How the Black Swan damages the harvest: Extreme weather events and the fragility of agriculture in development countries

Nadine Marmai; Maria Franco Villoria; Marco Guerzoni

doi:10.1371/journal.pone.0261839

Abstract

Climate change constitutes a rising challenge to the agricultural base of developing countries. Most of the literature has focused on the impact of changes in the means of weather variables on mean changes in production and has found very little impact of weather upon agricultural production. Instead, we focus on the relationship between extreme events in weather and extreme losses in crop production. Indeed, extreme events are of the greatest interest for scholars and policy makers only when they carry extraordinary negative effects. We build on this idea and for the first time, we adopt a conditional dependence model for multivariate extreme values to understand the impact of extreme weather on agricultural production. Specifically, we look at the probability that an extreme event drastically reduces the harvest of any of the major crops. This analysis, which is run on data for six different crops and four different weather variables in a vast array of countries in Africa, Asia and Latin America, shows that extremes in weather and yield losses of major staples are associated events. We find a high heterogeneity across both countries and crops and we are able to predict per country and per crop the risk of a yield reduction above 90% when extreme events occur. As policy implication, we can thus assess which major crop in each country is less resilient to climate shocks.

Citation: Marmai N, Franco Villoria M, Guerzoni M (2022) How the Black Swan damages the harvest: Extreme weather events and the fragility of agriculture in development countries. PLoS ONE 17(2): e0261839. https://doi.org/10.1371/journal.pone.0261839

Editor: Shamsuddin Shahid, Universiti Teknologi Malaysia, MALAYSIA

Received: June 11, 2021; Accepted: December 10, 2021; Published: February 2, 2022

Copyright: © 2022 Marmai et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The data underlying the results presented in the study are available from http://faostat3.fao.org http://unstats.un.org/unsd/methods/m49/m49regin.htm.

Funding: The author(s) received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Introduction

This paper investigates the effect of temperature and precipitation extremes on major staple crops in different regions of Asia, Africa, and Latin America. We provide two key contributions to the literature. First, we adopt a conditional dependence model for multivariate extreme values developed by [1]. Although this modelling approach already has some applications in environmental or food chemicals studies, see e.g. [2, 3], there has been no previous attempt to use the model in the context of extremes in weather and crop production losses. Secondly, we provide evidence that these extreme events are associated and we are able to estimate their dependence structure.

The relation between climate and agriculture is a highly debated issue. By focusing on the agricultural sector of countries in Latin America, Africa and Asia, numerous studies discussed the impact on weather variables on food production [4–13]. For example, [4] found both positive and negative impacts of temperature and precipitation trends for different major crops at the global level. Specifically, trends in temperature affect mainly yield, whereas precipitation influences inter-annual changes in crop production. More recently, their estimates have been corroborated by [12].

This stream of research makes use of linear regression models and focuses on the mean effects of weather on average crop production [4, 14, 15]. The results of, e.g., [16] reveal that time constant country-fixed effects and time trends explain most of the variation in yields of different agricultural products. Regional characteristics such as soil quality or crop management and country-specific trends, e.g., technological progress in crop production or warming, are the most crucial factors, whereas annual mean changes in weather provide only a minor explanation of the overall variation. This approach suggests that the relation between weather and production is non-linear and difficult to model with linear regression analysis. [17] discuss the importance of non-linear responses to temperature in agricultural and non-agricultural production for both rich and poor countries: production peaks at an annual average temperature of 13 degrees Celsius and decreases substantially at higher temperatures. However, the average temperature is higher in poor countries, which leads to stronger effects of temperature on production in these countries. Projections suggest that further warming will reduce productivity and income in countries with high average temperatures.

The assessment of weather impacts on crop production includes not only the focus on changes in the mean values of weather variables but also on the probability, frequency, and severity of extreme events, which substantially influence yields [18–22]. Several empirical works, such as [23] in European countries, [24] in the United States, [25] in the UK, and [26] in France show a substantial impact of extreme weather events on the yields for major crops. Moreover, changes in the frequency of extreme weather events also determine the quality of the crop harvest [27]. [22] estimate the impact of heat stress and drought oat the global level over the period 1980–2010.

This evidence deserves a deeper analysis, since both the frequency and the magnitude of extreme weather events such as heat-waves are likely to rise due to warming climate conditions [25]. Looking at countries globally, [28] find extreme heat events to be unfavourable for major producing regions and lower income countries and, according to [25], extreme high and low temperature can seriously harm crops or even cause plant death, whereas intensive precipitation can lead to contamination of ground water and soil erosion. In addition to the direct effects of heat, drought, and flooding, extreme events indirectly affect crops through pests, changing soil processes and nutrient dynamics [29]. Similarly, [20] analyses the damages of extreme weather disasters on crop production and find that drought and extreme heat significantly harm national staple crop production worldwide. More recently, [30] make use of threshold effects of extreme weather events to estimate cereal yields in India.

Based on these studies, we surmise that the missing piece in the literature consists of relating the extreme events in weather with substantial losses in production. Indeed, extreme events are of the greatest interest for scholars and policy makers when they have extraordinary negative impacts [31]; otherwise the extreme events are irrelevant. Specifically, we test the hypothesis that the tails of weather variables and crops yields are associated. We also test that this association is heterogeneous across countries and staple crops. In the remainder of the paper, we model the conditional extreme dependence of extraordinary yields losses with four different weather variables for six different staple crops, separately for different regions in Asia, Africa and Latin America.

Materials and methods

Data

This study covers countries in Africa, Asia, and Latin America ranging from low to upper middle income countries [32] and includes annual data observations from 1961 to 2002. The staple crops of interest are wheat, rice, maize, soybeans, barley, and sorghum which constitute the six most commonly cultivated crops worldwide [6]. Country-level data on yields are available from the FAO website(http://faostat3.fao.org).

We make use of [4]’s precipitation and temperature data; the authors constructed weather data based both on [33]’s crop calendar to derive the growing season of each crop and on the agricultural maps of [34] to identify the growing regions of each crop. [4] extracted growing season- and region-specific weather data from the CRU TS 2.1 historical climate data set [35] and created national precipitation and temperature aggregates for each year. Different growing regions and different growing seasons among countries result in different precipitation and temperature data by crop and by country. Table 1 gives an overview of the variables.

Download:

Table 1. Overview of variables.

https://doi.org/10.1371/journal.pone.0261839.t001

The initial sample consists of 42 yearly observations for each country and each variable of analysis. We omit year-country pairs when data are missing for one of the variables. Because 42 observations constitute a small sample to conduct an analysis of extreme values, we pool countries into regions based on the UN Statistics Division composition of geographical regions (http://unstats.un.org/unsd/methods/m49/m49regin.htm). Table 2 summarizes the geographical areas, which consist of countries geographically related but still not identical in terms of economic and weather conditions. To account for this heterogeneity within each pool of countries, we standardize the production and weather data independently for each country, subtracting the country-specific mean and dividing by the country-specific standard deviation. This standardization ensures comparability and avoids missing extremes for some countries when pooling data.

Download:

Table 2. Geographical regions.

https://doi.org/10.1371/journal.pone.0261839.t002

Fig 1 shows scatter plots of the standardized yields (Yield), precipitation (Prec), maximum temperature (tMax) and minimum temperature (tMin), using maize data in Eastern Africa as a representative example of the data. Fig 1 shows that a regression analysis is not a convenient option because there are no grounds to assume a clear positive or clear negative dependence of the yield and any of the weather variables. Instead, we model the extremes, that is for instance, the set of observations in the bottom right corner of Fig 1a.

Download:

Fig 1. Annual standardized yield and precipitation (a), maximum temperature (b), and minimum temperature (c), using maize data in Eastern Africa (1961–2002).

https://doi.org/10.1371/journal.pone.0261839.g001

We define extreme events in precipitation, temperature, and yield as the values of the corresponding variable above (below) a threshold value, so that the observations subject to the analysis are located in the upper (lower) tail of the distribution [36]. We are mostly interested in the effect that extremely high or low temperatures and high or low precipitations have on the probability of observing severe yield losses.

Methods

Let X = (X₁, …, X_d) be a vector variable with unknown distribution function F(x), and marginal distribution functions , with i = 1, …, d. The idea is to model the joint tail of F(x) and, more specifically, the conditional distribution of X_−i|X_i > x when x is large, where X_−i denotes the vector X excluding the i^th component. In this framework, we follow [1], who proposed a semi-parametric model for the marginal distributions based on the generalized Pareto distribution (GPD), where (β_i,ξ_i) are the scale and shape parameters of a GPD for the exceedances over the threshold and is the empirical distribution of X_i.

Following [37], we use the estimated distributions to transform X component-wise to follow Laplace marginal distributions: (1)

The aim is to model the distribution of Y_−i|Y_i = y for y large. For that purpose, univariate extreme value theory is extended to a multivariate context. Assume that there exist normalizing functions a_|i(x), b_|i(x): / ∀ fixed and any sequence of y_i values such that y_i → ∞ (i.e., high enough): (2)

Denote by G_i the i^th marginal distribution of G_|i, a non-degenerate distribution function with lim_z→∞{G_i(z)} = 1 ∀i. The method assumes that the limiting distribution holds ∀ for a suitable high threshold . When Y_i = y_i, with , the (standardized) random variable Z_|i is defined as: (3) and the limiting distribution of Z_|i: (4)

Under this assumption, conditionally on , as , the variables and Z_|i are independent in the limit and their limiting marginal distributions are exponential and G_|i(z_|i), respectively [37].

The extremal dependence behaviour is characterized by a_|i(y), b_|i(y) and G_|i; hence estimates of the three are needed to derive the conditional distribution. To do so, [1] propose a semi-parametric model. The parametric part involves estimating a_|i(y) and b_|i(y) using the regression model: (5)

Specifically, a_|i(y) and b_|i(y) are expressed in terms of y as a_|i(y) = a_|i y and , with the restrictions (a_|i, b_|i) ∈ [−1, 1]^d−1 × (−∞, 1)^d−1. Further joint constraints on the parameters have been imposed by [37] to avoid problems of inconsistent inferences with respect to the marginal distributions and parameter identification. Positive and negative dependence are defined by a_j|i, the component of a_|i linked to Y_j and large Y_i, being 0 < a_j|i ≤ 1 and −1 ≤ a_j|i < 0 respectively [37]. Assuming that (a_|i, b_|i) are known, G_|i can be estimated non-parametrically using the empirical (or kernel smoothed) distribution of replicates of the random variable : for . Pseudo-samples can then be generated using the fitted model to estimate the conditional probability of interest. Confidence intervals can be obtained using bootstrap methods, see [1] for computational details.

Results

Empirical analysis

In the case of Prec, we consider both extreme high and low precipitation values. Because [1] model the upper tail of the distribution, for extreme low precipitation, we consider the reflection of the variable Prec, defined as PrecRefl. The same procedure applies to Yield because we are interested in yield losses and we therefore consider the reflection YieldRefl. For each crop, we define four two-dimensional vectors X = (X₁, X₂) with unknown distribution function F(x), where X₁ is always YieldRefl of the specific crop and X₂ is one of the four weather variables. We thus model the extreme values in a bivariate context and run a separate analysis for each pair of crop and weather variables. To illustrate the procedure of fitting the dependence model, we report results of the two variables YieldRefl and Prec using maize data of Eastern Africa from 1961 to 2002. Results for the set of specifications with different marginal weather variables and for different crops or regions are available on request. The analysis was conducted using the R packages texmex [38], evd [39], ggplot2 [40] and rworldmap [41].

We first fit a generalized Pareto distribution separately for each variable and then transform X component-wise following Eq (1) to obtain identical Laplace marginal distributions. The fitting of the GPD requires an appropriate selection of the threshold above which the GPD model is valid. For this reason, linearity of the mean residual life plots has to be ensured, which is already the case for very low thresholds of Prec and YieldRefl for maize data in Eastern Africa. However, probability, quantile and return level plots suggest that the estimated distribution function is a reasonable estimate of the theoretical function only above the threshold of the 80^th percentile [36], which we then choose as the threshold for Prec and YieldRefl.

Based on these marginal variables with identical Laplace distributions, we describe the dependence structure of the variables. We explore the behaviour of the variable YieldRefl conditional on extreme values of the variable Prec. We choose the threshold over which the limiting distribution holds by examining the threshold stability of the estimated parameters a_|i and b_|i of the dependence model (5) using the 50^th to 90^th quantiles of the conditioning variable Prec as potential thresholds. The 90^th quantile was found to be adequate, leading to parameter estimates and . Imposing the ordering constraints to the values of the parameters as explained in Section Methods, parameter estimates can be sensitive to the initial value of the optimization procedure for the estimation [42]. However, constrained dependence parameter estimates are located in the maximum of the profile likelihood surface, as shown in Fig 2. With , the variables YieldRefl and Prec exhibit positive extremal dependence [37]. Using maize data from Eastern Africa, we see that the form of dependence between YieldRefl and the weather variables varies. In contrast to YieldRefl and Prec, the pair YieldRefl and PrecRefl shows an extremal negative dependence with an . The results of the other regions and for the other crops show that the dependence structure does not only change for different conditioning variables but also for different crops and regions. The uncertainty of the parameter estimates was evaluated by means of bootstrapping, see [1] for details.

Download:

Fig 2. Constrained dependence parameter estimates

(a) and

(b) of the conditional distribution of (YieldRefl|Prec) conditional on Prec > qu_Prec, with qu_Prec being the 90^th quantile of the conditioning variable Prec, correspond to the maximum of the profile likelihood surface using maize data of Eastern Africa.

https://doi.org/10.1371/journal.pone.0261839.g002

Finally, plotting Z_|i against Y_i for values of Y_i over the threshold chosen to fit the model, we conclude that the standardized variable and the conditioning variable are independent. Moreover, the fitted quantiles of the conditional distribution and the observations on the original scale match indicating that the model fits well.

Extrapolation

The semiparametric model fitted in the previous section is used to simulate from the joint distribution of (YieldRefl, Prec) conditional on Prec > qu_Prec, where qu_Prec is the 91^st to 99,99^th quantile of the conditioning variable Prec. We simulate 1000 observations, which are then used to calculate the conditional probabilities P(YieldRefl > qu_YieldRefl|Prec > qu_Prec), where qu_YieldRefl is always set as the 90^th quantile of the variable YieldRefl. Thus, we are interested in the change of the probability of high losses in basic food production given rising extremes in precipitation. The uncertainty of each point estimate is assessed by creating 95% confidence intervals of the conditional probabilities based on 100 bootstrap samples.

Fig 3 contains conditional probabilities and the corresponding confidence intervals of high losses in maize production given a range of thresholds, i.e. from 91^st to 99.99^th quantiles, for high precipitation. The plots in Fig 3 cover all considered regions in Asia, Africa and Latin America. In Middle & Southern Africa, the conditional probabilities of lower confidence interval bounds are equal to zero, indicating no evidence of an association between extremes in high precipitation and high yield losses. The probability of high losses in maize production given increasing extremes in high precipitation increases sharply up to 94% in Eastern Africa but with widening confidence intervals. Eastern Africa aside, the probabilities do not change significantly with increasing thresholds for high precipitation among the different regions and are mostly below 25%. For the opposite case of extremes in low precipitation, Fig 6 in S1 Annex plots the conditional probabilities for all regions. South America is affected more than the other regions because losses in its maize production are more likely to occur given extremely low rainfall. Point estimates up to 74% show higher uncertainties, although the confidence intervals do not include zero. In Central America & Caribbean, the lower confidence interval bounds of the conditional probabilities are equal to zero. The conditional probabilities for Eastern Africa are approximately 25% and are less than 25% for the other regions. Summing up, events of extremely high precipitation affect maize production the most in Eastern Africa and the least in Middle and South Africa, while drought has severe consequences on Maize production in South America and limited ones in In Central America & Caribbean.

Download:

Fig 3. Point estimates and 95% confidence intervals of the conditional probability P(YieldRefl > qu_YieldRefl|Prec > qu_Prec), where qu_YieldRefl is always set as the 90^th quantile of the variable YieldRefl and qu_Prec is the 91^st to 99,99^th quantile of the conditioning variable Prec.

Estimation is done using maize data from 1961 to 2002. In Eastern Africa conditional probabilities sharply increase with widening confidence intervals for high thresholds of the conditioning variable. In Middle & Southern Africa the conditional probabilities or the lower confidence interval bounds are zero indicating no evidence of an association between extremes in high precipitation and high yield losses.

https://doi.org/10.1371/journal.pone.0261839.g003

In the case of extremes in minimum temperature, see Fig 7 in S1 Annex, the lower confidence interval bounds are equal to zero in South America, Western Africa and South & South-Eastern Asia & Melanesia. The other regions do not show remarkable differences. In Central America & Caribbean and Eastern Africa, the conditional probabilities slightly increase with higher extremes in minimum temperature. However, the point estimates are around or below 25%. The picture looks similar for the maximum temperature as the conditioning variable, which is shown in Fig 8 in S1 Annex. The conditional probabilities rarely exceed 25%, and the lower confidence interval bounds are equal to zero in Western Africa and South & South-Eastern Asia & Melanesia. The conditional probability plots for barley, rice, sorghum, soy and wheat also reveal a mixed picture and are shown in the supporting material to this paper [43, pp. 20–35]. Production losses due to weather extremes are not equally likely among the regions. Depending on the crop, the weather extremes and the region, the occurrence of severe production losses is more or less likely. The results emphasize the complexity of interaction factors.

Fig 4 visually summarizes this heterogeneity and displays the highest point estimates and the 95% confidence intervals of the probability of observing a high loss in yield, i.e., the probability to observe a reduction of Yield below the 10%, conditional to the occurrence of a weather extreme in the 98^th quantile.

Download:

Fig 4. Highest point estimates of the conditional probabilities, i.e. the probabilities of yield losses above the 90^th quantile given minimum temperature, maximum temperature or high precipitation extremes above the 98^th quantile or low precipitation extremes below the 2^nd quantile.

If the 95% confidence interval includes zero, the conditional probabilities are not shown.

https://doi.org/10.1371/journal.pone.0261839.g004

Fig 4 shows that each of the considered regions is likely to have severe losses in its production of staple food due to extreme weather events. However, the effect of extreme weather varies among regions depending on the crop and the type of weather extreme. Sorghum, which is the main staple in Africa, has the highest risk of severe yield losses for different weather extremes in the three African regions. Extreme high maximum temperature and low precipitation are the main weather conditions leading to extraordinary yield losses, which is in line with the distribution of sorghum in arid regions in Africa or in regions where precipitation is erratic and characterized by short periods of high precipitation [44]. In Eastern and Western Africa extreme high minimum temperature lead to severe reduction in sorghum yield. High losses in maize production are due to extreme high minimum temperature in Middle & Southern Africa and are due to extreme high precipitation in Eastern Africa. Noteworthy because of the importance of rice, which is a main staple in Western Africa [45], high losses in rice production are likely to occur with extreme events in high precipitation.

Central America & Caribbean exhibits the highest conditional probabilities due to weather extremes for rice and maize, whereas South America shows it for barley and soybean which are among the most important staple crops in the regions. Weather patterns are diverse in Latin America. Whereas South America experiences rising temperature and changing precipitation patterns leading to mixed effects on agriculture, an increase in temperature severely damages agricultural output in Central America. Moreover, even though Central America is marked by a decline in precipitation, floods remain among the most frequent extreme weather events [46]. The estimated conditional probabilities suggest that in Central America & Caribbean maize production reacts mainly to extremes in high precipitation and temperature and rice production reacts to extreme low precipitation. In South America, barley and soybean production losses are likely to increase with extremes in temperature and low precipitation respectively.

In Asia a main staple crop is rice where 90% of rice production and consumption is concentrated [45]. The dependency on rice is reflected by the fact that we obtain the highest conditional probabilities for rice in South & South-Eastern Asia & Melanesia. Losses in rice production are likely due to extreme high minimum temperature. The finding is in line with [45] who state that higher minimum temperatures become increasingly a major cause of yield losses of rice in Asia. On the other side, the occurrence of very low rice yields is likely due to high precipitation. Only in South Asia yield losses of rice due to floods are about 4 million t per year [45]. Interestingly, the probability of production losses given extreme low precipitation is the highest for maize in the tropical region of South & South-Eastern Asia & Melanesia. The finding is plausible as maize production is predominantly rain-fed in South and South-Eastern Asia [47].

The point estimates of conditional probabilities shown in Fig 4 are at maximum 40%, which is the case for sorghum in Eastern Africa given extreme low precipitation. Overall, the highest conditional probabilities given different weather extremes are in Eastern and Middle & Southern Africa as well as in South America. In contrast, the conditional probabilities are less than 30% for all types of weather extremes in Central America & Caribbean and South & South-Eastern Asia & Melanesia.

Summing up the results, we find that, first, maize and sorghum have the highest conditional probabilities of extreme high losses in crop production given the occurrence of extreme weather conditions. Second, extreme low precipitation and extreme high maximum temperature are the defining weather extremes, except for Latin America and Western Africa. In Latin America, the probabilities do not change significantly among different conditioning variables, and in Western Africa, the highest conditional probability is due to extreme high precipitation in the case of rice. Third, losses in staple crop production given extreme weather events are more likely in the African regions and South America.

Finally, Fig 5 shows the worst-case scenario by displaying the maximum upper bound of the 95% confidence interval of the conditional probability estimates, i.e., the probability estimates of yield losses above the 90^th quantile given the minimum temperature, maximum temperature or high precipitation extremes above the 98^th quantile or low precipitation extremes below the 2^nd quantile. The maximum upper bound of the 95% confidence interval is chosen from all of the weather variables and crops in a region. Each region has a different worst-case scenario, i.e., a different crop and weather variable for which we obtain the maximum upper confidence interval bound. In Central America & Caribbean, South America, Eastern Africa and Middle & Southern Africa, extreme high temperature is the condition that leads to the highest production losses. Whereas in the African regions, the associated crop is sorghum, the associated crop is maize in Central America & Caribbean and barley in South America. The worst-case scenario for Western Africa constitutes high losses in rice yields due to extremes in high precipitation. In the Asian region, extremes in low precipitation is the defining weather variable in the worst-case scenario and rice is the affected crop. The worst-case scenarios occur with different probability in the different regions. The probability ranges from around 25% in Central America & Caribbean and South & South-Eastern Asia & Melanesia to 56% in Eastern Africa and South America, whereas probabilities of approximately 50% are found in Western and Middle & Southern Africa.

Download:

Fig 5. Worst-case scenario: The maximum upper bound of the 95% confidence interval of the conditional probability estimates, i.e. the probability estimates of yield losses above the 90^th quantile given minimum temperature, maximum temperature or high precipitation extremes above the 98^th quantile or low precipitation extremes below the 2^nd quantile.

https://doi.org/10.1371/journal.pone.0261839.g005

This worst case scenario can be employed as a fragility index which can be used to assess strengths and weakness of different areas conditional to the type of crops they mostly harvest. It can thus became a valuable instrument for policy design. As a matter of fact, the same index can be applied to different areas other than agriculture, but still linked with climate change, such as desertification, diffusion of deceases, and poverty rate.

Discussion and conclusion

The influential paper on African countries by [16] reports an expected 7% reduction of aggregate agriculture productivity with a probability of 95% probability due to change in the mean of weather variables in Sub-Saharan Africa with an R-squared between 0.5 and 0.7 according to the type of staple. Their work, as any other standard approach for assessing the impact of weather on crops has largely focused on the mean values. In line with a recent stream of research on extreme weather events, we do not focus on the mean but on the tails of the distributions. We find systematic evidence that extreme losses in production of major food staples are likely to occur in times of extreme conditions in weather and the probability depends on the type of weather extreme, the crop, and the region. The alleged increase in the number of extreme events should be taken seriously because the potential damage can be extraordinary. As a comparison with [16], independently on the mean, seasonal extreme events of drought can reduce Sorghum production in Sub-Saharan Africa by 40%, that is 7 times higher than the projected reduction of 7%. We are able to provide this measure of risk for each region, that is specific to both the crop and the extreme weather event. This outcome can also be used as a index for the resilience of a region and to help policy makers to intervene and set priorities. In the short run, a country cannot change the risk of incurring high losses due to weather events. However, the resilience of a country depends on many factors, such as the adoption of advanced irrigation technology, the diffusion of fertilizer, and the introduction of new and resistant crops, which can be highly influenced by the policy makers. This work is highly complementary with the ones focusing on the mean impact of the present trend of weather variables. While these works [6, 16, 24, 48, 49] can provide an assessment of the future long term development on average, we highlight that this path will be punctuated by events with a much higher disrupting effect. Thus, long term policy for climate change intervention should be accompanied by immediate action for counteracting the impact of very likely extreme events.

This work leaves some questions unanswered due to data availability. Specifically, the same type of weather extremes can lead to different crop responses based on the time of the year and the crop growth stage [26]. The use of growing-season national aggregates of weather to account for weather extremes instead of using precise weather data is critical because growing-season aggregates do not to capture inter-annual ups and downs and extremes within a growing period. The drawback occurs because extreme weather conditions are particularly harmful in certain stages of plant growth [50]. In addition, studies that focus on the occurrence of extreme weather events have to be accompanied by studies that also include the severity of extremes, such as the length of heatwaves or floods. The main constraint for a large fraction of countries worldwide is the lack of access to long-term weather data with high time and spatial resolution [51], which are now available for case studies limited in time and space. We try to mitigate this problem by looking at aggregates of weather in the growing season and for different growing regions and we analyse the occurrence of periods of extreme hot, dry or wet conditions from a long-term perspective. The value of this paper is to provide an indication of the probability of extreme losses in basic food production due to extreme weather events in a specific region under historic climate conditions. This analysis provides a perspective that is complementary to more detailed localized studies focusing on one particular region with higher time and spatial resolution data.

To conclude, studying weather changes and the impacts on agricultural production remains a challenging task. To evaluate crop production responses the uncertainty of weather changes needs to be further addressed. The uncertainty in the evaluation of current and future impacts of weather on agricultural production stems from uncertainties that arise in the estimation of crop responses to changes in average growing season temperature and precipitation [48]. We believe that the uncertainties also come from the extreme events that are by nature difficult to model. Given that future climate is likely to be prone to a higher frequency of extreme weather events, our results contribute to this discussion by providing the probabilities of severe staple crop production losses conditioned on extremes in temperature and precipitation.

Supporting information

S1 Annex.

https://doi.org/10.1371/journal.pone.0261839.s001

(ZIP)

References

1. Heffernan JE, Tawn JA. A conditional approach for multivariate extreme values (with discussion). Journal of the Royal Statistical Society Series B. 2004;66(3):497–546.
- View Article
- Google Scholar
2. Keef C, Tawn J, Svensson C. Spatial risk assessment for extreme river flows. Journal of the Royal Statistical Society: Series C (Applied Statistics). 2009;58(5):601–618.
- View Article
- Google Scholar
3. Paulo M, Van der Voet ht, Wood J, Marion G, Van Klaveren J. Analysis of multivariate extreme intakes of food chemicals. Food and chemical toxicology. 2006;44(7):994–1005. pmid:16457929
- View Article
- PubMed/NCBI
- Google Scholar
4. Lobell DB, Schlenker W, Costa-Roberts J. Climate trends and global crop production since 1980. Science. 2011;333(6042):616–620. pmid:21551030
- View Article
- PubMed/NCBI
- Google Scholar
5. Jones PG, Thornton PK. The potential impacts of climate change on maize production in Africa and Latin America in 2055. Global environmental change. 2003;13(1):51–59.
- View Article
- Google Scholar
6. Lobell DB, Field CB. Global scale climate–crop yield relationships and the impacts of recent warming. Environmental research letters. 2007;2(1):014002.
- View Article
- Google Scholar
7. Kristensen K, Schelde K, Olesen JE. Winter wheat yield response to climate variability in Denmark. The Journal of Agricultural Science. 2011;149(01):33–47.
- View Article
- Google Scholar
8. Lobell DB, Bänziger M, Magorokosho C, Vivek B. Nonlinear heat effects on African maize as evidenced by historical yield trials. Nature Climate Change. 2011;1(1):42–45.
- View Article
- Google Scholar
9. Rowhani P, Lobell DB, Linderman M, Ramankutty N. Climate variability and crop production in Tanzania. Agricultural and Forest Meteorology. 2011;151(4):449–460.
- View Article
- Google Scholar
10. Welch JR, Vincent JR, Auffhammer M, Moya PF, Dobermann A, Dawe D. Rice yields in tropical/subtropical Asia exhibit large but opposing sensitivities to minimum and maximum temperatures. Proceedings of the National Academy of Sciences. 2010;107(33):14562–14567. pmid:20696908
- View Article
- PubMed/NCBI
- Google Scholar
11. Knox J, Hess T, Daccache A, Wheeler T. Climate change impacts on crop productivity in Africa and South Asia. Environmental Research Letters. 2012;7(3):034032.
- View Article
- Google Scholar
12. Zhao C, Liu B, Piao S, Wang X, Lobell DB, Huang Y, et al. Temperature increase reduces global yields of major crops in four independent estimates. Proceedings of the National Academy of Sciences. 2017;114(35):9326–9331. pmid:28811375
- View Article
- PubMed/NCBI
- Google Scholar
13. Das B, Nair B, Reddy VK, Venkatesh P. Evaluation of multiple linear, neural network and penalised regression models for prediction of rice yield based on weather parameters for west coast of India. International journal of biometeorology. 2018;62(10):1809–1822. pmid:30043218
- View Article
- PubMed/NCBI
- Google Scholar
14. Kucharik CJ, Serbin SP. Impacts of recent climate change on Wisconsin corn and soybean yield trends. Environmental Research Letters. 2008;3(3):034003.
- View Article
- Google Scholar
15. Tao F, Yokozawa M, Liu J, Zhang Z, et al. Climate-crop yield relationships at provincial scales in China and the impacts of recent climate trends. Climate Research. 2008;38(1):83–94.
- View Article
- Google Scholar
16. Schlenker W, Lobell DB. Robust negative impacts of climate change on African agriculture. Environmental Research Letters. 2010;5(1):014010.
- View Article
- Google Scholar
17. Burke M, Hsiang SM, Miguel E. Global non-linear effect of temperature on economic production. Nature. 2015;. pmid:26503051
- View Article
- PubMed/NCBI
- Google Scholar
18. Wheeler TR, Craufurd PQ, Ellis RH, Porter JR, Prasad PV. Temperature variability and the yield of annual crops. Agriculture, Ecosystems & Environment. 2000;82(1):159–167.
- View Article
- Google Scholar
19. Moriondo M, Giannakopoulos C, Bindi M. Climate change impact assessment: the role of climate extremes in crop yield simulation. Climatic Change. 2011;104(3-4):679–701.
- View Article
- Google Scholar
20. Lesk C, Rowhani P, Ramankutty N. Influence of extreme weather disasters on global crop production. Nature. 2016;529(7584):84–87. pmid:26738594
- View Article
- PubMed/NCBI
- Google Scholar
21. Schewe J, Gosling SN, Reyer C, Zhao F, Ciais P, Elliott J, et al. State-of-the-art global models underestimate impacts from climate extremes. Nature communications. 2019;10(1):1–14. pmid:30824763
- View Article
- PubMed/NCBI
- Google Scholar
22. Zampieri M, Ceglar A, Dentener F, Dosio A, Naumann G, Van Den Berg M, et al. When will current climate extremes affecting maize production become the norm? Earth’s Future. 2019;7(2):113–122.
- View Article
- Google Scholar
23. Ciais P, Reichstein M, Viovy N, Granier A, Ogée J, Allard V, et al. Europe-wide reduction in primary productivity caused by the heat and drought in 2003. Nature. 2005;437(7058):529–533. pmid:16177786
- View Article
- PubMed/NCBI
- Google Scholar
24. Schlenker W, Roberts MJ. Nonlinear effects of weather on corn yields. Applied Economic Perspectives and Policy. 2006;28(3):391–398.
- View Article
- Google Scholar
25. Semenov MA. Development of high-resolution UKCIP02-based climate change scenarios in the UK. Agricultural and Forest Meteorology. 2007;144(1):127–138.
- View Article
- Google Scholar
26. van der Velde M, Tubiello FN, Vrieling A, Bouraoui F. Impacts of extreme weather on wheat and maize in France: evaluating regional crop simulations against observed data. Climatic change. 2012;113(3-4):751–765.
- View Article
- Google Scholar
27. Porter JR, Semenov MA. Crop responses to climatic variation. Philosophical Transactions of the Royal Society B: Biological Sciences. 2005;360(1463):2021–2035. pmid:16433091
- View Article
- PubMed/NCBI
- Google Scholar
28. Deryng D, Conway D, Ramankutty N, Price J, Warren R. Global crop yield response to extreme heat stress under multiple climate change futures. Environmental Research Letters. 2014;9(3):034011.
- View Article
- Google Scholar
29. Rosenzweig C, Iglesias A, Yang X, Epstein PR, Chivian E. Climate change and extreme weather events; implications for food production, plant diseases, and pests. Global change & human health. 2001;2(2):90–104.
- View Article
- Google Scholar
30. Fontes F, Gorst A, Palmer C. Threshold effects of extreme weather events on cereal yields in India. Climatic Change. 2021;165(1):1–20.
- View Article
- Google Scholar
31. Taleb NN. The black swan: The impact of the highly improbable fragility. vol. 2. Random House; 2010.
32. UN. World Economic Situation and Prospects 2014; 2014.
33. Sacks WJ, Deryng D, Foley JA, Ramankutty N. Crop planting dates: an analysis of global patterns. Global Ecology and Biogeography. 2010;19(5):607–620.
- View Article
- Google Scholar
34. Monfreda C, Ramankutty N, Foley JA. Farming the planet: 2. Geographic distribution of crop areas, yields, physiological types, and net primary production in the year 2000. Global biogeochemical cycles. 2008;22(1).
- View Article
- Google Scholar
35. Mitchell TD, Jones PD. An improved method of constructing a database of monthly climate observations and associated high-resolution grids. International Journal of Climatology. 2005;25(6):693–712.
- View Article
- Google Scholar
36. Coles S, Bawa J, Trenner L, Dorazio P. An introduction to statistical modeling of extreme values. vol. 208. Springer; 2001.
37. Keef C, Papastathopoulos I, Tawn JA. Estimation of the conditional distribution of a multivariate variable given that one of its components is large: Additional constraints for the Heffernan and Tawn model. Journal of Multivariate Analysis. 2013;115:396–404.
- View Article
- Google Scholar
38. Southworth J Harry; Heffernan. texmex: Statistical modelling of extreme values. R package Version 21. 2013;.
39. Stephenson A. evd: Functions for Extreme Value Distributions. R package Version 23. 2015;.
40. Wickham W Hadley; Chang. ggplot2: An Implementation of the Grammar of Graphics. R package Version 20. 2015;.
41. South A. Mapping Global Data. R package Version 13-6. 2016;.
42. Southworth H, Heffernan JE, Southworth MH. Package texmex. Statistics. 2013;.
43. Marmai N, Franco Villoria M, Guerzoni M. How the Black Swan damages the harvest: statistical modelling of extreme events in weather and crop production in Africa, Asia, and Latin America. Department of Economics and Statistics Cognetti de Martiis Working Paper Series University of Turin. 2016;.
44. Taylor J. Overview: Importance of sorghum in Africa. In: Proceedings of AFRIPRO Workshop on the Proteins of Sorghum And Millets: Enhancing Nutritional and Functional Properties for Africa, Pretoria, South Africa. vol. 9; 2003.
45. Maclean J, Hardy B, Hettel G. Rice Almanac: Source Book for One of the Most Important Economic Activities on Earth. IRRI; 2013.
46. Galindo LM, De Miguel CJ. Economics of climate change in Latin America and the Caribbean: summary. Agriculture Markets and Crop profiles. 2010;.
- View Article
- Google Scholar
47. Wani SP, Rockström J, Oweis TY, et al. Rainfed agriculture: unlocking the potential. vol. 7. CABI; 2009.
48. Lobell DB, Burke MB. Why are agricultural impacts of climate change so uncertain? The importance of temperature relative to precipitation. Environmental Research Letters. 2008;3(3):034007.
- View Article
- Google Scholar
49. Lobell DB, Cassman KG, Field CB. Crop yield gaps: their importance, magnitudes, and causes. Annual Review of Environment and Resources. 2009;34(1):179.
- View Article
- Google Scholar
50. Porter JR, Gawith M. Temperatures and the growth and development of wheat: a review. European Journal of Agronomy. 1999;10(1):23–36.
- View Article
- Google Scholar
51. Easterling DR, Evans J, Groisman PY, Karl T, Kunkel KE, Ambenje P. Observed variability and trends in extreme climate events: a brief review*. Bulletin of the American Meteorological Society. 2000;81(3):417–425.
- View Article
- Google Scholar

[ref1] 1. Heffernan JE, Tawn JA. A conditional approach for multivariate extreme values (with discussion). Journal of the Royal Statistical Society Series B. 2004;66(3):497–546.
View Article
Google Scholar

[2] View Article

[3] Google Scholar

[ref2] 2. Keef C, Tawn J, Svensson C. Spatial risk assessment for extreme river flows. Journal of the Royal Statistical Society: Series C (Applied Statistics). 2009;58(5):601–618.
View Article
Google Scholar

[5] View Article

[6] Google Scholar

[ref3] 3. Paulo M, Van der Voet ht, Wood J, Marion G, Van Klaveren J. Analysis of multivariate extreme intakes of food chemicals. Food and chemical toxicology. 2006;44(7):994–1005. pmid:16457929
View Article
PubMed/NCBI
Google Scholar

[8] View Article

[9] PubMed/NCBI

[10] Google Scholar

[ref4] 4. Lobell DB, Schlenker W, Costa-Roberts J. Climate trends and global crop production since 1980. Science. 2011;333(6042):616–620. pmid:21551030
View Article
PubMed/NCBI
Google Scholar

[12] View Article

[13] PubMed/NCBI

[14] Google Scholar

[ref5] 5. Jones PG, Thornton PK. The potential impacts of climate change on maize production in Africa and Latin America in 2055. Global environmental change. 2003;13(1):51–59.
View Article
Google Scholar

[16] View Article

[17] Google Scholar

[ref6] 6. Lobell DB, Field CB. Global scale climate–crop yield relationships and the impacts of recent warming. Environmental research letters. 2007;2(1):014002.
View Article
Google Scholar

[19] View Article

[20] Google Scholar

[ref7] 7. Kristensen K, Schelde K, Olesen JE. Winter wheat yield response to climate variability in Denmark. The Journal of Agricultural Science. 2011;149(01):33–47.
View Article
Google Scholar

[22] View Article

[23] Google Scholar

[ref8] 8. Lobell DB, Bänziger M, Magorokosho C, Vivek B. Nonlinear heat effects on African maize as evidenced by historical yield trials. Nature Climate Change. 2011;1(1):42–45.
View Article
Google Scholar

[25] View Article

[26] Google Scholar

[ref9] 9. Rowhani P, Lobell DB, Linderman M, Ramankutty N. Climate variability and crop production in Tanzania. Agricultural and Forest Meteorology. 2011;151(4):449–460.
View Article
Google Scholar

[28] View Article

[29] Google Scholar

[ref10] 10. Welch JR, Vincent JR, Auffhammer M, Moya PF, Dobermann A, Dawe D. Rice yields in tropical/subtropical Asia exhibit large but opposing sensitivities to minimum and maximum temperatures. Proceedings of the National Academy of Sciences. 2010;107(33):14562–14567. pmid:20696908
View Article
PubMed/NCBI
Google Scholar

[31] View Article

[32] PubMed/NCBI

[33] Google Scholar

[ref11] 11. Knox J, Hess T, Daccache A, Wheeler T. Climate change impacts on crop productivity in Africa and South Asia. Environmental Research Letters. 2012;7(3):034032.
View Article
Google Scholar

[35] View Article

[36] Google Scholar

[ref12] 12. Zhao C, Liu B, Piao S, Wang X, Lobell DB, Huang Y, et al. Temperature increase reduces global yields of major crops in four independent estimates. Proceedings of the National Academy of Sciences. 2017;114(35):9326–9331. pmid:28811375
View Article
PubMed/NCBI
Google Scholar

[38] View Article

[39] PubMed/NCBI

[40] Google Scholar

[ref13] 13. Das B, Nair B, Reddy VK, Venkatesh P. Evaluation of multiple linear, neural network and penalised regression models for prediction of rice yield based on weather parameters for west coast of India. International journal of biometeorology. 2018;62(10):1809–1822. pmid:30043218
View Article
PubMed/NCBI
Google Scholar

[42] View Article

[43] PubMed/NCBI

[44] Google Scholar

[ref14] 14. Kucharik CJ, Serbin SP. Impacts of recent climate change on Wisconsin corn and soybean yield trends. Environmental Research Letters. 2008;3(3):034003.
View Article
Google Scholar

[46] View Article

[47] Google Scholar

[ref15] 15. Tao F, Yokozawa M, Liu J, Zhang Z, et al. Climate-crop yield relationships at provincial scales in China and the impacts of recent climate trends. Climate Research. 2008;38(1):83–94.
View Article
Google Scholar

[49] View Article

[50] Google Scholar

[ref16] 16. Schlenker W, Lobell DB. Robust negative impacts of climate change on African agriculture. Environmental Research Letters. 2010;5(1):014010.
View Article
Google Scholar

[52] View Article

[53] Google Scholar

[ref17] 17. Burke M, Hsiang SM, Miguel E. Global non-linear effect of temperature on economic production. Nature. 2015;. pmid:26503051
View Article
PubMed/NCBI
Google Scholar

[55] View Article

[56] PubMed/NCBI

[57] Google Scholar

[ref18] 18. Wheeler TR, Craufurd PQ, Ellis RH, Porter JR, Prasad PV. Temperature variability and the yield of annual crops. Agriculture, Ecosystems & Environment. 2000;82(1):159–167.
View Article
Google Scholar

[59] View Article

[60] Google Scholar

[ref19] 19. Moriondo M, Giannakopoulos C, Bindi M. Climate change impact assessment: the role of climate extremes in crop yield simulation. Climatic Change. 2011;104(3-4):679–701.
View Article
Google Scholar

[62] View Article

[63] Google Scholar

[ref20] 20. Lesk C, Rowhani P, Ramankutty N. Influence of extreme weather disasters on global crop production. Nature. 2016;529(7584):84–87. pmid:26738594
View Article
PubMed/NCBI
Google Scholar

[65] View Article

[66] PubMed/NCBI

[67] Google Scholar

[ref21] 21. Schewe J, Gosling SN, Reyer C, Zhao F, Ciais P, Elliott J, et al. State-of-the-art global models underestimate impacts from climate extremes. Nature communications. 2019;10(1):1–14. pmid:30824763
View Article
PubMed/NCBI
Google Scholar

[69] View Article

[70] PubMed/NCBI

[71] Google Scholar

[ref22] 22. Zampieri M, Ceglar A, Dentener F, Dosio A, Naumann G, Van Den Berg M, et al. When will current climate extremes affecting maize production become the norm? Earth’s Future. 2019;7(2):113–122.
View Article
Google Scholar

[73] View Article

[74] Google Scholar

[ref23] 23. Ciais P, Reichstein M, Viovy N, Granier A, Ogée J, Allard V, et al. Europe-wide reduction in primary productivity caused by the heat and drought in 2003. Nature. 2005;437(7058):529–533. pmid:16177786
View Article
PubMed/NCBI
Google Scholar

[76] View Article

[77] PubMed/NCBI

[78] Google Scholar

[ref24] 24. Schlenker W, Roberts MJ. Nonlinear effects of weather on corn yields. Applied Economic Perspectives and Policy. 2006;28(3):391–398.
View Article
Google Scholar

[80] View Article

[81] Google Scholar

[ref25] 25. Semenov MA. Development of high-resolution UKCIP02-based climate change scenarios in the UK. Agricultural and Forest Meteorology. 2007;144(1):127–138.
View Article
Google Scholar

[83] View Article

[84] Google Scholar

[ref26] 26. van der Velde M, Tubiello FN, Vrieling A, Bouraoui F. Impacts of extreme weather on wheat and maize in France: evaluating regional crop simulations against observed data. Climatic change. 2012;113(3-4):751–765.
View Article
Google Scholar

[86] View Article

[87] Google Scholar

[ref27] 27. Porter JR, Semenov MA. Crop responses to climatic variation. Philosophical Transactions of the Royal Society B: Biological Sciences. 2005;360(1463):2021–2035. pmid:16433091
View Article
PubMed/NCBI
Google Scholar

[89] View Article

[90] PubMed/NCBI

[91] Google Scholar

[ref28] 28. Deryng D, Conway D, Ramankutty N, Price J, Warren R. Global crop yield response to extreme heat stress under multiple climate change futures. Environmental Research Letters. 2014;9(3):034011.
View Article
Google Scholar

[93] View Article

[94] Google Scholar

[ref29] 29. Rosenzweig C, Iglesias A, Yang X, Epstein PR, Chivian E. Climate change and extreme weather events; implications for food production, plant diseases, and pests. Global change & human health. 2001;2(2):90–104.
View Article
Google Scholar

[96] View Article

[97] Google Scholar

[ref30] 30. Fontes F, Gorst A, Palmer C. Threshold effects of extreme weather events on cereal yields in India. Climatic Change. 2021;165(1):1–20.
View Article
Google Scholar

[99] View Article

[100] Google Scholar

[ref31] 31. Taleb NN. The black swan: The impact of the highly improbable fragility. vol. 2. Random House; 2010.

[ref32] 32. UN. World Economic Situation and Prospects 2014; 2014.

[ref33] 33. Sacks WJ, Deryng D, Foley JA, Ramankutty N. Crop planting dates: an analysis of global patterns. Global Ecology and Biogeography. 2010;19(5):607–620.
View Article
Google Scholar

[104] View Article

[105] Google Scholar

[ref34] 34. Monfreda C, Ramankutty N, Foley JA. Farming the planet: 2. Geographic distribution of crop areas, yields, physiological types, and net primary production in the year 2000. Global biogeochemical cycles. 2008;22(1).
View Article
Google Scholar

[107] View Article

[108] Google Scholar

[ref35] 35. Mitchell TD, Jones PD. An improved method of constructing a database of monthly climate observations and associated high-resolution grids. International Journal of Climatology. 2005;25(6):693–712.
View Article
Google Scholar

[110] View Article

[111] Google Scholar

[ref36] 36. Coles S, Bawa J, Trenner L, Dorazio P. An introduction to statistical modeling of extreme values. vol. 208. Springer; 2001.

[ref37] 37. Keef C, Papastathopoulos I, Tawn JA. Estimation of the conditional distribution of a multivariate variable given that one of its components is large: Additional constraints for the Heffernan and Tawn model. Journal of Multivariate Analysis. 2013;115:396–404.
View Article
Google Scholar

[114] View Article

[115] Google Scholar

[ref38] 38. Southworth J Harry; Heffernan. texmex: Statistical modelling of extreme values. R package Version 21. 2013;.

[ref39] 39. Stephenson A. evd: Functions for Extreme Value Distributions. R package Version 23. 2015;.

[ref40] 40. Wickham W Hadley; Chang. ggplot2: An Implementation of the Grammar of Graphics. R package Version 20. 2015;.

[ref41] 41. South A. Mapping Global Data. R package Version 13-6. 2016;.

[ref42] 42. Southworth H, Heffernan JE, Southworth MH. Package texmex. Statistics. 2013;.

[ref43] 43. Marmai N, Franco Villoria M, Guerzoni M. How the Black Swan damages the harvest: statistical modelling of extreme events in weather and crop production in Africa, Asia, and Latin America. Department of Economics and Statistics Cognetti de Martiis Working Paper Series University of Turin. 2016;.

[ref44] 44. Taylor J. Overview: Importance of sorghum in Africa. In: Proceedings of AFRIPRO Workshop on the Proteins of Sorghum And Millets: Enhancing Nutritional and Functional Properties for Africa, Pretoria, South Africa. vol. 9; 2003.

[ref45] 45. Maclean J, Hardy B, Hettel G. Rice Almanac: Source Book for One of the Most Important Economic Activities on Earth. IRRI; 2013.

[ref46] 46. Galindo LM, De Miguel CJ. Economics of climate change in Latin America and the Caribbean: summary. Agriculture Markets and Crop profiles. 2010;.
View Article
Google Scholar

[125] View Article

[126] Google Scholar

[ref47] 47. Wani SP, Rockström J, Oweis TY, et al. Rainfed agriculture: unlocking the potential. vol. 7. CABI; 2009.

[ref48] 48. Lobell DB, Burke MB. Why are agricultural impacts of climate change so uncertain? The importance of temperature relative to precipitation. Environmental Research Letters. 2008;3(3):034007.
View Article
Google Scholar

[129] View Article

[130] Google Scholar

[ref49] 49. Lobell DB, Cassman KG, Field CB. Crop yield gaps: their importance, magnitudes, and causes. Annual Review of Environment and Resources. 2009;34(1):179.
View Article
Google Scholar

[132] View Article

[133] Google Scholar

[ref50] 50. Porter JR, Gawith M. Temperatures and the growth and development of wheat: a review. European Journal of Agronomy. 1999;10(1):23–36.
View Article
Google Scholar

[135] View Article

[136] Google Scholar

[ref51] 51. Easterling DR, Evans J, Groisman PY, Karl T, Kunkel KE, Ambenje P. Observed variability and trends in extreme climate events: a brief review*. Bulletin of the American Meteorological Society. 2000;81(3):417–425.
View Article
Google Scholar

[138] View Article

[139] Google Scholar

Figures

Abstract

Introduction

Materials and methods

Data

Methods

Results

Empirical analysis

Extrapolation

Discussion and conclusion

Supporting information

S1 Annex.

References