Ensembles for Viticulture Climate Classifications of the Willamette Valley Wine Region

Abstract

Future climate projections provide an opportunity to evaluate cultivar climate classification and preferred styles of wine production for a wine grape growing region. However, ensemble selection must account for downscaled archive model skills and interdependence rather than be arbitrary and subjective. Relatedly, methods for generalizing climate model choice remain uncertain, particularly for identifying optimal ensemble subsets. In this study we consider the complete archive of the thirty-two Coupled Model Intercomparison Project Phase 5 (CMIP5) daily Localized Constructed Analogs (LOCA) downscaled historic datasets and their observational data that were used for downscaling and bias corrections. We apply four model averaging methods to determine optimal ensembles for the computation of six common climate classification indices for the Willamette Valley (WV) American Viticultural Area (AVA). Among the four methods evaluated, elastic-net regularization consistently performed best with identifying optimal ensemble subsets. Variation exists among the optimal ensembles computed for each of the six bioclimatic indices. However, a subset of approximately seven to ten climate models were consistently excluded across all six indices’ ensembles. While specific to the archive and wine region, optimal ensemble sizes were noticeably larger than ensemble sizes commonly employed in published studies. Results are reported such that they can be used by researchers to independently perform analyses involving any one of the six bioclimatic indices throughout the WV AVA while using historic and future LOCA CMIP5 climate projections. The data and methods employed herein are applicable for other wine regions.

Temperature-based Climate Projections of Pinot noir Suitability in the Willamette Valley American Viticultural Area

“…It is now understood that assuming archive model independence results in a multi-model mean that is biased by the duplicative information that is contributed by the similar models (Sanderson et al, 2015;Herger et al, 2018), which can confound assessments of model agreement about changes in future climate and alter the statistics of identified correlations (Sanderson et al, 2015). One can address the knowledge uncertainty associated with model choice by applying methods that account for model skill and interdependence, wherein skill is measured by comparison with observed data (Massoud et al, 2019;Massoud et al, 2020;Sanderson et al, 2017;Skahill et al, 2021). Ensemble weight assignment depends on the modelling objective (i.e., the prediction) and study location (Sanderson et al, 2017).…”

“…to-measurement misfit using the elastic net penalty (Zou and Hastie, 2005) configured in the same manner as it was applied in Skahill et al (2021).…”

“…Application of future climate projections requires location and cultivar specific evaluations of the GSR model (Parker et al, 2020b). Skahill et al (2021) computed viticulture climate classification directed optimal ensemble compositions and weights for the Willamette Valley (WV) American Viticultural Area (AVA) for six bioclimatic indices, one being the GST. Their study was based on a comprehensive evaluation of the complete archive of the thirty-two CMIP5 daily Localized Constructed Analogs (LOCA) downscaled historic datasets and their observational data that were used for downscaling and bias corrections (Pierce et al, 2015a;Pierce et al, 2015b;Livneh et al, 2013).…”

“…It was also motivated by recent and consistent observed warm to hot vintages and the lack of previous studies of climate change impacts on viticulture in the WV AVA (Wine Spectator, 2020;Schultze and Sabbatini, 2019). The LOCA CMIP5 ensemble compositions and weights computed by Skahill et al (2021) varied by viticulture climate classification index. Of the thirty-two models which constitute the entire LOCA CMIP5 archive, ten of its members composed the unique ensemble for optimal computation of the GST index throughout the WV AVA.…”

“…To our knowledge, this is the first study to produce future projections of the GST index and Pinot noir specific GSR model applications for the WV AVA. Our focus herein differs from past studies in that our calculations are based on a consideration of the complete archive of the thirty-two LOCA CMIP5 downscaled datasets rather than an arbitrary and subjective selection or a small sampling of climate projections, or an ensemble weighting strategy that does not account for model archive skill and interdependence (Skahill et al, 2021). Spatiotemporal computations of the GST climate index and Pinot noir specific applications of the GSR model enable the opportunity to explore relationships between their computed values, with one intent being to provide updated GST ranges that better align with current temperature-based modelling understanding of Pinot noir grapevine phenology (Parker et al, 2020a;Parker et al, 2020b;van Leeuwen et al, 2013) and the viticultural application of LOCA CMIP5 climate projections for the WV AVA (Skahill et al, 2021).…”