BCIs provide simplistic illustrations of viticulture and/or variety suitability (Kenny and Harrison, 1992; Tonietto and Carbonneau, 2004; Hall and Jones, 2010; Anderson et al., 2012). They commonly place numerical or descriptive envelopes around summed or averaged daily or monthly growing-season climate variables to express suitability ranges. They do not necessarily resolve the full range of climatic processes, intra-annual variability, seasonality of weather, or critical daily or hourly timescale events, which can also influence climate-viticulture relationships (Jones et al., 2009). Furthermore, BCI suitability classification envelopes have been derived from observations of varieties in established regions, which may not adequately reflect full variety suitability ‘potential' or the adaptive capacity of viticulture through vineyard management techniques (Jones and Storchmann, 2001; Webb et al., 2008; Tomasi et al., 2011). BCIs are, in essence, crude measures of suitability that may mask or overstate true viticulture potential in a specific location. However, notwithstanding their limitations, they have been extensively applied in different regions, over different timescales, at different spatial resolutions and driven by both observed and modelled climate data as a means of depicting variance and broad suitability for viticulture and winegrape varieties. In this work, BCIs are employed to provide comparative spatial and temporal models to illustrate variability, trends and analogues that could affect viticulture suitability in the UK. BCI selection is restricted to those that are more commonly used to represent indicative suitability (Supplementary Information, Table 2): GST (Jones, 2006) and GDD (from which the Winkler Index is derived; Amerine and Winkler, 1944). These are complemented by GSP (Blanco-Ward et al., 2007; Santos et al., 2012; here extended to include October) and mid-March–May air frost frequency, as high values of these two meteorological variables have been previously shown to have a negative impact on UK wine yields (Nesbitt et al., 2016).

The GST ‘Cool' climate and the GDD Region Ia (Supplementary Information, Table 2) classifications are representative of BCI values associated with most UK vineyards over the last 30–40-years. In the development of the Winkler Index, used to classify GDD, Amerine and Winkler (1944) did not specify a lower-class limit for Region I or an upper-class limit for Region V. Subsequent research (Jones et al., 2010; Hall and Jones, 2010), using high-resolution spatial climate data, identified these limits for California, Oregon, Washington, Idaho, and Australia. The results provided a lower bound to Region I of 850 GDD and an upper bound to Region V of 2700 GDD. Additional research in other wine regions resulted in Region I being sub-divided into a Region Ia (very early ripening varieties, mostly hybrid grapes) and Region Ib (early ripening varieties, mostly Vitis vinifera L.) (Anderson et al., 2012; Jones et al., 2012). The GST ‘Cool' climate classification, meanwhile, is broadly representative of viticulture regions with the GST climate-maturity requirements for varieties such as Müller-Thurgau, Pinot Gris and Gewürztraminer and some regions where Riesling, Chardonnay, Pinot noir and Sauvignon Blanc are grown (Jones, 2006).

Few vineyards in the UK have long-term historic site-specific weather data available for analysis. Therefore, Met Office regional monthly average temperature and precipitation data (1981–2020) were used to calculate GST and GSP values for the south-east and south-central England region—where most vineyards are currently located (Figure 2). Anomalies were calculated relative to a 1981–2010 climate normal period (baseline) to quantify and illustrate climate trends and to highlight growing season characteristics in low (2012) and high (2018) yielding years (Figure 3). The 1981–2010 climate normal period has been widely used in climate change research and in previous climate and wine work (Bois et al., 2018; Nesbitt et al., 2018; Dunn et al., 2019; Hewer and Brunette, 2020). Met Office regional monthly average temperature data are derived from station daily means and rainfall data from summed daily station totals, interpolated onto a 5 × 5 km grid before being averaged for a region, enabling a macroclimatic regional analysis of historic growing season conditions (Met Office, 2020a).

We also use HadUK 1 × 1 km gridded data (interpolated from meteorological station data onto a uniform grid; Met Office et al., 2019) to calculate UK BCIs for other specific reference periods (Figures 4, 5, 7 and 8). In addition, we employ the ERA5-Land re-analysis (ECMWF, 2020; Hersbach et al., 2020) as our source of high-resolution gridded European air temperature, with a spatial resolution of ~9 km, to support the calculation of historic GST and GDD values for European viticultural regions (Supplementary Information, Tables 3 and 4, and Figure 1). For equivalent GSP values in Europe, we use historic nearest station measurement data (Infoclimat, 2021).

UKCP18 (Met Office, 2020b; Met Office et al., 2021) provides a set of spatially coherent future climate projections for the globe based on 15 Met Office Hadley Centre model variants: HadGEM-GC3.05 (Perturbed Parameter Ensemble 15 (PPE-15)) and 13 other global climate models/simulations from the CMIP5 multi-model ensemble that was used to inform the IPCC 5th^{Assessment (CMIP5-13). Regional (Europe) and local (UK only) projections (12 km and 2.2 km spatial resolution, respectively) were downscaled from the global projections (PPE-15) using the Met Office Hadley Centre model: HadREM3-GA705 (to regional level) and then HadREM-GARA11M (to local level), leading to 12 high-resolution simulations in total (Gohar et al., 2018). Only 12 of the PPE-15 were made available through UKCP18 due to computational restrictions and following assessment of model performance against observations in the historical period (Murphy et al., 2018). For the local projections, the original climate model gridded data at 2.2 km spatial resolution had been regridded to 5 km resolution for the British National Grid (Ordnance Survey National Grid) prior to use in this work (Met Office, 2019).}

RCP8.5 most closely reflects society’s limited response thus far to climate change mitigation, and, as a result, this scenario has been the context for the most detailed climate modelling for the UK undertaken within UKCP18 (at 2.2 km resolution). Hence, to present changes over the near term 2021–2040 period—when climate projections are not very sensitive to emission pathways and emissions are not the largest source of uncertainty—we focus on analysing these 2.2 km projections (using daily values of maximum and minimum temperature and precipitation, regridded to 5 km) for 2021–2040 driven by RCP8.5, and the difference between that period and 1999–2018 (HadUK). UKCP18 does provide a longer-term (2061–2080) time slice, but this is not assessed here because large uncertainties exist around the world’s RCP track and response to climate change over this longer timeframe, i.e., there is an opportunity now to change course from this ‘high-emissions’ or ‘business as usual’ scenario. The 2021–2040 time horizon is highly relevant to current viticultural decision-making as investment and grower focus is currently on the next 20–30-years. Hence, we assess uncertainties in climate projections by focusing on climate model uncertainty (using a multi-model ensemble) and upon inter-annual climate variability rather than through alternative emission scenarios.

Regional grape yield data for the south-east and south-central England, where most vineyards are located (Figure 2), are not available. Nationally aggregated yield data—the only ‘official’ wine yield data that is available in the UK—were therefore used to examine the relationship between south-east and south-central England GST and GSP and wine yields in 2012 and 2018. 1989 marks the start of ‘official’ records of UK wine yields. UK yield data (hL/ha) was obtained from the Wine Standards Branch of the Food Standards Agency via Wines of Great Britain (WineGB, 2020a). The use of national, non-regionally specific yield data could lead to some distortion of climate‒yield relationships; however, national yield values were deemed indicative of those in our focus region due to its significant contribution to the total UK vineyard area (74 %; WineGB, 2020b). An analysis of national weather/climate data, meanwhile, would disproportionally include areas where few or no vineyards exist.

Table

Intra- and inter-annual weather variability can drive grape production and quality parameters beyond ‘acceptable' vintage variation bounds. Kenny and Harrison (1992) evaluated the frequency of viticulturally suitable or unsuitable years (1951–1980) in Europe and based their work on the premise that the frequency of ‘good' or ‘bad' years is more important than average conditions over a 30-year period. As an additional dimension to our analysis, we hypothesise that, particularly in the UK's relatively marginal climate (Kenny and Harrison, 1992; Nesbitt et al., 2016), vulnerability to weather variability is a limiting factor to viticulture viability, at annual or longer timescales. Through this work, investment and variety adaptation decisions are also, therefore, meaningfully guided by the frequency of bioclimatic values akin to those of 2018, over a 20-year period (2021–2040), as well as by projected averages.

ArcGIS (ESRI, 2020) was used for climate data and vineyard location integration and as a rapid means of spatial analysis at high resolution. No ‘official’ database of UK vineyard locations nor size (ha) was publicly available. The UK Vineyards List (Skelton, 2020), although not independently verified, was deemed the most reliable and up-to-date (November 2020) source. Postcodes from this list, however, were often found to relate to premises and not precise vineyard locations, so, to maximise model accuracy, individual vineyards in south-east England (≥ 1 ha) were visually located utilising Google Earth (Google, 2020). Once located, the presence of Pinot noir grape production in the vineyards was confirmed with producers (through online material or personal correspondence). Coordinates (British National Grid) of approximate vineyard centres were imported as point features into ArcGIS to enable an analysis of their growing season climates when overlain with HadUK and UKCP18 gridded BCI data layers. The BCI data were imported into ArcGIS as Network Common Data Form (NetCDF) files.

European vineyard areas were mapped using the CORINE landcover data (Copernicus, 2020; which omits UK vineyards) and integrated into ArcGIS. Specific vineyard areas in the Montagne de Reims, Côte des Bar, Côte de Nuits and Baden, where Pinot noir grapes of international renown are grown, were identified using guides and maps from Lewin (2011) and Johnson and Robinson (2001). An illustration of the mapped European vineyard areas, gridded ERA5-Land climate data overlay, and the employed data extraction process is shown in Supplementary Information, Figure 1. Data were not extracted from grid cells in which only very small areas of viticulture existed.

Annual GST and GSP for the south-east and south-central England region (the dominant viticulture areas; Figure 2) for 1989–2020 are presented in Figure 3, expressed as anomalies relative to a 1981–2010 average baseline of 13.6 °C and 421 mm, respectively. There was an even split (16 of 32-years) of positive and negative GSP anomalies, although twice as many growing seasons (8) were > 20 % drier than wetter (4). Of the 32 growing seasons, 21 were warmer than the GST baseline, and 16 of these occurred since 2000. Of these 21, 16 were > 14.0 °C GST, which is 1.0 °C above the 13.0 °C climate-maturity baseline for ‘cool climate’ wine production, as provided in the groupings set out in Supplementary Information, Table 2 (Jones, 2006). Indeed, the peak in 2006 of 15.1 °C just reached the ‘intermediate’ climate maturity classification (15.0–17.0 °C GST; Jones, 2006), which constitutes the climate-maturity threshold for a wider selection of commercially popular grape varieties: Pinot Gris, Gewürztraminer, Riesling, Pinot noir, Chardonnay, Sauvignon Blanc, Semillon and Cabernet Franc.

Correlations in this same geographical region between growing season weather variables, acute/extreme events and UK wine yields have previously been explored, and their relationships with high inter-annual yield variability documented (Nesbitt et al., 2016). The lowest UK wine yield on record (6 hL/ha) was recorded in 2012 (Figure 1) and was attributed to wet and cold weather during flowering and fruit set. Conversely, 2018 was an outlier which saw the UK's highest wine yield of 48 hL/ha (Figure 1; WineGB, 2020a) following an exceptionally dry and warm season (Figure 3).

Table

The suitability of UKCP18 projections for application in this study was evaluated through a bias assessment between modelled (UKCP18) and observed (HadUK) gridded historic (20-yr mean) data for the UK (Figure 4—right-hand panels). The bias assessment was undertaken using ESRI ArcGIS (resampling HadUK to a 5 × 5 km grid and employing the Spatial Analyst Map Algebra function to calculate biases; ESRI, 2020). UKCP18 1981–2000 GST minus HadUK GST for the same period yielded a spatial mean UKCP18 bias of -0.3 °C (i.e., a cold bias; range: -3.2 to 3.5 °C). The UKCP18 GDD mean bias was -22 GDD (range: -332 to 898 GDD). In the dominant viticulture areas (Figure 2) of south-east and south-central England, the GST bias range was circa -0.5 to 0.3 °C and the GDD bias range circa -100 to 50 GDD (Figures 4a and 4b). The UK-wide comparison between UKCP18 and HadUK 1981–2000 GSP revealed a -9.7 mm GSP mean bias (i.e., a dry bias; range: -774 to 474 mm). The GSP mapped bias generally ranges from 50 mm to -100 mm (10 to -15 %) in the dominant viticulture areas (Figure 4d). Mid-March to May Air Frost days were higher in the UKCP18 model for the 1981–2000 period, with a mean UK bias of 0.9 days (i.e., a cold bias). The bias range in south-east and south-central England was circa ± 3 days. Subsequently, UKCP18 values used in this study have been subjected to spatially dependent bias-adjustments using ESRI ArcGIS Data Management and Spatial Analyst tools.

Figure 4a highlights a recent warming in the observational record of ~1 °C GST in much of south-east and eastern England (Cambridgeshire and East Anglia) between the 1981–2000 and 1999–2018 periods, a shift which has been one of the key enablers for growth and variety change in the UK viticulture sector over this time interval, significantly expanding the area within the ‘cool climate’ GST and Region Ia GDD (Figure 4b) classifications (Supplementary Information, Table 2). Over the same time, Air Frost days during the sensitive mid-March to end of May season have become less common but have remained a hazard everywhere (Figure 4c). Meanwhile, trends in GSP are less homogeneous and impactful in the main UK viticulture regions: south-east England has, in general, become marginally drier by 1–50 mm during the growing season and areas of south-central England and East Anglia marginally wetter by a similar amount (Figure 4d).

Table

The evidence of recent warming during the growing season in south-east and south-central England and recent rapid growth of the sector illustrates the impact of climate change on viticulture viability. Future UK climate change may further impact viability, variety suitability, and wine style potential. Figure 5 shows the projected 20-year multi-model mean BCI changes in the UK under RCP8.5 from 1999–2018 (HadUK) to 2021–2040 (UKCP18; bias-adjusted).

Table

The UKCP18 multi-model 12-member ensemble means for 2021–2040, presented in Figure 5, do not reveal the ensemble range and, therefore, the associated model uncertainty, nor the inter-annual variability, which is so critical to viticulture (Nesbitt et al., 2016). To illustrate these important outputs, Figure 6 presents bias-adjusted GST from the 12 simulations (see Methods—Climate Change Projections section), as well as the ensemble means for each year during 1981–2000 and 2021–2040 for a single 5 × 5 km grid cell in southern England, in which a cluster of established vineyards are located.

Table

2018 was a year that produced exceptional yields and grape quality across most of the UK’s grape-growing regions (WineGB 2020a). BCI values derived from HadUK grid cells (1 × 1 km) overlain onto 97 vineyards (> 1 ha) producing Pinot noir (predominantly for sparkling wine) in south-east England show that GST during 2018 ranged from 14.1–15.5 °C, GDD from 974–1223 and GSP between 273–449 mm. Based on the bias-adjusted UKCP18 12-member ensemble for individual years during 2021–2040 (n = 12 × 20 = 240), the number of years when these BCI ranges are projected to occur was calculated (Figure 7).

Table

[Note that London and Greater London are often outside the upper end of this temperature range].

Mean BCI values for 1999–2018 in the Champagne, Burgundy and Baden regions are presented in Supplementary Information, Table 4. The GSP values for 1999–2018 are sourced from a set of weather stations for which a complete record was available for the period and which are within or near the regions of interest. They provide a useful indication of the 20-year mean GSP within the regions, but we acknowledge they are not fully representative of the GSP spatial variability across these areas. Lack of GSP is not a viticulture limiting climatic variable in the UK and nor is it projected to be so during the next 20-years, even allowing for likely enhanced evapotranspiration as a function of positive air temperature trends. The main viticulture areas in the UK and areas that are projected to fall within temperature-based BCI ranges of Baden and Burgundy during 2021–2040 (Figure 8) are projected (under RCP8.5) to have a GSP of 300–450 mm during 2021–2040 (Figure 5c), with a mean which is close to the 1999–2018 Champagne average but lower than that for Burgundy and Baden (where warm-season convective downpours are more common).

Therefore, in this work, only the GST and GDD temperature-based BCI 20-year mean values in Supplementary Information, Table 4, were used as analogue ranges (from across all grid cells) to define areas in the UK which satisfy their requirements under both current (1999–2018) and bias-adjusted UKCP18 multi-model ensemble 2021–2040 conditions (Figure 8).

Table

Table 1 summarises ‘expert’ scores for high-quality Champagne (2008 and 2012) and Burgundy (2005 and 2015) vintages over the period 1996–2015. Another year, 1996, was also rated as a high-quality Champagne vintage, but ‘ratings’ were not available from all ‘experts’. Pinot noir wines from Baden were not included in this vintage assessment or analogy as there were fewer published and reliable vintage results available for the region from the same ‘expert’ sources.

Table

titre du tableau
	Vintage	Decanter (out of 5)	Jancis Robinson(non-numerical score)	Berry Bros & Rudd(1 = poor; 10 = outstanding)	Wine Spectator (95–100 = classic; 90–94 = outstanding)	Robert Parker (90–95 = outstanding; 96–100 = extraordinary)
Champagne	2008	5	‘Initially, a difficult, damp year with widespread mildew; drier conditions in August and a fine, warm September proved many producers’ saving grace. Classically styled wines with fresh acidity balanced by sound ripeness achieved late in the season.’	10	97	99
	2012	5	‘An exceptional vintage in Champagne. Despite low yields thanks to frost, hail and disease early in the season, August saw conditions improve dramatically, resulting in exemplary maturity, acidity and grape health at harvest.’	9	95	96
Burgundy	2005	5	‘As in Bordeaux, a quite exceptionally good vintage, although many wines may go through a prolonged stage of chewy adolescence.’	10	98	98
	2015	5	‘Low yields and warm weather allowed for ample ripeness, small berries and an early harvest. Quality is looking extremely fine, with some people whispering comparisons with the outstanding 2005 vintage. Acid levels in individual wines may be crucial.’	9	98	98

The BCI values in which such exceptional vintages were realised are presented in Supplementary Information, Table 3.

The two regional sets of vintage BCI ranges provide analogues to define spatial BCI patterns in the UK when applied as the projected years in 20 repetitions during 2021–2040 of top-quality vintage Champagne and Burgundy years as shown in Figure 9.

Table