Conceived and designed the experiments: EL MZ. Performed the experiments: EL. Analyzed the data: EL. Contributed reagents/materials/analysis tools: EL EHG. Wrote the paper: EL.
The authors have declared that no competing interests exist.
Winter chill is one of the defining characteristics of a location's suitability for the production of many tree crops. We mapped and investigated observed historic and projected future changes in winter chill in California, quantified with two different chilling models (Chilling Hours, Dynamic Model).
Based on hourly and daily temperature records, winter chill was modeled for two past temperature scenarios (1950 and 2000), and 18 future scenarios (average conditions during 2041–2060 and 2080–2099 under each of the B1, A1B and A2 IPCC greenhouse gas emissions scenarios, for the CSIRO-MK3, HadCM3 and MIROC climate models). For each scenario, 100 replications of the yearly temperature record were produced, using a stochastic weather generator. We then introduced and mapped a novel climatic statistic, “safe winter chill”, the 10% quantile of the resulting chilling distributions. This metric can be interpreted as the amount of chilling that growers can safely expect under each scenario. Winter chill declined substantially for all emissions scenarios, with the area of safe winter chill for many tree species or cultivars decreasing 50–75% by mid-21st century, and 90–100% by late century.
Both chilling models consistently projected climatic conditions by the middle to end of the 21st century that will no longer support some of the main tree crops currently grown in California, with the Chilling Hours Model projecting greater changes than the Dynamic Model. The tree crop industry in California will likely need to develop agricultural adaptation measures (e.g. low-chill varieties and dormancy-breaking chemicals) to cope with these projected changes. For some crops, production might no longer be possible.
Cool temperatures in the winter are essential for successful cultivation of many tree crops
Insufficient winter chill can severely reduce crop yields and crop quality. When chilling requirements are not completely fulfilled, trees display irregular and temporally spread out flowering, leading to inhomogeneous crop development. This process ultimately results in varying crop sizes and maturity stages at the time of harvest, which can substantially reduce yield amount and value
Agricultural scientists have developed mathematical models that are used by growers to select tree cultivars with chilling requirements that correspond to available chilling at a specific location. However, a grower's understanding of available winter chill is likely to reflect conditions of the past rather than those expected for a warmer future. Since orchards often remain in production for decades, consideration of future expected winter chill is necessary in times of imminent climatic changes. Without such considerations, many orchards might receive inadequate chilling by the time they reach physiological maturity, even though at the date of planting, climatic conditions were optimal for the chosen cultivars. Depending on the pace of winter chill decline, the consequences for California's fruit and nut industries could be devastating.
While a few studies have investigated the impact of climate change on winter chill
For generating the hourly temperature records needed for quantifying winter chill from daily records, which are more readily available, we correlated short-term hourly with long-term daily temperature records by Partial Least Squares regression (for accuracy estimates refer to
When using the Chilling Hours Model to calculate winter chill, safe winter chill was about 130 Chilling Hours lower than mean winter chill on average over all locations and scenarios analyzed (shown for Davis in
In box plots, the central line indicates the median of the distribution, the edges of the boxes are the 25% and 75% quantiles, error bars are the 10% and 90% quantiles, and dots indicate outliers.
In all scenarios, winter chill in California declined substantially over time. The MIROC GCM projected the greatest warming and thus the greatest decreases in winter chill, followed by the HadCM3 and CSIRO GCMs. Since none of these models can be clearly identified as being more accurate than the others, we only show winter chill averaged over all three models. Supporting
Since most of the state's fruit and nut production is located in the Central Valley (
Future winter chill was quantified using the A2 IPCC greenhouse gas emissions scenario.
Changes projected by the Dynamic Model were less severe than for the Chilling Hours Model, but nevertheless likely to strongly impact fruit and nut production. By the end of the 21st century, this model projected decreases in winter chill between 30 and 60% of 1950 conditions (
Future winter chill was quantified using the A2 IPCC greenhouse gas emissions scenario.
Decreases in winter chill between 1950 and 2080–2099 varied geographically between the Northern Sacramento Valley (NSacV;
Year | 1950 | 2000 | Mid 21st century (2041–2060) | End 21st century (2080–2091) | ||||
Emission scenario | B1 | A1B | A2 | B1 | A1B | A2 | ||
Northern Sacramento Valley | 993±43 | 870±70 | 697±91 | 647±100 | 654±100 | 577±112 | 498±125 | 439±137 |
Southern Sacramento Valley | 1015±71 | 784±55 | 634±71 | 572±73 | 578±73 | 494±75 | 400±77 | 334±77 |
Northern San Joaquin Valley | 1101±79 | 876±90 | 757±93 | 702±93 | 704±94 | 632±96 | 537±97 | 476±96 |
Southern San Joaquin Valley | 979±70 | 844±58 | 697±69 | 647±70 | 649±71 | 587±71 | 489±74 | 423±74 |
Northern Sacramento Valley | 73.0±2.0 | 70.8±2.0 | 62.8±3.3 | 60.2±3.7 | 61.1±3.6 | 56.7±4.1 | 51.3±5.0 | 48.7±5.5 |
Southern Sacramento Valley | 73.8±2.9 | 69.7±2.9 | 61.1±4.0 | 58.0±4.3 | 59.2±4.3 | 54.1±4.5 | 47.5±4.9 | 44.3±5.2 |
Northern San Joaquin Valley | 74.8±3.3 | 71.4±2.4 | 63.6±3.4 | 60.5±3.6 | 62.0±3.6 | 57.4±3.7 | 51.4±4.0 | 48.7±4.1 |
Southern San Joaquin Valley | 67.0±3.5 | 64.3±2.9 | 54.5±3.6 | 50.6±3.8 | 52.2±3.8 | 47.6±3.9 | 41.6±4.0 | 37.9±4.2 |
Future projections were done for the B1 (low), A1B (moderate) and A2 (high) IPCC emission scenarios.
Winter chill decline strongly affected the spatial extent of areas suitable for the cultivation of tree crops with chilling requirements. For cultivars requiring 200 Chilling Hours, such as low-chill almonds, winter chill conditions are unlikely to become critical by the end of the 21st century (
Observed historic and future projected temperature increases in California strongly decreased the availability of winter chill under all greenhouse gas emissions scenarios, regardless of the model used to quantify this important climatic parameter for fruit production. On a global scale, it is likely that most other growing regions of subtropical fruit and nut trees with chilling requirements will be similarly affected by declining winter chill. Our projections showed that for many tree crops that now cover large areas within the Central Valley, climatic conditions will become less suitable and in many cases potentially prohibitive for production. Areas where safe winter chill exists for growing walnuts, pistachios, peaches, apricots, plums and cherries (>700 Chilling Hours) are likely to almost completely disappear by the end of the 21st century. For cultivars with chilling requirements above 1000 Chilling Hours, such as apples, cherries and pears, very few locations with safe chilling levels were found to exist today, and our modeling results project that virtually none will exist by mid century.
The resulting reductions in crop yield and quality could severely impact California's tree crop growers. According to the USDA Agricultural Census of 2002, the state had 38,693 fruit and nut orchard farms, covering 1.2 million hectares of land and driving a US$ 8.7 billion industry
Given the long life spans of orchards compared to annual crops and the substantial investments required for orchard establishment, tree crop growers will be much more vulnerable to the long and medium term effects of climate change than growers of annual crops, making the development of predictive temperature models for tree crop yields crucial for strategic planning of orchard operations.
Improved orchard management might have potential for alleviating winter chill decline, since planting density, pruning practices and irrigation regime can influence orchard microclimate
Research on chilling models in many subtropical regions has indicated that the Chilling Hours Model is not very precise in this climatic zone
While we are confident of the general trend of declining winter chill, some locations within the Central Valley might remain suitable even for crops with high chilling requirements. Locations with cooler microclimatic conditions might be found along major rivers, in the foothills of Sierra Nevada and Coastal Range, where cold air tends to drain, as well as close to the Sacramento Delta and in those parts of the Central Valley, where frequent fogs reduce temperatures during the winter. On the other hand, it is likely that warmer temperatures will reduce the incidence of fog in many places, leading locally to stronger deterioration in winter chill than projected in this study.
The high sensitivity of the commonly used Chilling Hours Model to climate change
While this study focused only on winter chill, climate change may have other (positive and negative) effects on tree crop production. Rising summer temperatures can be expected to be beneficial to some crops, while having a negative impact on others
Hourly temperature records are required for estimating winter chill with all common methods without resorting to idealized daily temperature curves. We obtained records of hourly temperatures for all 205 (active and inactive) stations of the California Irrigation Management Information System
Since the CIMIS network was only established in 1982, it is not very suitable for analyzing long-term climatic changes. We therefore obtained daily measurements of minimum and maximum temperatures and precipitation from all 113 weather stations in California that belong to the cooperative weather station network administered by the National Climatic Data Center
For using daily measurements to estimate hourly temperatures, each CIMIS weather station was associated with a nearby NCDC station. Using the Euclidean Allocation function of a Geographical Information System (GIS; ArcGIS 9.2, ESRI, Redlands, CA, USA), each CIMIS station was assigned the closest weather station of the other network, resulting in pairs of weather stations that were on average 20 km apart (max. distance was 76 km). The daily and hourly datasets of these station pairs were then joined. To remove records that were considered faulty, all hourly temperature records that were more than 5°C above the daily maximum or below the daily minimum of the NCDC record were eliminated from the dataset.
When analyzing observed or modeled weather records, long-term trends are often obscured by interannual variation. Temperatures observed during a particular year are often substantially warmer or cooler than the long-term running average for that year. This constraint can be overcome by generating synthetic weather records, which allow correction for interannual variation and facilitate statistical evaluation of weather records
We used the LARS-WG stochastic weather generator
Since all common winter chill models require hourly temperatures as inputs, such records had to be derived from the daily datasets. To establish a relationship between daily and hourly temperatures, we performed separate Partial Least Squares
In order to achieve the most accurate predictive equations, we used a cross-validation procedure (JMP 7, SAS Institute, Cary, NC, USA) to identify the most appropriate dimension for the regression models. This procedure splits the dataset into two or more groups and fits a regression model to all groups except one. The resulting model is then used to predict the values in the omitted group. This process is repeated for all groups and errors are quantified, providing an estimate of overall model accuracy. The number of latent factors is then chosen to maximize the overall accuracy in estimating hourly temperatures.
Two climate scenarios representing 1950 and 2000 conditions were based on temperatures observed during the historic record. To obtain representative conditions for these two years, we calculated separate linear regression analyses for each month of the year from the entire daily temperature record for each NCDC weather station that was used to estimate hourly temperatures. Regressions were calculated for minimum and maximum daily temperatures, as well as for daily precipitation. Based on the resulting equations, representative values for all three parameters were obtained for both 1950 and 2000, and converted into separate climate scenario input files for LARS-WG for each weather station pair.
Future winter chill conditions were estimated based on statistically downscaled climate projections for minimum and maximum daily temperatures (averaged monthly) from three General Circulation Models—UKMO-HadCM3, CSIRO-MK3.0, and MIROC3.2(medres)—each run under the SRES A2, A1B, and B1 greenhouse gas emissions scenarios from the Intergovernmental Panel on Climate Change AR4
We calculated winter chill according to two methods that are currently used in California. The most common chilling model used in the state is the Chilling Hours Model [sometimes referred to as Weinberger Model; 17,18]. In this model, chilling is quantified by simply adding up all hours, during which temperatures range between 0 and 7.2°C [refer to ref. 8 for equations describing both models]. As commonly practiced in California, we quantified accumulated winter chill between Nov 1st and Mar 1st of each winter season.
In recent years, growers of cherries in California have adopted the Dynamic Model, developed in Israel
For each time period analyzed (1950, 2000, 2041–2060, and 2080–2099), we calculated winter chill for 100 replications of each year, which allowed statistical evaluation of winter chill estimates. That is, rather than simply producing one value representing the winter chill accumulated during a given year, we used the variability produced by the stochastic weather generator to evaluate the distribution of winter chill over 100 replications of that year. This provided the ability to estimate the percentage of years, during which particular amounts of winter chill are likely to be available to fruit and nut growers.
While trends in winter chill are often analyzed using the mean of the chilling distribution, this measure is of subordinate importance to growers, because they economically depend on obtaining good yields in most (e.g. 90%) years rather than in an average year. Inadequate winter chill as often as once in ten years can threaten the economic sustainability of a farming operation. Here we present a novel climate change metric called “safe winter chill”, which we define as the 10% quantile of the winter chill distribution. This metric specifies the maximum chilling requirement that will be fulfilled in 90% of all years at a given site. In addition to calculating the mean of the winter chill distribution, we also calculated this safe winter chill metric.
Using the procedure outlined above, we estimated safe winter chill for all twenty climate scenarios at all suitable CIMIS weather stations in California. This procedure provided point estimates of safe winter chill, which needed to be interpolated to cover all of the state. We used the Kriging interpolation technique [with a spherical semivariogram, variable search radius and based on the 12 nearest data points; ref. 39] to create winter chill surfaces at a 20 arc-minute spatial resolution. While the resulting surface should fairly accurately describe safe winter chill in the relatively flat Central Valley, its validity in more mountainous terrain is limited, because elevation has a strong effect on winter chill and many locations are at substantially lower or higher elevations than the closest CIMIS station. To adjust for this effect, we estimated the elevation error of the interpolated surface by generating a Kriging surface through all point elevations of the weather stations. This surface was then subtracted from a Digital Elevation Model of California
We then estimated the rate at which safe winter chill increases with increasing elevation separately for each climate scenario by calculating simple linear regressions between estimated safe winter chill and elevation across all weather stations. While the resulting regression equations were relatively poorly defined (coefficients of determination of <0.1 in many cases), all regressions were statistically significant at p<0.05, and their slopes should represent a reasonable estimate of the additional effect of elevation on safe winter chill. Multiplying the resulting rates with the elevation error surface and adding the results onto the original interpolated chilling grids produced error-adjusted safe winter chill surfaces for the entire state for each climate scenario.
For easier interpretation of the results, we finally created average surfaces for the chilling estimates resulting from the different General Circulation Models. This process resulted in eight winter chill surfaces, representing climatic conditions in 1950, 2000 and in 2041–2060 and 2080–2099 under the B1, A1B and A2 greenhouse gas emissions scenarios, respectively. To facilitate data processing, we implemented most analysis steps in JSL, the scripting language of JMP 7 and in the ArcGIS ModelBuilder.
Accuracy estimate of winter chill projections. This text file describes the methods used to assess the accuracy of interpolated winter chill surfaces. Results are displayed in
(0.05 MB PDF)
Error estimates of projected winter chill. Qualitative error estimates of winter chill projections caused by elevation differences between the interpolated location and the closest CIMIS station (a) and by distance to the closest station (b).
(1.54 MB TIF)
Safe winter chill throughout California (in Chilling Hours). Safe winter chill (10% quantile of distribution over 100 modeled repetitions for each year) in California, quantified with the Chilling Hours Model for eight climate scenarios, representing climate conditions observed around 1950 and 2000, and projected for 2041–2060 and 2080–2099 under the B1, A1B and A2 IPCC greenhouse gas emissions scenarios.
(4.70 MB TIF)
Safe winter chill throughout California (in Chill Portions). Safe winter chill (10% quantile of distribution over 100 modeled repetitions for each year) in California, quantified with the Dynamic Model for eight climate scenarios, representing climate conditions observed around 1950 and 2000, and projected for 2041–2060 and 2080–2099 under the B1, A1B and A2 IPCC greenhouse gas emissions scenarios.
(4.63 MB TIF)
We acknowledge Ron Neilson and the MAPPS group from United States Forest Service/Oregon State University for developing and producing the downscaled GCMs used in this paper. We also acknowledge the GCM modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP's Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multi-model dataset.