Site description
The primary grape producing areas in Michigan are located in the west coast of the lower peninsula spread throughout four American Viticultural Areas, which are geographically defined areas that designate where grapes are grown based on local geographic features. There are four AVAs within the state of Michigan, two of which are located in the southwest corner (Fennville, Lake Michigan Shore) and two in the northwest portion of the lower peninsula (Leelanau, Old Mission) (Fig. 1). The reasons for the location near the shores of Lake Michigan are: (a) the climate moderating effects of the Lake and (b) topographic influences, which allow for drainage of cold air during the spring and fall seasons. Growing vinifera grapes farther inland or in flat regions in Michigan is not recommended, as spring/fall frosts and harsh winter temperatures can combine to potentially damage or even kill vines (Zabadal and Andresen 1997). These potentially dangerous temperatures still occur in the coastal areas where vinifera grapes are grown, but site selection and vineyard management are key to mitigating these potential damages. Thus, large-scale losses are less common in these regions as compared to areas as little as 50 km inland. However, in areas close enough to the close to experience the temperature moderating effects of the lakes, growers have success growing a number of V. vinifera types including Cabernet Franc, Chardonnay, Gewurztraminer, Pinot Noir and Riesling among other vinifera and hybrid varieties.
Southwest Michigan, located around 41° North latitude, is classified as a Dfa, humid continental climate in the Köppen Climate Classification system (Köppen 1900; Geiger 1965). Northwest Michigan, located around 45° north latitude, is classified as a Dfb climate with shorter summers and colder winters than areas to the south. However, the small areas that are located within the Leelanau and Old Mission AVAs are areas located on peninsulas (the AVAs are named for their respective peninsula) in Lake Michigan and Grand Traverse Bay. The ability to grow vinifera grapes in these regions is due almost entirely to the presence of favorable microclimates where the temperature during fall, winter and spring are much warmer than surrounding areas. It is likely that the microclimates in the northwest are much more similar to the Dfa climates found in the southwest Michigan AVAs. A combination of microclimates and the close proximity of the Lake help to limit temperature extremes compared to areas farther inland due to consistent lake and land breezes (Moroz 1967).
Data collection
This research relies on future climate projections from the fifth phase of the Coupled Model Intercomparison Project (CMIP5) suite of climate models. CMIP5 was developed to answer the many questions posed by the Intergovernmental Panel on Climate Change (IPCC)’s fourth assessment report (Solomon 2007). Among the many potential improvements to the experiments, one group of hypothetical scenarios is of particular interest to our research. Included in the model experiments were four different, transient greenhouse gas (GHG) scenarios wherein the amount of global emissions of GHGs followed different potential cases. These “representative concentration pathways” (RCP) scenarios would project global GHG emissions in the coming decades using different econometric and social models where the number following “RCP” represents the increase in radiative forcing values (measured in W/m2) by the year 2100 relative to pre-industrial values (Van Vuuren et al. 2011). In one such scenario (RCP4.5), GHG emissions are reduced on a global scale at a certain point in the future on the assumption of policy action by global leaders. In another scenario (RCP8.5), very little to no action is taken. As such, these scenarios involve different reactions by global temperatures to these hypothetical GHG emission scenarios (Taylor et al. 2012).
The results of many of the CMIP5 models were released in 2013. However, one of the limitations of such a large undertaking when creating such a large model projection dataset is that model resolution has to be sacrificed limited by computational power, storage space among other factors. Most model runs in the CMIP5 models have a resolution of approximately 100 km. Projections at such a resolution is useful for continental to global scale studies, but problematic for regional scale applications. This problem, according to CMIP5’s executive summary, presents an issue with point observations and with the spatial areal issue, in that a spatially averaged value (one grid-cell) is not representative of a point observation within the grid-cell (Taylor et al. 2012). One method to manage this concern is through the downscaling of grid cells from a lower resolution to a higher resolution. The NASA Earth Exchange Downscaled Project (NEX-DCP30) is one project that downscaled a number of CMIP5 model runs down to a resolution of 800 m for the entire contiguous 48 United States. The dataset was created using the bias correction spatial disaggregation (BCSD) methodology established in Wood et al. (2004). This methodology takes in climate model data with larger pixel resolutions and time-steps and uses real-world observations and interpolation methods to bring the data down to finer resolution in both the spatial and temporal dimensions. NEX-DCP30 allows users to use future climate projections to perform environmental analysis at a manageable resolution. These projections are based on data obtained from the Parameter-elevation Relationships on Independent Slopes Model (PRISM) temperature data and this data transitions seamlessly into 32 different CMIP5 models until the year 2100 at a monthly time step for three variables: temperature max (Tmax), temperature min (Tmin) and precipitation. This downscaled analysis of the climate models allows for a high-resolution analysis of future trends. It is of particular interest in an area like the Great Lakes Region, where the land–water interface is highly difficult to resolve in models where the resolution is bigger than 50 km.
The NEX-DCP30 downscaled data was compiled from the National Climate Change Viewer (http://www.usgs.gov/climate_landuse/clu_rd/apps/nccv_viewer.asp), managed by the United States Geological Survey (USGS). The downscaled data was downloaded on a county scale and averaged over the county in focus. The two counties focused on in this study were Berrien county (southwest MI) and Leelanau county (northwest MI). These counties, combined, account for a significant portion of Michigan’s vinifera production and are likely to expand in acreage in the coming decades, potentially following the near 300 % trend in vinifera acreage growth from 2000 to 2011 (USDA-NASS 2012). The dataset, downscaled to a resolution of 800 m and then averaged over the county area, does introduce uncertainty as the data was downscaled from the much larger climate model scaled projections. However, such a fine resolution is needed for studies where microclimates are a part of the study and is crucial in a region like the Great Lakes, where the land–water interaction is either idealized or roughly estimated due to the coarse resolutions of the models. In fact, viticultural studies at large could benefit greatly from high-resolution climate model data even in areas where the water-land interaction is less important. The western United States is one such area (Jones et al. 2010) with its heavy topography, or in Portugal, an area with a multitude of microclimates in even a small area (Fraga et al. 2015).
Averaging the NEX-DCP30 data over a county is also reflective of the dataset for yields used in this study. The data obtained from the National Grape Co-operative was taken as the average from 25 plots from around the southwest portion of Michigan. The age of the vineyards was between 15 and 30 years old. Own-rooted Concord vines were trained to 1.8-m high bilateral cordons with a north–south row orientation. Vines were spaced 2.38 m in row and 3.05 m between row (Jasper and Holloway, personal communication, August 1, 2012). The use of Concord vines, rather than vinifera vines, is due to the robust size of the Concord vine phenology dataset. Data for V. labrusca grapes in Michigan exists for several decades, while the equivalent data for V. vinifera does not exist for nearly the same length of time as vinifera phenology tracking in Michigan has only been occurring in the recent past.
Experimental design
In order to describe how the continual warming trend has affected and will continue to affect Michigan’s wine grape industry, it was necessary to use data from historical sources in conjunction with future projections. First, temperatures (max, min and mean) were calculated for the growing season (1 April–31 October) for both regions considered in the study. The NEXDCP30 dataset (future projections) has data obtained from 32 model runs plus one ensemble mean of all models run in the RCP4.5 and RCP8.5 GHG emission scenarios. However, these model simulations begin in the year 1950. The historical data from these models was developed by incorporating PRISM temperature and precipitation data when the creators were using the BCSD method of downscaling (Wood et al. 2004). This downscaled data was used as historical climate in southwest and northwest Michigan in this study from 1950 to 2005. From 2006 to 2099, there were 32 different model runs for each RCP scenario, and there was one ensemble mean for all models. This historical and future projection data was then used as the input for the analysis in this paper.
This analysis includes a multilinear regression based on past climate and grape yields in order to predict future trends for grapes. Similar concepts in methodology for the application of future climate projections into statistical models for estimating past and future grape yields can be found in Lobell et al. (2006), Santos et al. (2011), Fraga et al. (2014). This multilinear regression is limited only to the southwest of Michigan, as that is where the yield data has been recorded for decades by the National Grape Cooperative. This regression calculates potential yield in future years based on both historical data and monthly and seasonal climate projections in the RCP4.5 and RCP8.5 scenarios based on five variables: average GST, growing degree day totals (GDD), potential early season frost occurrence (Frost), total season precipitation (PPT) and early season GDD accumulation (eGDD). NEXDCP30 data was used to directly obtain two data sources (GST, PPT, eGDD) and to indirectly obtain the other variables using regressions (Frost, GDD).
$$\begin{aligned} \frac{T}{acre} & = - 27.662 + (0.0778 \times PPT) + (1.578 \times GST) - (0.542 \times Frost) \\ & \quad + (0.0017 \times GDD) - (0.022 \times eGDD) \\ \end{aligned}$$
(1)
Regression model for approximation of potential yield of V. labrusca where T/acre = tons per acre of production, PPT = total season precipitation, GST = average growing season temperature, GDD = growing degree day total, Frost = potential early season frost occurrence, and eGDD = early season GDD accumulation.
This regression is prone to uncertainty due to the data sources (downscaled climate model data), but the goal of using this model is to look at how the RCP4.5 and RCP8.5 scenarios potentially affects grape yields under future scenarios in southwest Michigan.
The study also includes analysis on the region’s changing climate and how it affects the aforementioned three primary concerns for vinifera cultivation in Michigan. Future GST, monthly precipitation distributions and season length are all considered under the RCP4.5 and RCP8.5 scenarios in the future out to the year 2099 for southwest and northwest Michigan and the changes are discussed. Finally, there is a discussion of Michigan’s potential future varieties using the future climate projections and known climate thresholds for a number of vinifera varieties. This section is meant as a hypothetical scenario for future decades where vinifera acreage in Michigan continues to expand and new varieties are considered for the region.