Movatterモバイル変換

[0]ホーム
URL:

Skip to main content
NCBI home page
Search in PMCSearch
  • View on publisher site icon
As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health.
Learn more:PMC Disclaimer | PMC Copyright Notice
UKPMC Funders Author Manuscripts logo
. Author manuscript; available in PMC: 2020 Mar 2.

Climate change impacts on banana yields around the world

Varun Varma1,Daniel P Bebber1,*
1Department of Biosciences, University of Exeter, Exeter EX4 4QJ, UK
*

Corresponding author: Daniel P. Bebber,d.bebber@exeter.ac.uk, tel. +44 1392 725851.

Issue date 2019 Oct.

Users may view, print, copy, and download text and data-mine the content in such documents, for the purposes of academic research, subject always to the full Conditions of use:http://www.nature.com/authors/editorial_policies/license.html#terms

PMCID: PMC6774771  EMSID: EMS83798  PMID:31579401
The publisher's version of this article is available atNat Clim Chang

Abstract

Nutritional diversity is a key element of food security1–3. However, research on the effects of climate change on food security has, thus far, focussed on the major food grains4–8, while the response of other crops, particularly those that play an important role in the developing world, are poorly understood. Bananas are a staple food and a major export commodity for many tropical nations9. Here we show that for 27 countries – accounting for 86% of global dessert banana production – a changing climate since 1961 has increased yields by an average of 1.37 T.ha-1. Past gains have been largely ubiquitous across the countries assessed and African producers will continue to see yield increases into the future. However, global yield gains could be dampened or disappear in the future, reducing to 0.59 T.ha-1and 0.19 T.ha-1by 2050 under the RCP 4.5 and 8.5 climate scenarios, respectively, driven by declining yields amongst the largest producers and exporters. By quantifying climate-driven and technology-driven influences on yield, we also identify countries at risk from climate change and those capable of mitigating its effects, or capitalising on its benefits.

Bananas are widely cultivated in tropical and sub-tropical regions around the world, where they can provide a substantial proportion of affordable calories, dietary diversity and income9–11. Bananas are also ubiquitous in their availability in non-producing regions through international trade, which accounts for 15% of global production12. This international trade supplements nutritional diversity in non-producing countries, while making a large contribution to local and national economies in producing countries. For example, bananas and their derived products constitute the second largest agricultural export commodity of Ecuador and Costa Rica13. Globally, bananas (together with plantains) are amongst the top ten crops in terms of area of cultivation, yield and calories produced10. Given the importance of this crop for subsistence and trade, it is surprising how poorly represented bananas are in global assessments of climate change impacts on food and nutritional security4–6.

Quantifying the optimal climatic conditions for banana productivity is central to assessing the crop’s climate sensitivity, and thereafter, predicting the potential impacts of climate change on banana production systems. Ideally, this requires the collation of data from experiments and field trials conducted over a range of environmental conditions, including sub-optimal combinations. While few such experiments have been conducted14–19, some of which constitute the core of most banana production models currently used (e.g. Global Agro-Ecological Zones; GAEZ), major challenges remain. Firstly, the small number of published studies, coupled with small sample sizes within them and the limited breadth of environmental conditions assessed, are inadequate to derive generalisable estimates of optimal conditions for a crop so widely cultivated across the world. Second, estimates of productivity-climate relationship parameters have not been rigorously validated against large quantities of observed production data. Consequently, the representation of bananas in existing crop models is likely to be based on abstractions derived from shared plant characteristics1, which may not accurately predict effects of climate change on productivity.

Here we assess the climate sensitivity of global dessert banana (banana, hereafter) productivity or yield using a combination of national and sub-national production datasets from 27 countries (Table 1) spanning varying time periods (Supplementary table S1), coupled with previously published expert information on banana physiology. In all, the data used in our analyses account for approximately 86% of the world’s banana production and covers 80% of global area under cultivation. The selected countries include the world’s largest and regionally important producers, as well as the largest exporters of bananas, e.g. Ecuador, Colombia, Costa Rica, Ivory Coast, Philippines, etc.20. This large geographically stratified set of nations is exposed to diverse climatic conditions – ideally suited for climate sensitivity assessments. We statistically fitted observed yield data from the 1281 geographic units over multiple years to elevation corrected mean annual temperature and total annual precipitation using a beta function21. This allowed us to empirically identify the optimum climate space for banana productivity, and develop a climate-driven relative yield coefficient model for bananas. Model fitting involved partitioning the observed data into six regional subsets (Table 1), resulting in six regional models and a single global model. Models were constructed at these two scales to assess the validity of using a single global model for bananas, given that regions and countries can vary widely in cultivation practices and cultivars of bananas grown. Thereafter, we employed the regional models to quantify historical climate change effects on productivity, as well as the contribution of change in cultivation efficiency over time (technology trend), and project the future impacts of climate change.

Table 1.

Banana producing countries (grouped by regions) used in this analysis. Country code is the abbreviation by which countries are referred to in figures associated with our analyses. Values for area harvested and production are for 2016 (ref.21). The Exp:Prod variable represents the proportion of production that is exported. For regional names: LAC – Latin America and the Caribbean, SEAA – South East Asia + Australia.

RegionCountryCountry codeArea harvested (ha)Production (T)Exp:Prod
AfricaAngolaAO1314553858066<0.001
BurundiBI195248911193<0.001
CameroonCM7235911875470.25
D.R. CongoCD834133110870
EthiopiaET632135383020.02
GuineaGN45459212874<0.001
Ivory CoastCI73553309460.98
KenyaKE632991288588<0.001
RwandaRW32200930379620
TanzaniaTZ46847035596390.005
BrazilBrazilBR46971167643240.009
ChinaChinaCN43004613324337<0.001
IndiaIndiaIN846000291240000.004
LACBelizeBZ247270619>0.99
ColombiaCO8463720436680.9
Costa RicaCR4241024095430.98
Dominican RepublicDO2683410797810.36
EcuadorEC18033765296760.91
GuatemalaGT7820637751500.57
HondurasHN244277071200.93
MexicoMX7832223847780.19
NicaraguaNI16801064370.86
PanamaPA64552588910.96
SEAAAustraliaAU16612354241<0.001
IndonesiaID13996470071250.001
MalaysiaMY280363095080.08
PhilippinesPH45664158291420.24

In our global model, optimum mean annual temperature for banana productivity was estimated at 26.7°C (95% confidence interval = ± 0.04°C;Fig 1;Supplementary table S2), which is very similar to the commonly used optimum temperature of 27°C10,22,23. However, the optimum temperatures for regional models varied considerably, ranging from 20.1°C (95% CI = ± 0.1°C) for Brazil, to 30.4°C (95% CI = ± 0.1°C) for Africa (Fig 1;Supplementary table S2). Optimum total annual rainfall in our global model was estimated at 1673 mm (± 13 mm), which falls within the range previously reported23 (900 mm – 1700 mm). But again, regional models varied substantially (Fig 1;Supplementary table S2), with India showing the lowest optimum rainfall of 327 mm.y-1 (95% CI = ± 16 mm.y-1) and China requiring the highest (2924 mm.y-1; 95% CI = ± 27 mm.y-1). Hence, relying on a single global model to understand the climate sensitivity of banana productivity is likely to result in considerable error at regional, national and sub-national scales.

Figure 1. Climate-yield model parameter estimates for banana cultivation.

Figure 1

Cardinal values for mean annual temperature and total annual rainfall estimated using beta functions for banana yields. Optimum temperature and rainfall values are represented by think black vertical lines. Estimates are presented for each of the six regional model and a single global banana climate-yield model. For clarity, 95% confidence limits around the estimated parameters are presented insupplementary table S2. Curves fitted to observed data are presented inSupplementary figures S1-S7. (*) SeeMethods for notes on interpreting results for Brazil.

Our hindcast analysis, which utilised the regional climate-yield models, suggests that climate change over the recent past (1961 to 2016) has had a net benefit on global banana yields (Fig 2a), which have increased at a rate of 0.024 T.ha-1.y-1 (95% CI = ± 0.006 T.ha-1.y-1). Over the 56 years of the hindcast assessment this translates to an average global yield increase of 1.37 T.ha-1 (95% CI = ± 0.33 T.ha-1). Of the 27 countries included in our analyses, 21 showed a positive effect of recent climate change on banana yields, two (Kenya and Colombia) showed no effect, and four (Brazil, Indonesia, Malaysia and Philippines) showed climate-driven yield declines (Fig 2a;Supplementary table S3). On aggregating national yield trends to regional averages, we find that Africa, China, India, as well as Latin America and the Caribbean (LAC) show positive effects of climate change on banana yields, while Brazil, and south-east Asia and Australia (SEAA) have been negatively affected (Supplementary figure S8a; Supplementary table S3). Changes in yield appear to be primarily driven by consistent increases in temperature over the recent past (Supplementary figure S9; Supplementary table S4). Countries where warming has resulted in banana growing areas experiencing more optimal temperatures have seen productivity increases, while countries where temperatures have exceeded the regional optimum, show declines. However, we note that the inclusion of irrigated production in our analyses may obscure the influence of change in precipitation on yields.

Figure 2. Effects of past and future climate change on banana yields.

Figure 2

Modelled contribution of climate change between 1961 and 2016 on banana yields in major producing countries from a hindcast analysis (a). b,c Predicted changes in banana yields by 2050 relative to yields modelled using long-term average climatic conditions (1970-2000) under RCP 4.5 (b) and RCP 8.5 (c) climate scenarios. Black horizontal lines and grey areas in each panel represent the global averaged change in yield due to climate change, and associated 95% confidence bounds. Error bars represent 95% confidence intervals for each country. Each bar is associated with a country name abbreviation (seetable 1) and colour coded by region.

In countries such as Brazil and Malaysia, modelled climate-driven yield declines (Fig 2a) are also reflected in observed country level yield declines (Supplementary figure S10). However, large increases in observed yields in Indonesia and Philippines (Supplementary figure S10) run counter to modelled climate-driven yield declines. This could be attributed to the large positive effect of changes in cultivation efficiency or a positive technology-yield trend (an aggregate term that captures changes in cultivation practices, inputs, land management and investment in infrastructure, such as irrigation, etc.), which overwhelms the relatively smaller negative effect of climate change (Fig 3). Technology-yield trends were positive for most countries considered here, and of magnitudes much greater than climate-yield trends, thus enhancing observed country-scale yields. However, in some cases a strong negative technology-yield trend completely counteracted, and reversed yield gains due to climate change. For example, countries such as Cameroon, Ethiopia and Panama show a positive effect of climate change on yields, but a strong negative technology trend (Fig 3), resulting in an overall decline in observed yields over time (Supplementary figure S10). Hence, the capacity to capitalise on the benefits of increasingly suitable climatic conditions for banana cultivation, or mitigating against future change is strongly dependent on how countries invest in maintaining and improving their cultivation efficiency.

Figure 3. Effect of changes in climate and cultivation efficiency (technology) on banana yields (1961 to 2016).

Figure 3

Error bars represent 95% confidence intervals. Countries are labelled using a two letter abbreviation (seetable 1) and grouped by region. Note – Brazil, China and India are considered to be stand-alone regions.

Regional model based forecasting revealed that by 2050, past positive effects of climate change on average global banana yields, though likely to continue, will be of lower magnitude. Yield increases could decline to 0.59 T.ha-1 (95% CI = ± 1.38 T.ha-1) and 0.19 T.ha-1 (95% CI = ± 1.86 T.ha-1) under the RCP 4.5 and more extreme RCP 8.5 climate scenarios, respectively, relative to yields modelled using long-term climate averages for 1970-2000 (Fig 2b and 2c;Supplementary tables S5 and S6). Unlike the hindcast analysis, where only four countries in our assessment showed a negative effect of past climate change on yield, negative responses could be more widespread amongst countries in the future. Ten countries are predicted to show at least a negative trend, if not strong declines in yields (RCP 4.5 scenario). Importantly, these include India (the world’s largest producer and consumer of bananas), Brazil (fourth largest producer), as well as Colombia, Costa Rica, Guatemala, Panama and Philippines, all of which are major exporters. Some countries could continue to see benefits, or indeed increased benefits, of climate change in the future. These include all 10 African countries in our assessment, as well as Ecuador (the world’s largest exporter) and Honduras (also a major exporter). When aggregating future yield trends regionally, Africa unsurprisingly emerges as a key winner, while the strong positive effects of past climate change in the LAC countries declines to a positive trend. Similarly, China may not see any benefits of climate change in the future, as it did in the recent past. India could experience a major reversal with predicted negative effects of future climate change compared to positive effects in the past. Lastly, both Brazil and SEAA countries will continue along a negative trajectory into the future (Supplementary figures S8b and S8c; Supplementary tables S5 and S6).

Combining forecasted climate-driven changes in yield with the technology-yield trend estimates – which we assume to represent a country’s capacity to adapt to production risks in the future – we qualitatively classified the climate risk to banana production in each of the 27 countries included in this study (Fig 4;Supplementary figure S11). Countries where forecasted climate-driven yield changes were negative, and that had negative or flat technology-yield trends in the past, were classified as ‘at risk’. These included Malaysia, Panama, Nicaragua, the world’s fourth largest producer – Brazil, as well as Colombia – a major exporter. The two largest producers, India and China, along with many LAC countries that are important exporters, as well as Australia, Indonesia and Philippines were classified as ‘adaptable’. These countries showed potential negative effects of climate change on yields, but strong positive technology-yield trends that may mitigate climate-driven yield declines. Amongst the countries classified as at an ‘advantage’ - where forecasted changes in yield are strongly positive – are some of the largest current exporters (Ecuador and Honduras), and all 10 African countries that were assessed. However, it is important to note here, that our classification of risk is climate centric. Hence, realising the climate-driven advantage to banana productivity in many of the African countries will also be contingent on reversing negative technology-yield trends, e.g. by improving cultivation practices, investing in infrastructure, etc., in the future. If cultivation efficiency in the African nations can be improved, it could bolster local and regional nutritional security. In addition, it could also modify the existing configuration of the banana export market, especially given the negative impacts expected in some of the major current exporters.

Figure 4. Future climate risk assessment for major banana producing countries (by 2050).

Figure 4

This categorisation was carried out by combining changes in predicted yield under the RCP 8.5 climate change scenario and the effect of cultivation efficiency (technology trend) on past yields. Countries classified as ‘at risk’ are those where yields are predicted to decline due to climate change and that have shown a negative technology trend on past yields. ‘Adaptable’ countries could see future climate-driven yield declines, but mitigation could be possible given past positive technology trends. Countries at an ‘advantage’ are those predicted to experience climate-driven increases in yield by 2050. Countries are grouped by region. Note – Brazil, China and India are considered to be stand-alone regions.

In summary, our study quantified region specific climate-yield relationships for banana cultivation that suggests that climate change in the recent past has been beneficial to global banana productivity, but will be less so in the near future. However, we note that our analysis is based on average climatic conditions and does not account for other climate change driven threats of increased frequencies of extreme events24, as well as the risk posed by established and emerging diseases25–27. In addition, our climate-yield models are based on observed production data rather than experimental assays. Hence, model fits are likely to be influenced by agro-economic factors in addition to banana plant physiology, and therefore, model interpretation requires caution (seeMethods). Previous studies have largely assessed climate driven changes in the extent of land suitable for banana cultivation10,22,23,28,29, without considering the potential for competition with other staple crops and land-use types30. In contrast, analyses here focused on the more practical quantity of yield changes where bananas are already being grown. In addition, we assessed the climate risk to major producer and exporter countries. We infer that future climate risks to banana production could largely be mitigated to secure local nutritional diversity and security. Nevertheless, securing supply to non-producing countries, where banana consumption is an important contributor to dietary diversity, is likely to require a reorganisation of the export market.

Methods

Data sources

We used a time series of annual banana production data from 27 countries (Table 1), which included all the major producer and exporter countries. Our analyses focus on dessert bananas and we exclude data on plantain production. Consequently, we did not include Uganda in this study (despite having the highest per capita banana consumption) because the East African Highland Banana which makes up the majority of Ugandan production, is (a) classified as plantain in production data available from the FAO, and (b) is cultivated at higher elevations (1400 – 2000m) than other production systems considered in this study. Data sources, spatial resolution and time period over which production data were available differed between countries (seesupplementary table S1). Hereafter, we refer to the finest administrative scale at which data were available within each country as a geographic unit (GU). For example, production data were available at the whole country scale for countries such as Angola and Malaysia. Hence, in these cases, ‘country’ was assigned as the GU. On the other hand, for countries such as India and China, data were available at the district and province scale, respectively. Hence, individual districts or provinces were assigned as a GU. Data were ‘cleaned’ to remove non-sense values (e.g. where yield values were unrealistically high, or where production was reported, but area under cultivation was zero) and other values which indicated poor data quality or reliability. For instance, national scale data available from FAOSTAT (typically from 1961 to 2016) can contain a combination of officially reported values, as well as computed values (presumably when official data are not available). In such cases, we subsetted the time series of production data, such that the first data point corresponded to the first officially reported values. In the case of data from Brazil, there were many cases where data from the same GU in successive years were identical, suggesting that data were in fact not collected annually. In these cases, where three or more successive identical production and area harvested values were encountered, only the first was retained. Production and area under cultivation was used to calculate yield (T.ha-1) – the response variable of interest – for each GU-year combination in the dataset.

Climate data from the CRU TS 4.01 product31,32 was used to model climate sensitivity of banana yields, conduct the hindcast analysis and identify the contribution of change in cultivation efficiency (technology trend) on banana productivity over time. Mean annual temperature and total annual precipitation were extracted from the CRU dataset and assigned to the respective GU-year combinations in the production data. The temperature data was corrected for elevation. This was done to account for the lack of a good quality distribution map of banana growing areas, as well as the coarse resolution of the CRU dataset, which could result in individual pixels representing average temperatures over areas of high elevation, where bananas are less likely to grow. In other words, the relatively large area (approximately 360 km2) covered by a single CRU pixel could encompass areas of low and high elevation, and hence, uncorrected temperature values within a pixel represent the average temperature experienced across these elevations. Since bananas are less likely to grow at higher elevations, these uncorrected temperature values do not represent the temperatures experienced in banana plantations. The elevation based temperature correction was conducted by overlaying the 90 m resolution Shuttle Radar Telemetry Mission (SRTM) digital elevation model (DEM) with the CRU temperature dataset. A lapse rate of -0.0065 °C.m-1 was used to recalculate temperatures within each CRU pixel at the resolution of the DEM. Recalculated temperature values in DEM pixels where elevation was greater than 2000 m were eliminated (we assumed that there is a very low probability that bananas grow at elevations > 2000 m). For the remaining DEM pixels the elevation weighted average temperature was calculated. Weights were calculated using a logistic function 1/{1 + e[0.005(elevation – 1000)]}. The logistic function used assigns weights that tend to one for elevations < 500 m above sea level, declines in a near linear fashion from 500 m to 1500 m, with an inflection point at 1000 m (i.e. weight of 0.5 at 1000 m elevation), and tends to zero for elevations > 1500 m.

In addition to the elevation correction of temperature, climate extraction for Mexico and Australia (both with country scale production data) was restricted to administrative units where banana cultivation had previously been reported in published literature10,22,23. This was done to avoid climate data from extremely arid environments (where bananas are unlikely to be cultivated, and which would not be accounted for with an elevation based temperature correction) influencing the analysis. For all 10 African countries in this study (country scale production data), no usable published information of banana growing areas could be used to inform the data extraction. Therefore, modelled banana cultivation areas33 were used to restrict climate data extraction.

For future climate, mean annual temperature and total annual precipitation for 2050 was extracted from WorldClim CMIP5 downscaled projections34,35 (5-minute resolution bioclimatic variables) of 19 and 17 GCMs for the RCP 4.5 and RCP 8.5 climate scenarios, respectively. Change in banana yields under future climate scenarios (forecast analysis) was calculated relative to yields modelled using the long term (1970 to 2000) climate averages extracted from WorldClim 2.036,37. Mean annual temperature data from WorldClim 2.0 and CMIP5 downscaled projections were also elevation corrected (as had been done with the CRU TS 4.01 temperature data).

Climate-yield relationship estimation

Our objective was to statistically fit observed annual banana yield data to prevailing mean temperature and total annual precipitation. The resulting relationship would be used for further analyses. We expected the relationship along both climate axes to be represented by a bell-shaped or unimodal function, which was not necessarily symmetrical around its peak (i.e. around the optimum). A beta function38 has been suggested as a suitable candidate model for such a non-linear process39. Here we implemented a modified version of the beta function22 that has also been used in past banana physiology studies14,17. The beta function along the temperature and precipitation variables are as follows:

Rt=(Tmax−TobsTmax−Topt)(Tobs−TminTopt−Tmin)(Topt−Tmin)(Tmax−Topt)(1)
Rp=(Pmax−PobsPmax−Popt)(Pobs−PminPopt−Pmin)(Popt−Pmin)(Pmax−Popt)(2)

Where:

  • Rt = Yield coefficient for temperature

  • Tmax = Maximum annual average temperature, beyond which banana production stops

  • Tmin = Minimum annual average temperature, below which banana production stops

  • Topt = Optimum temperature for banana production

  • Tobs = Observed annual average temperature

  • Rp = Yield coefficient for precipitation

  • Pmax = Maximum total annual precipitation, beyond which banana production stops

  • Pmin = Minimum total annual precipitation, below which banana production stops

  • Popt = Optimum total annual precipitation for banana production

  • Pobs = Observed total annual precipitation

As bothequations (1) and(2) result in a value between zero and one, a scaling coefficient (S) is also estimated while fitting the yield data to the climate variables. The scaling coefficient represents the maximum average yield under optimum conditions of temperature and precipitation. As such the function being fit is given by:

Yield=S.Rt.Rp(3)

The yield and climate data were partitioned in to six regional sub-sets (seeTable 1), which were then fit toequation (3). Regional subsets were created to account for differences in cultivation practices and cultivars grown (cultivar identity was not explicitly included in our analyses as available production data does not differentiate between cultivars or varieties). Fitting models to regional, rather than country-level data subsets was also done to increase resolving power of the data, especially along the temperature axis. This attention to resolving power is important as tropical countries, in particular the small and/or island nations, can show a very limited temperature ranges in space and time, resulting in spurious non-linear model fits with unrealistic parameter estimates. Hence, relatively small countries, or those for which we only had country-scale data for were grouped into regions, while large countries (covering a larger climate space) with sub-national data formed regions by themselves (e.g. India, China and Brazil). In addition, a single ‘global’ model was also fitted. Here, regions were equally weighted, to account for differences in regional sample sizes.

Data were first fitted toequation (3) by brute force (107 iterations), such that residual sums of squares were minimised. The observed data occupied a restricted range along the temperature axis. This range logically represents the temperature space where commercial cultivation of bananas is viable, but not necessarily the physiological limits of the banana plant. Hence, estimation of Tmin and Tmax inequation (1) needed to be informed by expert opinion or previous estimates10,22,23. During the brute force fitting procedure, Tmin was constrained between 10°C and 20°C, while Tmax was constrained between 30°C and 35°C. Estimation of precipitation parameters inequation (2) were not similarly constrained as land management or cultivation practices could alleviate restrictions imposed by low precipitation (use of irrigation) or high precipitation (infrastructure to promote soil drainage). Hence, we assume that banana cultivation is feasible under sub-optimal/marginal precipitation conditions, subject to management intervention. As a consequence of production data including both rainfed and irrigated production systems, we treat estimated precipitation parameters with caution.

As parameters estimated forequation (3) by brute force may not represent a true optimised fit, estimated brute force parameters were used as starting values in a constrained non-linear least squares curve fitting function (implemented using theScipy optimize module in Python 3.6). Measures of variation around estimated parameters using the curve fitting function may be unreliable as estimated parameters often fell outside the bounds of the data (e.g. bananas are unlikely to be cultivated at their physiological minimum or maximum temperatures). Hence, the non-linear curve fitting procedure was bootstrapped (100 iterations), where each iteration was fitted using a resampled dataset with replacement. These bootstrapped estimates were used for interpretation and further analyses.

Past contribution of climate change (hindcast) and cultivation efficiency (technology) to banana yields

Regional models fromequation (3) were used to calculate modelled yield for each CRU climate dataset pixel within the GUs of interest from 1961 to 2016. These modelled annual yields were averaged at a country-scale. Where sub-national production data were available, the average yield was weighted by the area under cultivation of GUs within the country. A generalised least squares (GLS) regression was then fitted to the modelled yield data over time for each country, using thegls function from thenlme package in R. The GLS regression allowed for a 1st order autoregressive correlation structure in the residuals, to account for the correlation over time in climate data. The parameters of the GLS regression represented the climate-driven trend in yields. Modelled annual yield data were then regionally averaged (weighted by area under cultivation within GUs). Regional climate-driven yield trends were then estimated using region specific GLS regressions.

To determine if observed yield trends were driven by changes in temperature or precipitation, the relative yield coefficients for temperature (Rt;equation 1) and precipitation (Rp;equation 2) were calculated for each pixel in each year from 1961 to 2016. As with the hindcast analysis above, Rt and Rp were aggregated to the scale of countries. A GLS regression was then fitted to Rt and Rp separately, to estimate the trends in Rt (temperature RYC trend) and Rp (precipitation RYC trend). Similarly, trends for changes in mean annual temperature and total precipitation were calculated. Country-scale temperature RYC trends and precipitation RYC trends were plotted against country-scale mean annual temperature and total precipitation trends (along with associated 95% confidence intervals) and visually inspected for consistent patterns.

Modelled annual country-scale yields were subtracted from annual yield data available from FAO20 for each of the 27 countries in the study. As modelled yields are solely climate-determined, the resulting difference in yields (tYield) represents the influence of ‘technology’ or cultivation efficiency on yield. We fitted a GLS regression to tYield over time and the parameter estimates of the regression represent the effect of a ‘technology trend’ on banana yields.

To evaluate model performance, we calculated the correlation coefficient and root mean squared error (RMSE) between (a) observed yields (country-scale) from the FAO over time and country-scale yields estimated using regional climate-yield models (hindcast yield values), and (b) observed yields (country-scale) from the FAO over time and hindcast yield values to which country-scale technology trends were added (Supplementary figure S30).

Future climate change impacts on banana yields (forecast) and climate-risk classification

The impact of future climate change on banana yields was quantified as the change in predicted yields by 2050 (future yields) relative to modelled yields given long-term average temperature and precipitation between 1970 and 2000 (current yield). Regional yield models fromequation (3) were used to calculate current yield for each WorldClim 2.0 pixel within the GUs of interest. Similarly, future yields were calculated using regional models and climate data from WorldClim CMIP5 downscaled projections for 2050. Future yields were predicted using downscaled data from 19 GCMs representing the RCP 4.5 scenario and 17 GCMs for the RCP 8.5 scenario. Differences between current and future yields were averaged across GCMs for each pixel, and pixel-scale yield differences were averaged to country-scale (averages weighted by area under cultivation of GUs within countries were used when sub-national production data were available). Regional average yield differences were then calculated, weighted by area under cultivation.

Climate risk for countries was classified by combining yield differences under the RCP 8.5 scenario and estimated technology trends. This was a climate centric classification. Hence, if a country was predicted to show increases in climate-driven yields, it was classified as being at an ‘advantage’, regardless of its past technology trend. If a country was predicted to show a negative effect or no effect of climate on yields, and a past positive technology trend, it was classified as being ‘adaptable’. Lastly, if a country was predicted to show negative effects of future climate on yields, and negative or flat past technology trend, it was classified as being ‘at risk’.

Caveats and unaccounted for sources of variation

The analysis presented has utilised the best and most comprehensive data sources available at the time of writing. However, we note that further improvements in available data quality would be beneficial to carry out an even more accurate and fine grained assessment. A few shortcomings of the data used include – varying spatial resolution of the datasets, non-uniform gaps in the production time series for different regions and countries within regions, lack of variety specific production data, quantities of agricultural inputs and use of irrigation for cultivation. Fertilisation effects of increased CO2 in the atmosphere could also have an effect on banana productivity, but has not been accounted for in our analyses. Consequently, a level of caution is required in interpreting results, especially where high variation in yield is observed across climatic gradients. For example, production data from Brazil showed a large variation across the range of mean annual temperatures and levels of total annual precipitation encompassed by the dataset. In addition, the best fit beta model for yield revealed the lowest optimum temperature (20.06°C) compared to the other regions assessed in this analysis. Such a result could be due to overlapping (and reinforcing) gradients of climatic conditions and economic indicators. Compared to the north, southern Brazil experiences cooler and more seasonal climatic conditions, and fares better in economic terms. Higher incomes in the south could facilitate greater productivity due to greater capacity for the use of agricultural inputs, and therefore, lower the estimated optimum temperature in our fitted model. Hence, model interpretation should be carried out with caution and further detailed region-/country-specific research that incorporates socio-economic variables are a logical next step.

Secondly, in the absence of robust experimental data, we have statistically modelled climate-yield relationships using a top-down approach and observed production data. However, it is important to note that the distribution of banana producing areas are not solely a consequence of the banana plant's physiology. Agro-economic considerations, such as available cultivation infrastructure, transport links and access to markets also influence where bananas are grown. The production data incorporate these factors, and hence, our model fits cannot be interpreted as a purely physiological climate-yield relationship. For example, cultivation efficiency and yields can be substantially be improved in drier areas with irrigation. However, the extent of irrigation in use was not accounted for in our analyses, and low optimum precipitation estimates (e.g. India) should be interpreted with care.

Lastly, we also acknowledge that our analyses only consider the average annual climatic condition, and do not account for seasonal variation, nor the occurrence of extreme climatic events. Future region- and country-specific research would benefit from including these parameters, especially if more detailed production data are available to cope with increased model complexity.

Supplementary Material

1

Acknowledgements

The study was funded by Global Food Security grant no. BB/N020847/1 and EC Horizon 2020 project ID 727624. The funders had no role in the study design or execution. The authors thank Fiona Savory for feedback on the manuscript. Base maps were created using administrative region polygons from GADM version 2.8 (https://gadm.org/).

Footnotes

Competing financial interests

The authors declare no competing financial interests.

Author Contributions

DB designed the study. VV collated and analysed the data in discussion with DB. VV wrote the paper.

Data Availability Statement

All data used are publicly available and open access. All banana production data sources are listed inSupplementary Table S1. All climatic and topographic data sources are listed in Methods.

Code availability statement

No custom code was used in the analysis.

References

  • 1.Wheeler T, von Braun J. Climate change impacts on global food security. Science. 2013;341:508–513. doi: 10.1126/science.1239402. [DOI] [PubMed] [Google Scholar]
  • 2.Springmann M, et al. Global and regional health effects of future food production under climate change: a modelling study. The Lancet. 2016;387:1937–1946. doi: 10.1016/S0140-6736(15)01156-3. [DOI] [PubMed] [Google Scholar]
  • 3.Hwalla N, Labban SE, Bahn RA. Nutrition security is an integral component of food security. Front Life Sci. 2016;9:167–172. [Google Scholar]
  • 4.Welch JR, et al. Rice yields in tropical/subtropical Asia exhibit large but opposing sensitivities to minimum and maximum temperatures. Proc Natl Acad Sci. 2010;107:14562–14567. doi: 10.1073/pnas.1001222107. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Knox J, Hess T, Daccache A, Wheeler T. Climate change impacts on crop productivity in Africa and South Asia. Environ Res Lett. 2012;7 034032. [Google Scholar]
  • 6.Challinor AJ, et al. A meta-analysis of crop yield under climate change and adaptation. Nat Clim Change. 2014;4:287–291. [Google Scholar]
  • 7.Lobell DB, Gourdji SM. The influence of climate change on global crop productivity. Plant Physiol. 2012;160:1686–1697. doi: 10.1104/pp.112.208298. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Rosenzweig C, et al. Assessing agricultural risks of climate change in the 21st century in a global gridded crop model intercomparison. Proc Natl Acad Sci. 2014;111:3268–3273. doi: 10.1073/pnas.1222463110. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Heslop-Harrison JS, Schwarzacher T. Domestication, genomics and the future for banana. Ann Bot. 2007;100:1073–1084. doi: 10.1093/aob/mcm191. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.German Calberto G, Staver C, Siles P. An assessment of global banana production and suitability under climate change scenarios. In: Albehri Aziz., editor. Climate change and food systems: global assessments and implications for food security and trade. Food Agriculture Organisation of the United Nations (FAO); 2015. pp. 266–291. [Google Scholar]
  • 11.Vuylsteke D, Ortiz R, Ferris S. Genetic and agronomic improvement for sustainable production of plantain and banana in sub-Saharan Africa. Afr Crop Sci J. 1993;1 [Google Scholar]
  • 12.Turner DW, Fortescue JA, Thomas DS. Environmental physiology of the bananas (Musa spp.) Braz J Plant Physiol. 2007;19:463–484. [Google Scholar]
  • 13.World Bank. Available at:http://databank.worldbank.org
  • 14.Turner D, Lahav E. The Growth of Banana Plants in Relation to Temperature. Aust J Plant Physiol. 1983;10:43. [Google Scholar]
  • 15.Kallarackal J, Milburn J, Baker D. Water relations of the banana. III. Effects of controlled water stress on water potential, transpiration, photosynthesis and leaf growth. Aust J Plant Physiol. 1990;17:79. [Google Scholar]
  • 16.Eckstein K, Robinson JC. Physiological responses of banana (Musa AAA; Cavendish sub-group) in the subtropics. II. Influence of climatic conditions on seasonal and diurnal variations in gas exchange of banana leaves. J Hortic Sci. 1995;70:157–167. [Google Scholar]
  • 17.Thomas DS, Turner DW, Eamus D. Independent effects of the environment on the leaf gas exchange of three banana (Musa sp.) cultivars of different genomic constitution. Sci Hortic. 1998;75:41–57. [Google Scholar]
  • 18.van Asten PJA, Fermont AM, Taulya G. Drought is a major yield loss factor for rainfed East African highland banana. Agric Water Manag. 2011;98:541–552. [Google Scholar]
  • 19.Eckstein K, Robinson JC. Physiological responses of banana (Musa) AAA; Cavendish sub-group) in the subtropics. I. Influence of internal plant factors on gas exchange of banana leaves. J Hortic Sci. 1995;70:147–156. [Google Scholar]
  • 20.FAOSTAT. [Accessed: 3rd October 2017];2017 Available at:http://www.fao.org/faostat/en/#data/QC.
  • 21.Yan W, Hunt LA. An equation for modelling the temperature response of plants using only the cardinal temperatures. Ann Bot. 1999;84:607–614. [Google Scholar]
  • 22.Ramirez J, Jarvis A, Van den Bergh I, Staver C, Turner D. Crop adaptation to climate change. Wiley-Blackwell; Oxford, UK: 2011. Changing climates: effects on growing conditions for banana and plantain (Musa spp.) and possible responses; pp. 426–438. [Google Scholar]
  • 23.Van den Bergh I, et al. Climate change in the subtropics: the impacts of projected averages and variability on banana productivity. Acta Hortic. 2012:89–99. doi: 10.17660/ActaHortic.2012.928.9. [DOI] [Google Scholar]
  • 24.IPCC. Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. IPCC; 2014. [Google Scholar]
  • 25.Ordonez N, et al. Worse comes to worst: bananas and panama disease—when plant and pathogen clones meet. PLOS Pathog. 2015;11:e1005197. doi: 10.1371/journal.ppat.1005197. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Ploetz RC, Kema GHJ, Ma L-J. Impact of diseases on export and smallholder production of banana. Annu Rev Phytopathol. 2015;53:269–288. doi: 10.1146/annurev-phyto-080614-120305. [DOI] [PubMed] [Google Scholar]
  • 27.Bebber DP. Range-expanding pests and pathogens in a warming world. Annu Rev Phytopathol. 2015;53:335–356. doi: 10.1146/annurev-phyto-080614-120207. [DOI] [PubMed] [Google Scholar]
  • 28.Machovina B, Feeley KJ. Climate change driven shifts in the extent and location of areas suitable for export banana production. Ecol Econ. 2013;95:83–95. [Google Scholar]
  • 29.Sabiiti G, et al. Adapting agriculture to climate change: Suitability of banana crop production to future climate change over Uganda. In: Leal Filho W, Nalau J, editors. Limits to climate change adaptation. Springer International Publishing; 2018. pp. 175–190. [DOI] [Google Scholar]
  • 30.Foley JA, et al. Global consequences of land use. Science. 2005;309:570–574. doi: 10.1126/science.1111772. [DOI] [PubMed] [Google Scholar]
  • 31.Harris I, Jones Pd, Osborn Tj, Lister Dh. Updated high-resolution grids of monthly climatic observations – the CRU TS3.10 Dataset. Int J Climatol. 2014;34:623–642. [Google Scholar]
  • 32.CRU high-resolution gridded datasets. Available at:https://crudata.uea.ac.uk/cru/data/hrg/
  • 33.You L, et al. Spatial Production Allocation Model (SPAM) 2005 v3.2. 2018 Available at:http://mapspam.info.
  • 34.Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A. Very high resolution interpolated climate surfaces for global land areas. Int J Climatol. 2005;25:1965–1978. [Google Scholar]
  • 35.CMIP5 5-minutes. WorldClim - Global Climate Data. Available at:http://worldclim.org/CMIP5_5m.
  • 36.Fick SE, Hijmans RJ. WorldClim 2 new 1-km spatial resolution climate surfaces for global land areas. Int J Climatol. 2017;37:4302–4315. [Google Scholar]
  • 37.WorldClim Version2. WorldClim - Global Climate Data. Available at:http://worldclim.org/version2.
  • 38.Yin X, Kropff MJ, McLaren G, Visperas RM. A nonlinear model for crop development as a function of temperature. Agric For Meteorol. 1995;77:1–16. [Google Scholar]
  • 39.Archontoulis SV, Miguez FE. Nonlinear regression models and applications in agricultural research. Agron J. 2015;107:786–798. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1

Data Availability Statement

All data used are publicly available and open access. All banana production data sources are listed inSupplementary Table S1. All climatic and topographic data sources are listed in Methods.

Code availability statement

No custom code was used in the analysis.

ACTIONS

RESOURCES

[8]ページ先頭
©2009-2026 Movatter.jp