Another minor issue requiring the bootstrap approach is that the implicit potential yield estimation should change the degrees of freedom in the non-linear regressions when estimating the standard errors. In the lower panel of Figure 3.2, a histogram of positive shares is presented. That is, for each chill portion, the count of panel observations where the share of that chill portion was positive. The actual shares of the very low and very high portions are usually quite low. This shows the relatively small number of observations with low chill counts. The two yield effects curves look very similar in the relevant chill range. By both estimates, the yield loss is very close to 0 at higher chill portions, and starts declining substantially somewhere in the upper 60’s, as the experimental literature would suggest. Interestingly, the polynomial curve does not exceed zero effect, although it is not mechanically bounded from above like the logistic curve. This probably reflects the fact that historically, the average growing conditions has not deviated much from the optimal range. The “within” transformation hence did not deviate the potential yield much from the optimum in this case. At currently low chill portion ranges of 55-60, the effect is around −25%, again consistent with the stipulation of Pope et al. that a significant effect threshold would be located there. Considering alternate bearing and other factors contributing to the background fluctuation in yields, it is easy to understand how such effects on relatively small areas within the pistachio growing counties have not been picked up by researchers so far. Anecdotal yield losses due to low chill have happened on relatively small scale and passed undetected in the county-level statistics, especially when only one or two chill measures per county were considered. In this case, while the resulting curves are very similar, I find the structural approach more convincing. First, it has a smaller confidence area, and therefore seems more precise. Second, a polynomial of low order will not approximate the process described by agronomists very well. However,cut flower bucket estimating higher order polynomials results in estimates that are not statistically significant. The implications of my estimates for pistachio yields are depicted in the lower half of Figure 3.1.

The bottom left panel shows the effects on the 1/4 warmest years in 2000– 2018. They are mostly between 10-20% yield decline. These rates are easy to miss due to substantial yield fluctuations in pistachios. What do these estimates mean for the future of California pistachios? Prediction of yield effects for the years 2020–2040 are depicted in the bottom right panel, again for the 1/4 warmest years in the 2020-2040. They show substantial yield drops, which could amount to costs in the hundreds of millions of dollars. Chapter 4 in this dissertation explores the potential gains from a technology that could help deal with low chill in pistachios: applying kaolin clay mixtures on the dormant trees to block sunlight. Thee expected net present value of this technology is estimated at the billions of dollar in economic gains. Considering my results, there may be significant gains from using these technologies even in warmer years today. Concluding this chapter, I want to stress the fact that even in the era of “big data” in agriculture, data availability is still a challenge when estimating yield responses to temperature in some crops, especially perennials and local varieties. Weather information required for assessing potential damages and new technologies might not always be available for a researcher. This chapter develops a methodology to recover this relationship, using local weather data and techniques for dealing with aggregated observations. I use this setup to empirically assess the yield effects of insufficient chill in pistachios, recovering this relationship from commercial yields for the first time in the literature. I then look at the threat of climate change to pistachio production in southern California. As winters get warmer, lowering chill portion levels are predicted to damage pistachio yields and disrupt a multi-billion dollar industry within the next 20 years. These results were made possible by using precise local weather data, applying relevant statistical methods, and using agronomic knowledge in the modeling process.

This approach for information recovery from a small yield panel, with limited useful variability at first sight, could be useful for other crops as well.Introduced to California more than 80 years ago, and grown commercially since the mid 1970’s, pistachio was the state’s 8th leading agricultural product in gross value in 2016, generating a total revenue of $1.82 billion dollars. According to the California Department of Food and Agriculture , California produces virtually all pistachio in theUS, and competes internationally with Iran and Turkey . In 2016, five California counties were responsible for a 97% of the state’s pistachio crop: Kern , Fresno , Tulare , Madera , and Kings . Since the year 2000, the total harvested acres in these counties have been increasing by roughly 10% yearly. Each increase represent a 6 – 7 year old investment decision, as trees need to mature before commercial harvest . The challenge for California pistachios has to do with their winter dormancy and the temperature signals required for spring bloom. I discuss the dormancy challenge and the Chill Portion metric in Chapter 3. It is worth noting that in fact, for the areas covered in this study, chill portions are strongly correlated with the 90th temperature percentile between November and February, the dormancy season for pistachios. The correlation is very strong, with a goodness of fit rating of about 0.91. In essence, insufficient chill is a right side temperature tail effect, comparable with similar effects in the climate change literature. Chapter 3 estimates the yield response of pistachios to CP. Substantial losses are predicted below 60 CP. Compared to other popular fruit and nut crops in the state, this is a high threshold , putting pistachio on the verge of not attaining its chill requirements in some California counties. In fact, there is evidence of low chill already hurting yields .Chill in most of California has been declining in the past decades, and is predicted to decline further in the future. Luedeling, Zhang, and Girvetz estimate the potential chill drop for the southern part of San Joaquin valley, where virtually all of California pistachio is currently grown.

For the measure of first decile, i.e. the amount of CP attained in 90% of years, they predict a drop from an estimate of 64.3 chill portions in the year 2000 to estimates ranging between 50.6 and 54.5  in the years 2045-2060. Agronomists and stakeholders in California pistachios recognize this as a threat to this valuable crop . Together with increasing air temperatures, a drastic drop in winter fog incidence in the Central Valley has also been observed. This increases tree bud exposure to direct solar radiation, raising their temperature even further . The estimates cited above virtually cover the entire pistachio growing region, and the first decile metric is less useful for a thorough analysis of pistachios. I therefore need to create and use a more detailed dataset, in fact the same one described in Cahpter 3. Figure 3.1 shows the geographic distribution of chill and potential damage in the 1/4 warmest years of observed climate and predicted climate . While not very substantial in the past,flower display buckets these losses are predicted to reach up to 50% in some regions in the future.Figure 4.1 sketches the short run market model. The linear supply curves take weather as given. On an ideal weather season, the supply curve is S0. On a year with warm winter, the supply curve is multiplied by a coefficient smaller than one, i.e. shifts left and rotates counter-clockwise, resulting in curve S1. Without MCE, the intersection of demand with S1 determines the market equilibrium. Once that is solved, the welfare outcomes-consumer surplus, grower sector profits, and total welfare-are calculated as the areas above or under the appropriate curves. When MCE technology is available, a modified supply curve starts with a section overlapping S1, and then “bends” right towards S0. If demand is high enough, market equilibrium is attained at this bend. Again, the welfare outcomes with MCE are calculated with the equilibrium price and quantity, together with the demand and SMCE curves. The gains from MCE are the differences between these market outcomes, i.e. the outcomes with MCE minus the outcomes without it. Note that the expansion of supply byMCE is guaranteed to result in positive gains from MCE in terms of total welfare and consumer surplus: the price is lower and quantity is higher. As for the grower sector, it does enjoy extra profits from being able to produce more, but the resulting lower price also decreases its profits from the output that would have been produced anyway without MCE. Therefore, one cannot tell a priori if grower profits increase or decrease when MCE is available. The sign and magnitude will need to be determined in the simulations, given the various parameters and functional forms. The climate prediction data produce a point estimate of chill portions for each year in 2020-2040. For a given set of model parameters and climate predictions for 2020-2040, the model is solved numerically twice for each year in this range. The consumer, grower, and welfare gains are calculated for each year using these two simulations. Using a discount rate of 5%, I can calculate the Net Present Value of the MCE gains in 2019. For each scenario, I run this procedure for 100 “independent draws” of 2020-2040 prediction paths. For each one, an entire simulation is run to produce an NPV of the gains.

I report the Expected NPV , the mean of this distribution, and standard errors around it. More details on the numerical solution of the model can be found in appendix A.3.Before I present the simulated welfare gains, there is one more piece in the puzzle. The calibrated model is set with 2016 acreage . Pistachio acreage through 2020- 2040 is likely to be different, and most likely higher than that. However, the model does not include endogenous growth of planted and harvested pistachio acres. To give some bounds on the expected gains, I run the simulations with four different acreage growth scenarios, each specifying a different pistachio acreage growth path until 2040. All scenarios assume some growth path until 2030, when acreage stabilizes and stays fixed through 2040. The first scenario is “No Growth”, meaning that 2020-2040 climate predictions are cast over the 2016 acreage. This should give a lower bound for gains, as acreage is predicted to grow and not shrink. The second scenario is “Low Growth”, which sets the yearly growth of harvested acres until the year 2022 at 9.6%, the average rate since 2000, and then sets zero growth . The growth until 2022 is attributed to currently planted but not yet bearing acres. This assumes that we are on the brink of a dynamic equilibrium in growth, and therefore no new acres will be planted in California. This scenario should give estimates that are higher than the “No Growth” scenario, but still rather conservative. The third scenario is “High Growth”. This one sets the growth rate until 2022 at 14.6%, the average rate since 2010, and then lets pistachio acreage follow the historic path of almonds in California . That is, the growth rate of almonds when they had the corresponding pistachio acreage. This very optimistic growth prediction makes the “High Growth” scenario the upper bound for the gains from MCE. One potential concern with acreage growth is that growers might switch new acreage to unaffected counties, or plant more heat tolerant varieties. For this, the “High North” scenario takes the high growth rate, but all new acreage harvested from 2023 is located in an imaginary “North” county, where chill damages are virtually zero. Note that planting in the unaffected north has the same effect on supply as planting a more heat tolerant variety near the existing locations . This last scenario is, in my opinion, the most plausible in terms of MCE gain magnitudes. A summary of the growth rates is depicted in Figure 4.2. In all scenarios, demand grows by the total rate of acreage growth.