(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



Synchronous climate hazards pose an increasing challenge to global coffee production [1]

['Doug Richardson', 'Csiro Oceans', 'Atmosphere', 'Hobart', 'Tasmania', 'Jarrod Kath', 'Centre For Applied Climate Sciences', 'University Of Southern Queensland', 'Toowoomba', 'Queensland']

Date: 2023-03

Robusta regions and corresponding hazard definitions are to the right of the red line, plus northern Brazil. The map base layer is available from Natural Earth at https://www.naturalearthdata.com/downloads/110m-physical-vectors/110m-coastline/ .

The north of southern Brazil and eastern Ethiopia are the regions most susceptible to climate hazards, experiencing close to the maximum four hazards every year ( Fig 3 ). In general, everywhere experiences at least one hazard per year on average. Robusta regions appear to suffer from fewer hazards per year on average, though this may be due to the differences in hazard types. As we will show later, it is not uncommon for Arabica regions to experience VPD gr and T max,gr hazards concurrently, but it is rare for robusta regions to have too-cold minimum temperatures in the flowering season followed by too-warm minimum temperatures in the growing season (i.e. T min,fl < 15.8°C followed by T min,gr > 18.6°C). Arabica is native to Ethiopia, and so unsurprisingly parts of this country do not usually experience any hazards during the year.

Due to its higher inter-annual variability, precipitation could be the most important variable that affects large-scale coffee production due to the spatial heterogeneity of P an hazards. This is apparent for too-dry conditions in northern and southern Brazil ( Fig 2 e ) and for too-wet conditions in Colombia, Peru and Indonesia ( Fig 2 f ). By comparison, the majority of regions either never experience temperature-based hazards, or they are a feature of the climate.

Southern Brazil exhibits the greatest spatial variability, with varying degrees of susceptibility to VPD gr , T max,gr and low P an ( Fig 2 a , 2 b and 2 e ). As by far the largest grower of Arabica, global coffee production is heavily dependent on favourable climate conditions in southern Brazil, or on effective management of poor conditions.

Number of years between 1980 and 2020 during which climate variables surpass biophysical coffee thresholds for (a) high growing-season VPD (VPD gr ) or low flowering-season minimum temperature (T min,fl ), (b) high growing-season maximum temperature (T max,gr ) or minimum temperature (T min,gr ), (c) and (d) low and high growing-season mean temperature (T gr ), (e) and (f) low and high annual precipitation (P an ). Robusta regions and corresponding hazard definitions are to the right of the red line, plus northern Brazil. The map base layer is available from Natural Earth at https://www.naturalearthdata.com/downloads/110m-physical-vectors/110m-coastline/ .

As mentioned in Section 2.2, the biophysical hazard thresholds are not always ‘extreme’ values of each variable. The historical frequency of each climate hazard varies greatly by region ( Fig 2 ). In some regions, some hazards have never occurred in the period 1980–2020. The vast majority of coffee regions never experience too-cold growing season temperatures, for example ( Fig 2 c ). Conversely, there are regions in which a hazard occurs every year, such as growing season temperatures over 22°C in southern Brazil ( Fig 2 d ). In these cases, the ‘hazard’ is just a feature of the climate that must be managed to attain the best possible yields given those conditions. For example, regions with too-high temperatures often use shading as a management technique [ 58 ], while irrigation is employed to mitigate water stress in regions that do not receive optimal precipitation [ 59 , 60 ].

However, the region has experienced a concerning number of hazards since 2014. Since then, only two years have been hazard-free, with four years featuring multiple concurrent hazards. If the past seven years are indicative of the future, farmers may have to adapt to a hotter and drier climate to avoid negative impacts on coffee production.

Southern Brazil, the world’s most productive Arabica producing area, has experienced among the fewest climate hazards, with only seven years featuring hazards between 1980 and 2013 (right-hand bar plot of Fig 5 ). This implies that, by and large, the majority of coffee producers in southern Brazil have not often had to adapt cultivation practices or implement mitigation strategies, at least for the hazards analysed here.

The climate change signal evident for individual hazards is stark for compound events, with a clear shift towards warm or dry hazards. This shift is supported by trend test results. We find that the Mann-Kendall test yields a statistically significant (p = 0.0003) upward trend for the annual number of warm or dry hazards per year, and a significant (p = 0.001) downward trend for the number of cold or wet hazards per year. With the lack of evidence for trends in precipitation hazards, the increasing number and shift in the type of hazard is driven by increasing temperatures (and related increases in VPD).

The main panel shows the number of hazard events per region and year. Shading indicates whether the majority of hazards are ‘warm or dry’ (brown) or ‘cold or wet’ (green) according to Table 1 . On four occasions, one hazard from each of these classifications occurred (pink). The right-hand bar plot shows the number of hazards per region over the whole period. The top bar plot shows the number of hazards per year across all regions. These bar plots are shaded according to whether they are ‘warm or dry’ or ‘cold or wet’.

The annual number of hazards globally has increased since 1980 ( Fig 5 ). Most regions, with Uganda and India being the exceptions, appear to have experienced a greater number of hazards in the past decade. Since 2010, there have been five years during which at least 20 hazards occurred across all regions, compared to just once prior, in 1998 (top bar plot of Fig 5 ). This implies a greater risk to large-scale coffee supply from spatially compounding climate hazards. In 2016, for example, every region experienced at least one hazard, with most of the top growing regions (northern and southern Brazil, Colombia and Indonesia) experiencing multiple hazards simultaneously.

Region-level hazard events occur when a relatively large proportion (here, above one standard deviation) of the region’s grid cells experience a hazard. This definition means that there are no region-level events in regions for which the hazard is an ever-present feature of the climate. One example is for southern Brazil, which has never suffered a region-level event for T gr ( Fig 4 b ). This is because no grid cell has ever had growing season temperatures below 18°C ( Fig 2 c ). In addition, as virtually all grid cells are above 22°C every year ( Fig 2 d ), there is never a situation in which the proportion of the region experiencing the hazard is large enough to satisfy the definition of a region-level event. Changes in these events therefore reflect whether a region is becoming more or less susceptible to a hazard, rather than the region’s general suitability for coffee cultivation.

Time series show the occurrence of climate hazards for each region, for (a) growing season VPD (VPD gr ) and maximum temperature (T max,gr ; Arabica), and minimum temperature in the flowering (T min,fl ) and growing (T min,gr ) seasons (robusta), (b) growing season mean temperature (T gr ) and (c) annual precipitation (P an ).

Changes in temperature-related hazards show a clear climate change signal. Arabica regions are now much more likely to experience widespread high maximum temperatures and VPD ( Fig 4 a ). Until around 2000, robusta regions were susceptible to cold overnight temperatures in the flowering season. Since then, these hazards have become less common, replaced by warm minimum temperatures in the growing season. A similar picture is evident in growing season mean temperatures across all regions, with a clear shift over time from below- to above-optimal temperatures ( Fig 4 b ). In contrast, changes in sub-optimal precipitation totals do not display a clear trend ( Fig 4 c ).

3.3 Drivers of climate hazards and compound events

Fig 5 suggests that years with high numbers of climate hazards might be related to strong ENSO events. In particular, there were significant El Niño events in 1998, 2015, 2016 and 2019, and these years also featured high numbers of warm or dry hazards. To try and isolate any possible influence of the climate modes on the hazards from the climate change signal, we analyse their relationship using detrended data. Time series of the detrended region-level hazards are shown in Fig C in S1 Text.

Of the tropical ocean modes, ENSO is the most strongly correlated with precipitation and temperature (Fig 6). El Niño-like conditions favour warmer and drier conditions for every region except Southern Brazil, in which wetter conditions are more likely. The IOD has a relatively weak association with surface conditions. The strength of the IOD typically peaks in August to October, but the weak teleconnection is not to do with timing. This is because the growing seasons of each country, including those near the Indian Ocean where a strong teleconnection might be expected, span a wide array of months. The TNA exhibits similar (but weaker) relationships to surface variables as ENSO. The TNA and Atlantic Niño have similar teleconnection patterns to each other, and to the IOD. The exception is for Mexico, Central America and Colombia, for which the correlation sign is opposite (Figs D and E in S1 Text).

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 6. Relationship between climate modes and surface variables. Spearman correlation between detrended growing season ENSO (measured by Niño3.4), Indian Ocean Dipole (DMI), Tropical North Atlantic index (TNA), Madden-Julian Oscillation indices for phases 1 and 4 (MJO 1 or MJO 4 ), and detrended growing season mean temperature (left column) or annual precipitation (right column). Robusta regions are to the right of the red line, plus northern Brazil. The map base layer is available from Natural Earth at https://www.naturalearthdata.com/downloads/110m-physical-vectors/110m-coastline/. https://doi.org/10.1371/journal.pclm.0000134.g006

The MJO is quite strongly related to variations in temperature and precipitation, particularly in northern Brazil and Indonesia (Fig 6g–6j). The magnitude of these correlations does not vary greatly as a function of the MJO phase (also see Figs F and G in S1 Text), but the direction does. When the MJO is in phase 1 more often than normal, the majority of regions experience warmer temperatures and reduced precipitation, except in southern Brazil. Conversely, anomalously more frequent phase 4 days are linked to cooler and wetter conditions.

Our results do not feature the typical MJO teleconnection dipole pattern, whereby precipitation is enhanced over one half of the globe and suppressed in the other [35–37]. For example, we might expect MJO 4 , which indicates the extent to which the MJO is active (promoting precipitation) in the Maritime Continent, to be positively correlated with precipitation in Indonesia and negatively correlated with precipitation in northern Brazil. However, we see a positive correlation almost everywhere, including northern Brazil. This behaviour is potentially because the growing seasons over which we aggregate our MJO indices are much longer (four to nine months) than the time the MJO takes to traverse the planet (one to two months).

The El Niño year of 1998 is even more apparent in the detrended data than in the original series (Fig 7 and Fig C in S1 Text). Over all regions, 40 hazards occurred, 37 of which were warm or dry. As well as El Niño, that year saw positive values for the IOD, Atlantic Niño, TNA and TSA indices. In addition, there was a higher frequency of MJO activity near Africa, at the expense of activity over the Maritime Continent and western Pacific.

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 7. Detrended climate hazards and mode indices time series. (a) Number of hazards per year across all regions. Circles and triangles denote years that featured above average numbers of ‘cold or wet’ and ‘warm or dry’ hazards, respectively. (b) Mode indices averaged across all regions’ growing seasons. The indices are standardised by dividing by their standard deviation in time. The mode indices are for Niño3.4 (ENSO), the Indian Ocean Dipole Mode index (IOD), the Atlantic Niño index (Atl. Niño), the Tropical North and South Atlantic indices (TNA and TSA) and the Madden-Julian Oscillation indices for each of the eight phases (MJO 1 through MJO 8 ). https://doi.org/10.1371/journal.pclm.0000134.g007

Other years with a strongly positive Niño3.4 index, such as 1983, 1987, 2015 and 2016, also coincide with larger numbers of warm and dry climate hazards. Conversely, La Niña-like years such as 1989, 1999, 2000, 2008 and 2011 may be associated with some of the highest totals of cold and wet hazards.

As with so many studies analysing the ENSO teleconnection to surface climate, there is no one-to-one mapping of the ENSO phase to the climate hazards. The resultant Niño3.4 index after averaging across years with anomalously high numbers of warm/dry or cold/wet hazards (triangles and circles in Fig 7a) is around 0.6 or −0.7, respectively (Fig 8a). While the magnitude of these anomalies may seem small, they are statistically significant (according to a circular block bootstrap procedure carried out in the same manner as for the trend tests). As such, there is a clear association between ENSO phase and strength with the number and type of annual hazards (Fig 8b). An El Niño-like SST pattern tips the odds in favour of an increased number of warm or dry hazards globally, and vice versa for La Niña signature.

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 8. Climate modes during detrended spatially compounding warm/dry and cold/wet years. (a) Mean of standardised climate mode indices over years that featured above-average numbers of ‘warm or dry’ or ‘cold or wet’ hazards. Black circles indicate statistical significance. (b)-(e) Hazards per year plotted against ENSO (Niño3.4), Madden-Julian Oscillation phase 1 (MJO 1 ), TNA and Madden-Julian Oscillation phase 4 (MJO 4 ) indices, shaded according to ‘warm or dry’ (brown) and ‘cold or wet’ (green) hazards. https://doi.org/10.1371/journal.pclm.0000134.g008

Of the remaining ocean modes, the TNA has the highest magnitude anomalies on spatially compounding years, with an average anomaly of −0.38 during cold or wet years (Fig 8a). However there is not an obvious distinction in TNA index magnitude or phase between cold/wet and warm/dry years (Fig 8d). Given that a cooler TNA has been associated with June-September droughts across tropical regions [22], we might have expected a negative relationship between the TNA index and the number of warm or dry hazards. However, we find a potentially opposite relationship, although it is weak and likely influenced by the outlier year of 1998. The differences in results we find compared to [22] are perhaps not surprising due to the different regions and seasons analysed, plus we consider temperature variables as well as precipitation.

After ENSO, the MJO exhibits the strongest changes in behaviour during spatially compounding years. When these compound years are characterised by cold and wet hazards, the MJO is more active than usual in the Maritime Continent (phases 4 and 5) at the expense of activity in the western Hemisphere and Africa (phases 1 and 8; Fig 8a, 8c and 8e). During years where hot and dry hazards dominate, this activity is reversed.

These results implicate ENSO, and to a lesser extent the MJO, as the climate modes most influencing global climate hazards important for coffee production. However, as ENSO and the MJO are correlated it is difficult to isolate the effects of each of these climate modes. Still, while the overall relationship between ENSO and the MJO is difficult to ascertain, El Niño events have been shown to modulate MJO amplitude in boreal winter [35, 61].

For the growing seasons considered here, we find Niño3.4 to be relatively strongly correlated with the MJO in phases, 1, 4 and 8 (Spearman correlation between |0.48| and |0.94|, depending on the growing season), and modestly correlated with phases 3, 5 and 7 (between |0.38| and |0.84|; Fig H in S1 Text). The MJO indices are correlated with each other, as expected given their construction. The TSA index is positively correlated with the Atlantic Niño index, which is unsurprising given the spatial overlap (Fig 4a).

We use regression analysis, excluding strongly correlated mode indices as explanatory variables, to identify the relative importance of the climate modes in explaining hazard frequencies for each region. In general, we find the Gaussian GLM performs reasonably well, as indicated by the apparent normality of the residuals (Figs I and J in S1 Text).

ENSO is the most important mode, as it has the largest absolute standardised coefficient, for explaining climate hazard occurrences in tropical South American regions (i.e. excluding southern Brazil) and for Indonesia (Fig 9). ENSO is also an important predictor for Vietnam and Mexico, although no more or less important than the IOD (for Vietnam and Mexico) and the Atlantic Niño, phase 1 MJO and phase 6 MJO indices (for Vietnam only). In some regions for which ENSO is not selected in the models (Ethiopia, Guatemala, Uganda 2 and India), MJO phases that are strongly correlated to Niño3.4 are present.

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 9. Important climate modes for regional climate hazards. Standardised regression coefficients of explanatory variables for the best-performing Gaussian GLM models. Positive coefficient values indicate a positive relationship between the explanatory variable and the number of warm or dry hazards. Negative coefficient values indicate a positive relationship between the explanatory variable and the number of cold or wet hazards. The explanatory variables are Niño3.4 (ENSO), the Indian Ocean Dipole Mode index (IOD), the Atlantic Niño index (Atl. Niño), the Tropical North and South Atlantic indices (TNA and TSA) and the Madden-Julian Oscillation indices for each of the eight phases (MJO 1 through MJO 8 ). https://doi.org/10.1371/journal.pclm.0000134.g009

Southern Brazil being relatively less influenced by ENSO may be important in mitigating systemic risk to global coffee supply. Colombia, Vietnam, northern Brazil and Indonesia are also major producers, yet are susceptible to climate hazards during El Niño-like conditions. With southern Brazil’s production power, it may be capable of making up for other regions’ shortfalls during El Niño events.

Despite the weak correlation of the DMI to temperature and precipitation (Fig 6), and its ambiguous relationship with spatially compounding hazard years (Figs 7 and 8a), the regression models imply the IOD is important in explaining climate hazard frequency variability in seven of the 15 regions (Fig 9). A positive IOD is associated with higher numbers of cold or wet hazards near the eastern Indian Ocean (Vietnam and India), plus Peru. The central and north American regions, on the other hand, may experience an increase in warm or dry hazards during a positive IOD.

[END]
---
[1] Url: https://journals.plos.org/climate/article?id=10.1371/journal.pclm.0000134

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.