) PFTs, respectively. Solid curves indicate the average location of the limits for the period 1948–1977, and dashed curves indicate the period 1978–2007. The limit for DEC is a minimum tolerable limit corresponding to a coldest mean monthly temperature of −15 °C. The limit for EVG is a heat stress limit corresponding to a warmest mean monthly temperature of 22 °C.

[31] Within the boreal evergreen forest biome, EVG was the dominant PFT for the entire simulation, comprising nearly 100% of the upper canopy, including the part of the biome overlapping with the climatic domain of DEC. Despite adequate climatic conditions, lower canopy PFTs including shrubs and cool (C3) grasses did not grow within the boreal evergreen forest. Within the temperate deciduous forest where DEC was the dominant PFT, the only other PFT to grow was evergreen shrub (SHRUB). A small number of grid cells with a biome classification of dense shrubland were located in southeast South Dakota, extending slightly into Minnesota. Shrubland appeared here because for the beginning half of the simulation neither DEC nor EVG could grow due to bioclimatic limitation. Toward the end of the simulation, with the northward migration of the minimum tolerable bioclimatic limit for DEC, deciduous forest invaded most of this small region of shrubland.

3.4 Changes in NPP

[32] Trends in total NPP, which include both above- and below-ground NPP, increased throughout the region (Figure 4). On average, total NPP increased by 41 ± 30 g C m−2 across all grid cells, corresponding to a region-wide increase of 8%. Increases in NPP were larger in the temperate deciduous forest than in the boreal evergreen forest. Within some grid cells located along the TZ there were negative trends in total NPP. This was due to the less productive EVG outcompeting the more productive DEC within these grid cells. This occurred as a result of a small southward migration of the 22 °C heat stress limit for boreal PFTs and indicated that EVG was able to outcompete DEC once climatic conditions became favorable.

Figure 4.

Agro-IBIS simulated trends in annual NPP (g C m−2) during the 1948–2007 period for (a) total grid cell NPP, (b) temperate broad-leaved deciduous tree (DEC), (c) boreal conifer evergreen tree (EVG), and (d) evergreen shrub (SHRUB). Trends were calculated as the slope of the linear regression multiplied by 60 years. Dotted grid cells had linear trends that were statistically significant (p < 0.1).

[33] Most grid cells within the temperate deciduous forest had an increase in NPP, with an average total change of 96 ± 35 g C m−2 (25%) over the 60 year study period (Figure 4b).

[34] Within the boreal evergreen forest, NPP trends were positive, but lower in magnitude than for the temperate deciduous forest, and fewer trends were statistically significant (Figure 4c). On average, NPP within the boreal evergreen forest increased by 28 ± 14 g C m−2 (11%). Evergreen shrub (SHRUB), the only PFT in the lower canopy to actively grow within the study region, experienced a widespread average decline in NPP of −31 ± 19 g C m−2 (−16%) (Figure 4d).

3.5 Changes in Hydrology

[35] Figure 5 shows Agro-IBIS simulated trends in ET, GWR, surface runoff, and available water. Trends in annual ET were heterogeneous in space, with 49% (51%) of all grid cells showing an increase (decrease). Trends in ET across the study region ranged from approximately −100 to +150 mm, with an average total change of 5 ± 21 mm (1%) across all grid cells.

Figure 5.

Agro-IBIS simulated trends in (a) evapotranspiration, (b) groundwater recharge, (c) runoff, (d) and available water (P − ET) during the 1948–2007 period (mm yr−1). The total change for each variable was calculated as the slope of the linear regression multiplied by 60 years. Dotted grid cells had linear trends that were statistically significant (P < 0.1).

[36] Simulated trends in annual GWR (Figure 5b) exhibited a more homogeneous spatial pattern than for ET, with 80% of grid cells showing positive gains. Trends in GWR averaged 39 ± 35 mm (45%), ranging from approximately −100 to +200 mm. As with ET, simulated trends in surface runoff (Figure 5c) were heterogeneous in space, with 48% (52%) all grid cells having positive (negative) trends. Trends in runoff averaged 0 ± 3 mm (1%) and ranged from −10 to +8 mm.

[37] Available water, calculated as precipitation minus evapotranspiration (P − ET), provides a good indicator of soil moisture available to plants. Trends in available water (Figure 5d) showed increases in 94% of all grid cells, indicating that average soil moisture storage throughout the region increased over the 60 years. Trends in available water averaged 59 ± 44 mm (125%), ranging from approximately −50 to +200 mm.

[38] We also analyzed trends for the ratio of ET to precipitation (ET/P) (Figure 6a). This quantity decreased over the study period in 94% of all grid cells, with the total change ranging from approximately −0.25 to 0.05 and averaging −0.08 ± 0.06 (−8%). The ratio of GWR plus runoff to P (Figure 6b) increased in 94% of grid cells. The average total change was 0.06 ± 0.04 (64%), with a range of −0.05 to 0.2.

Figure 6.

Agro-IBIS simulated trends in (a) ET/P and (b) (runoff + groundwater recharge)/P during the 1948–2007 period. The total change for each ratio was calculated as the slope of the linear regression multiplied by 60 years. Dotted grid cells had linear trends that were statistically significant (P < 0.1).

3.6 Climate Drivers

[39] Climate drivers for each model variable were assessed using spatially averaged values of R2partial (Figure 7 and Table 5). All hydrologic variables were driven primarily by annual P. Groundwater recharge was especially correlated with annual P within the boreal evergreen forest. For total NPP, annual P and summer RH were the primary climate drivers, having comparable values of R2partial (0.29 and 0.27, respectively). Correlations between total NPP and annual P were stronger within the boreal evergreen forest, whereas correlations between NPP and summer RH appeared unaffected by biome. For DEC, the top climate drivers were also summer RH and annual P, both having comparable values of R2partial (0.39 and 0.36, respectively). The central part of the study region showed a particularly strong correlation between summer RH and DEC NPP. For EVG, the top climate drivers were summer RH and summer Tmax (0.24 and 0.16, respectively).

Figure 7.

Partial R2 between modeled ecosystem variables and their respective climatic drivers for the 1948–2007 period. ET and yearly precipitation (a); groundwater recharge and yearly precipitation (b); runoff and yearly precipitation (c); total NPP and yearly precipitation (d); total NPP and summer relative humidity (e); temperate broadleaf deciduous tree NPP and summer relative humidity (f); temperate broadleaf deciduous tree NPP and yearly precipitation (g); boreal conifier evergreen tree NPP and summer relative humidity (h); boreal conifer evergreen tree NPP and maximum summer temperature (i).

Table 5. Climatic Drivers for NPP and the Hydrologic Budget
Model VariableDriver 1R2partialDriver 2R2partial
Total NPPAnnual precipitation0.29Summer relative humidity0.27
NPP for temperate broadleaf deciduous treeSummer relative humidity0.39Annual Precipitation0.36
NPP for boreal conifer evergreen treeSummer relative humidity0.24Maximum summer temperature0.16
ETAnnual precipitation0.13Summer precipitation0.09
Surface runoffAnnual precipitation0.46Annual relative humidity0.09
Groundwater rechargeAnnual precipitation0.39Spring precipitation0.04

[40] All partial correlation coefficients between climatic quantities and SHRUB NPP were approximately zero. Further analysis of NPP trends in SHRUB and DEC showed these two model variables were negatively correlated (R2 = 0.40). From these results we inferred that competition between the two PFTs was responsible for suppressing SHRUB NPP and that climate was not a direct influence on SHRUB.

4 Discussion

4.1 Regional Climate Change

[41] The 60 year winter warming trend within the study region supports future projections for significant wintertime warming in the Upper Midwest [IPCC, 2007] and also agrees with previous Midwest climate change studies [Hayhoe et al., 2010; Kucharik et al., 2010]. In contrast, the recent 60 year trend of cooling summertime Tmax and a lower frequency of hot days is inconsistent with projected warming of 3–4 °C during summer by 2050 [IPCC, 2007] and projected increases in the frequency of hot summer days [Hayhoe et al., 2009]. The observation for summertime cooling is however consistent with previous observations of a “warming hole,” or localized area of summer cooling, in the central U.S. [Kunkel et al., 2006; Pan et al., 2004]. Using a regional climate model, Pan et al. [2004] linked the “warming hole” phenomenon to changes in low level circulations during summer that increased soil moisture and ET and suppressed daytime maximum temperatures. While our results did not show uniform changes in summer ET over the study region, trends in increased summer RH, decreased summer solar radiation, and increased available water do support the findings of Pan et al. [2004]. These climatic trends contributed to widespread increases in NPP in our simulations via a reduction in soil moisture stress and an increase in stomatal conductance in trees during summer.

4.2 Climate Impacts on NPP

[42] The important roles of summer RH and annual P on NPP support the notion that both of these climatic quantities are physically linked. However, their impacts on NPP were variable across space, evidenced by different spatial patterns of correlation (Figure 7d–7g). Within the boreal evergreen forest, for example, total NPP was weakly correlated with annual P, whereas it was moderately correlated with summer RH (Figure 7d and 7e). This suggests that RH can affect ecosystems in a manner that is independent from P. The important role of increased summer RH raises an interesting question regarding the source of influential climate change in the Midwest region. If, as some studies suggest, intense agricultural production of the Midwest Corn Belt is responsible for the observed increase in summertime humidity [Changnon et al., 2003], this implies that human land use practices can have a significant effect on local climate as well as surrounding ecosystems. Another implication is that local-scale forcings have the potential to alter regional- or global-scale forcings, perhaps confounding future climate change projections.

[43] There were two biological mechanisms by which climate change increased NPP in Agro-IBIS simulations. The first was a reduction in soil moisture stress that led to an increase in photosynthetic rate. The increase in water availability occurred as a result of increased P and perhaps a decrease in potential evapotranspiration (PET) attributable to an increase in summer RH, a decrease in summer Tmax, and a decrease in summer radiation (Figures 2s, 2d, and 2x). The second mechanism was related to the impact of increased summer RH on stomatal conductance, which is positively correlated with photosynthetic capacity [Wong et al., 1979]. The empirical model for stomatal conductance used within Agro-IBIS, based on Collatz et al. [1991], includes a numerator term containing leaf surface humidity, indicating that increased humidity at the leaf surface increases stomatal conductance and subsequently C assimilation and NPP. Humidity at the leaf surface is calculated within the canopy physics sub-model using specific humidity, which is obtained by converting the input RH driver data. Humidity at the leaf surface is then converted back to RH before being passed to the stomatal conductance sub-model. Thus, the value of RH that is used to compute stomatal conductance is originally a value of specific humidity that is adjusted according to environmental conditions and leaf processes within the canopy. Other models of stomatal conductance predict an increase in photosynthesis with increased humidity, including the well-used Jarvis-Stewart model [Jones and Higgs, 1989; Stewart, 1988].

[44] Twine and Kucharik [2009] investigated trends in NPP for crops as well as deciduous and conifer forests of the eastern U.S. over the 1950–2002 time period. Using the Agro-IBIS model over a larger domain at coarser spatial resolution (0.5° × 0.5°), they observed an increase in deciduous forest annual NPP of 26.5 g C m−2 over the 53 year period. This value compares reasonably well with our simulated change in total NPP over the Upper Midwest from 1948 to 2007 of 41 g C m−2. Twine and Kucharik [2009] attributed the increase in NPP to increased summer P. Our results also showed the important role of P in driving these trends, yet because we considered additional climatic drivers, we found that summer RH also played a dominant role in driving changes in NPP. Twine and Kucharik [2009] found no significant change in NPP for conifer forests in the northern U.S., in rough agreement with our observations of small changes in NPP for EVG.

[45] Hicke et al. [2002] generated trends of NPP for North America using satellite records from 1982 to 1998. They observed increases in total annual NPP over the 17 year period of approximately 39.1 g C m−2 for deciduous broadleaved forests of the eastern U.S., and 34 g C m−2 for mixed forests located near the temperate-boreal transition. We calculated an increase in total NPP over the Upper Midwest from 1948 to 2007 to be 41 g C m−2, with the first 30 years (1948–1977) having a trend of −16.5 g C m−2 and the latter 30 years having a trend of 32.2 g C m−2. Between 1982 and 1998, the time period corresponding to the analysis of Hicke et al. [2002], we calculated the trend in total annual NPP to be 18.8 g C m−2. Our results differed from Hicke et al. [2002] over the 1982–1998 period by about a factor of two. This may imply that half of the increase in deciduous forest NPP was due to climate change and the remaining half due to factors we did not consider that promote increases in NPP, such as CO2 fertilization, land-use change, and nitrogen deposition [Caspersen et al., 2000; Friedlingstein et al., 1995; Townsend et al., 1996]. The difference in trends between our study and Hicke et al. [2002] may also be attributed to different spatial domains. As with Twine and Kucharik [2009], Hicke et al. [2002] attributed increasing trends in deciduous tree NPP primarily to an increase in P. Our findings corroborate this result, yet because we considered additional climatic variables we found that summer relative humidity could play an equally important role in driving future changes in NPP.

4.3 Climate Impacts on Hydrology

[46] Among the PFTs growing in the region, there were notable differences in water utilization (Figure 8). This was attributed to the differences in plant physiology as well as typical growing season temperatures within the two biomes. Because the maximum rate of photosynthesis for DEC is slightly greater than for EVG, it is likely that DEC was better able to utilize increases in P—especially during the spring and summer—in comparison to EVG. Net primary productivity for EVG was less responsive to increased P. Ground water recharge was more strongly correlated with annual P in the boreal evergreen forest than the temperate deciduous forest. Reasons for this may include that EVG was not able to utilize additional rainfall as well as DEC and that cooler temperatures in the north were associated with lower levels of evaporation.

Figure 8.

Scatter plot of annual ET (mm yr−1; y axis) versus annual precipitation (mm yr−1; x axis) for all grid cells and all years from 1948 to 2007. Black data points denote temperate deciduous forest and gray data points denote boreal evergreen forest. Light gray curves are functional fits the distributions.

[47] Our results suggest that if annual P were to continue to increase, ET, GWR and surface runoff would also increase. Annual precipitation is expected to increase over the next several decades in the Upper Midwest [IPCC, 2007], yet actual hydrologic partitioning will depend on interactions between precipitation, soil moisture storage, and vegetation dynamics [Brooks et al., 2011; Rodriguez-Iturbe et al., 1999]. Seasonal predictions of precipitation are difficult to assess, yet some evidence suggests that summers in the Midwest will become drier [Hayhoe et al., 2010]. Given that summer coincides with the growing season and also provides a large fraction of the annual precipitation budget, this could have important implications for both natural and managed ecosystems of the region.

[48] Given the differential response of groundwater recharge within the two biomes of the study region, plant productivity appears to play a role in the amount of GWR that occurs. This suggests that vegetation can play a potential role in mitigating groundwater and surface flooding, which has been a problem in the Upper Midwest in recent years [Changnon and Kunkel, 1999; Pielke and Downton, 2000; Villarini et al., 2011]. Given that groundwater recharge was shown to increase considerably for the case of potential vegetation, increased water availability and flooding may remain a problem within the Upper Midwest. This might particularly be true in many agricultural areas where vegetation is only present during the growing season in the absence of cover crops.

4.4 Implications for Coupled Carbon and Water Cycling

[49] Despite significant increases in NPP across the region from 1948 to 2007 (Figure 4), there was still an overwhelming increase in plant available water throughout most of the region (Figure 5d). In general trends in the ratio of ET/P declined over most areas whereas the ratio of GWR and runoff to P increased dramatically. We offer three reasons for these results. First, given typical rates of daily ET (growing season) in the range of 3–5 mm day−1 for these natural plant communities, about 20–40 mm of weekly rainfall is required to replenish water in soil that is taken up by plants. However, spring and summer rainfall across this region can be delivered in sporadic, often extreme events, sometimes when soils are saturated. This can lead to excess water moving below the root zone and being unavailable to plants. A second explanation is that there is an ongoing trend of more plant available water than needed for maximum rates of growth in some regions, and thus, even under optimal conditions, plants are not able to fully utilize all available water. Lastly, some of the increases in precipitation are occurring outside of the growing season when there is little plant growth as well as minimal evaporative demand. This suggests that with increases in precipitation in the future, and in particular increased frequencies of extreme rainfall events [IPCC, 2007], natural vegetation may have a reduced capacity to support flood mitigation services in some regions.

[50] Given that annual precipitation is expected to increase in the future, we can ask whether or not NPP should be expected to also continue to increase. Figure 9, which shows the anomaly in total NPP as a function of annual precipitation (for all grid cells and all years of the CC simulation), suggests that increased precipitation may only support increased NPP to a limited extent within the region. This is a plausible result given the link between P, cloud cover, and radiation, and is supported by our observations for increased GWR and available water. More P likely involves a higher frequency of cloudy days, and at some point excess cloud cover would be expected to negatively impact NPP.

Figure 9.

Scatter plot of percent anomaly in annual NPP (%; y axis) versus annual precipitation (mm yr−1; x axis). NPP percent anomaly was calculated for each grid cell as the difference between yearly NPP and average NPP over the study period (1948–2007), divided by the average NPP. Vertical gray line denotes average annual precipitation for 1948–2007. Light gray curve is a functional fit to the distribution.

4.5 Model Limitations

[51] In this study, the model was able to simulate changes in climatic suitability for PFTs as well as the distribution of dominant PFTs, yet could not achieve realistic PFT diversity within grid cells. This was evidenced by the fact that very few grid cells had more than one tree PFT present at a time, including the area of overlapping climatic domains for DEC and EVG. Furthermore, there was a noticeable absence of C3 and C4 grasses in the southern regions of the study area, and a corresponding representation of savanna and grassland biomes that once dominated portions of the landscape. This lack of diversity limited the ability of the model to address competitive interactions between PFTs and thus any shift of the TZ resulting from competition. This limitation may stem from the fact that Agro-IBIS, like other models that simulate dynamic vegetation, does not account for species-level dynamics, many of which play an important role in vegetation migration and composition. These include small-scale disturbance [Sykes et al., 2001], seed dispersal, and life-history traits [Neilson et al., 2005]. An interesting question lies in the exact spatial scales at which DGVMs can adequately capture vegetation diversity, and whether or not errors occur as a result of missing ecological processes, or processes that are poorly represented.

[52] Despite poor performance in plant diversity at the grid-cell level, the model achieved a realistic distribution of vegetation biomes, mainly due to the use of appropriate bioclimatic limits. Between the periods 1948–1977 and 1978–2007, DEC experienced a northward expansion in its suitable climatic domain (Figure 3). EVG however did not undergo a shift in its domain, contrary to the expectation for a general northward retreat of boreal plant types [Soja et al., 2007]. This was mainly due to the fact that summer heat stress is responsible for limiting the southern extent of boreal PFTs within Agro-IBIS, and summer Tmax over the study period declined.

[53] We did not account for rising atmospheric CO2 in our study. The fertilization theory predicts that CO2 enrichment will increase water use efficiency and photosynthesis in plants [Ainsworth and Long, 2005; Long, 1991]; therefore our results may underestimate trends in NPP. The extent of underestimation is unclear however, given that the long term effects of CO2 enrichment are uncertain [Ainsworth and Long, 2005; Norby et al., 2005]. This research topic will be explored in future work.

5 Conclusion

[54] Our results indicate that climate change over the Upper Midwest from 1948 to 2007 was spatially variable, having little effect on biome distribution but significant impacts on coupled carbon and water exchange in natural ecosystems. Even though forest decline is expected within the Upper Midwest during the 21st century [Notaro et al., 2012; Scheller and Mladenoff, 2008], our results suggest that this effect has not yet occurred. Instead, our results suggest that gains in precipitation and relative humidity have enabled increased productivity throughout the region. A further result of increased precipitation was a large gain in GWR, indicating that plants were not able to use all excess available water. Future increases in P may therefore exacerbate flooding problems within the region.

[55] The temperate-boreal ecotone is expected to migrate north with warming, yet our results, based on climatic thresholds alone, suggest that this has not yet occurred within the Upper Midwest. The Tension Zone was not observed to shift over the study period, mainly because summers remained cool and boreal species were not negatively impacted. Continued research is needed however to assess if and when changes in climate become unfavorable for ecosystems within the region, and what impacts those changes may have.

[56] Summer relative humidity was an important driver of year-to-year variability for most NPP quantities. The significant role of humidity in driving increases in NPP was unexpected, mainly because previous regional studies have suggested that temperature and precipitation have been the key drivers of ecosystem change in the Midwest [Kucharik and Serbin, 2008; Twine and Kucharik, 2009]. Our study highlights the importance of considering other climatic quantities in addition to temperature and precipitation when studying the effects of climate change on ecosystems.

Acknowledgments

[57] This work was supported by the U.S. Department of Energy's Office of Science through the Midwestern Regional Center for the National Institute for Climatic Change Research at Michigan Technological University, under Award DE-FC02-06ER64158. We are grateful for helpful comments from Michael Notaro, John W. Williams, and an anonymous reviewer. We also thank Cécile Ané for statistics consulting.