This may be because breeding and management actions are generally assumed to outweigh any natural conditions that could result in alternate bearing at farm-, region- or nationwide scales. Similarly, agricultural research on yield and alternate bearing rarely integrates insights from masting literature. Such insights include the possibility that factors which increase yield in one year may result in a more severe reduction in the following year and the expectation that wind-pollinated taxa may be more prone to synchronous fluctuations in yield at larger spatial scales than insect-pollinated taxa. Here, we use global crop production data for plants known to be alternate bearing at an individual level to evaluate patterns of seed production at the national level. Specifically, we assess whether these crops are alternate bearing at national scales and whether patterns differ across pollination syndromes and are consistent with findings in masting systems. For this analysis, we use data from the Food and Agriculture Organization of the United Nations. The global, long-term nature of the FAO data offers a unique opportunity to study reproductive patterns in perennial plants, but also poses some limitations that could mask a signal of alternate bearing even if one exists. First, the FAO reports data at the national level, square pot and while the total area under production is included, it is not spatially explicit and thus cannot serve as a proxy for the extent of cultivation.
As such, we cannot directly test for synchrony among individuals or populations using these data. While a signal of alternate bearing in national-scale data would require synchrony at smaller scales, a lack of a signal does not preclude synchrony at the farm or regional level. Studies of synchrony in mast-seeding species suggest that we might expect signals of alternate bearing to be weaker in data at national compared to local scales. In that sense, our analysis is likely to underestimate the magnitude of alternate bearing at more local scales and thus their potential impact on farmer livelihoods. Furthermore, multiple crops are sometimes grouped together into a single FAO crop category , which could mask a species-specific signal. Finally, there is no information on the genetic variety or cultivation practices employed in each country that could influence the tendency toward alternate bearing within taxa. These features make any observed patterns in these data particularly salient. The presence of alternate bearing at national scales would highlight its ecological and socio-ecological importance. A national-scale analysis also allows for broad comparison with masting species where such synchrony has been observed. In this study, we characterized alternate bearing at national scales using three complementary metrics: lag-1 autocorrelation , coefficient of variation in seed production and bimodality of seed production . Classic alternate bearing would have all three metrics at a population level; in our study, we test whether they are detectable at a national scale. Throughout, when we discuss alternate bearing at a national scale, we refer to any or all of these variables.
We had strong a priori expectations that wind-pollinated species would show stronger alternate bearing at national scales, i.e. they would have stronger negative lag-1 autocorrelation, higher CV and We downloaded 58 years of data collected by the FAO on yield and production for 27 perennial crop taxa that have been reported in the horticultural literature as having a tendency toward alternate bearing at the individual level. For each crop, we selected up to 10 countries worldwide, each with at least 20 years of data, for inclusion in the analysis by their highest total production. The included nations were filtered for historical consistency and the crop names were modified for interpretability . The resulting dataset was comprised of 236 crop–country combinations . We performed the country selection and all subsequent analyses twice, once including only the past 25 years and once including all data on record . Patterns did not differ between the two time series; the results and discussion below describe the ‘long’ time series . To confirm that the FAO dataset reflected historical reproductive output of agricultural perennials, we scanned horticultural and economic online publications on focal crops that mentioned environmental shocks that occurred in exceptionally low yielding years at national scales and looked for signals of these in the detrended FAO time series . This validation exercise was done as a qualitative check of how well these aggregate time series represent on-the-ground experiences of growers; therefore, we scanned the available online information, rather than attempting a systematic review of trade publications.All statistical analyses were performed in R Studio Software 3.5.0. To focus on interannual variation in yield, rather than broader patterns of increasing or decreasing yield, we detrended all time series using generalized additive models fit to time series of yield through time.
These were calculated with defaults from the mcgv package and appeared to provide reasonable fits to the data based on visual inspection . To confirm that the results were not dependent on the detrending method, we repeated the analysis using locally estimated scatter plot smoothing and differencing as alternative detrending methods. The observed patterns persisted; we report the GAM-detrended results below . While national production was used for selecting the top-producing countries included in the analysis, yield was used for time series analysis because annual yield estimates should be less sensitive than production to additional variability from factors like farm expansion or contraction that are irrelevant to the present study. We characterized each time series with three common mast seeding metrics that provide complementary information about alternate bearing. The lag-1 autocorrelation is a measure of alternate bearing that captures the tendency of high years to follow low ones [38].We also inspected the autocorrelation function at lags up to 17 years as a descriptive tool to understand patterns of yield that differ from alternate bearing. Our second metric, the CV, is typically the standard deviation divided by the mean. However, because the mean of a detrended time series is 0, we used the standard deviation of the detrended time series divided by the mean of the raw time series, a technique commonly applied in climate modelling. Third, we used Hartigans’ dip statistic to test for bimodality. D is a measure of deviations from a unimodal distribution of values; significant differences would indicate that yield is multimodal . D was rarely significant , so we did not perform further tests to distinguish bimodality from other multimodal distributions. We analysed whether each masting metric differed as a function of pollination syndrome . We treat pollination syndrome as a binary variable as in previous masting studies, though relative dependence on insect pollinators may vary in more nuanced ways . Of the 27 crops, 6 were wind-pollinated and 21 were insect-pollinated . Because alternate bearing could be driven in part by phylogenetic relatedness, we conducted these tests using phylogenetically corrected linear mixed models with the trait as the predictor, the masting metric as the response, and random effects of plant phylogeny and country using the lme4 package . The plant phylogeny was constructed using the function ‘phylo.maker’ in the R package V.PhyloMaker using the GBOTB.extended mega-phylogeny as backbone and the default option ‘scenario 3’, in which the tip for a new genus is bound to the 1/2 point between the family root node and basal node. The resulting phylogeny had 35 tips because six of the crop categories contained multiple species. The phylogenetically corrected LMMs were run using the ‘phylo_lmm’ function on all possible combinations of FAO crop species of these six categories . Because the results were consistent across all iterations and varied minimally , the mean Chi-square and p-values are reported below . The only observed differences across iterations were driven by species in the ‘mango, mangosteen and guava’ category, blueberries in containers which are grouped by the FAO despite distant phylogenetic relatedness. We also performed a phylogenetically corrected LMM to compare the relationship between AC-1 and CV.Alternate bearing is common among managed perennial crops at the national level. A negative lag-1 autocorrelation is significantly more common in wind- than insect-pollinated plants.
These results are in line with previous findings in natural plant systems that masting is more common and more pronounced among wind-pollinated taxa. Bimodality was rare in these time series and did not differ between pollination syndromes, which is also consistent with natural systems where strong bimodality is uncommon and ‘partial masting’ prevails. Given that we observed a strong signal of alternate bearing despite limitations of the FAO dataset , our findings offer encouraging support that enhanced integration of the alternate-bearing and masting literature could offer mutual insight. Unsurprisingly, the CV of alternate-bearing crops reported here at national scales is smaller than has been observed for both masting and alternate-bearing taxa at population scales. In general, alternate bearing leads to lower CVs than mast-seeding at longer intervals. More importantly, we expect that in crop plants, as in wild plants, synchrony should decay with distance, and, at the present time, the scale of synchrony in crop plants is largely unknown. To explore this spatial scale somewhat quantitatively, we compared our results to data presented by Noble et al., one of the few published studies of yield variability at multiple scales in a crop plant. Noble et al. provide data on pistachio yield in four Californian counties. In their data, farm-level yield was more variable than county-level yield . These are all higher than estimates of pistachio CV at a national scale from the FAO data presented here . This very limited exploration suggests that synchrony across crop yields decays with distance, and that further exploration of farm- to regional-scale data for crop plants could be valuable for understanding spatial synchrony, especially in insect-pollinated species. Yield stability is frequently the goal of farmers and horticultural researchers, yet our results suggest that alternate bearing persists even at a national level. At the present time, studies on drivers of synchrony in agricultural settings are scarce and would benefit from integration with ecological theories about the causes of mast-seeding. For example, numerous studies of crop pollination are based on the premise of enhancing yield in a single year. However, if the alternate bearing is due to resource depletion after high-seed years, then maximizing yield in 1 year could lead to greater variability in yield, an undesirable outcome. The long-term impact of increased pollination will be affected by whether or not the crop tends to bear alternately at the floral initiation stage or in the flower-to-fruit conversion stage and by the relative cost of seeds to flowers. These aspects of plant development are rarely integrated into models of crop yield but would be straightforward to measure and implement to better align insect conservation and farmer priorities. We observed signals of numerous environmental shocks in crops at a national scale, but their role as a driver of synchrony in alternate bearing is largely unknown and presents another opportunity for masting theory to inform horticultural understanding. Environmental vetoes—external conditions that prevent seed set—have been well-supported as a driver of synchrony in masting systems. As a recent example, Schermer et al. studied frost-induced fruit losses in relationship to flower phenology in oaks. They concluded that a delay in flowering would lead to a more deterministic pattern of seed production characterized by a lower CV at the population level and a more pronounced lag-1 autocorrelation, which strongly resembles an alternate-bearing pattern . By contrast, an advancing flower phenology was predicted to increase the stochastic component of interannual variation characteristic of masting. Agricultural studies have offered stronger support for the latter scenario, i.e. advances in flower phenology with climate change, but have not explored the implications of advancing phenology on yield patterns. Greater stochasticity in natural systems can be advantageous as a pest control agent ; however, farmers are unlikely to experience a net benefit from increased variability given their reliance on a steady income. Our understanding of mast-seeding would similarly benefit from enhanced integration with agricultural crop data. First, given their direct socio-economic implications, long-term yield datasets are widely available for crop plants and can scale from individual plants to farms to national scales. Our results suggest these data present an underused resource for understanding perennial variability and synchrony.