Effects of diurnal temperature range and drought on wheat yield in Spain

42 Pages·2017·1.82 MB·English

This is a repository copy of Effects of diurnal temperature range and drought on wheat
yield in Spain.
White Rose Research Online URL for this paper:
http://eprints.whiterose.ac.uk/100058/
Version: Accepted Version
Article:
Hernandez-Barrera, S, Rodriguez-Puebla1, C and Challinor, AJ
orcid.org/0000-0002-8551-6617 (2017) Effects of diurnal temperature range and drought
on wheat yield in Spain. Theoretical and Applied Climatology, 129 (1-2). pp. 503-519. ISSN
0177-798X
https://doi.org/10.1007/s00704-016-1779-9
Reuse
Unless indicated otherwise, fulltext items are protected by copyright with all rights reserved. The copyright
exception in section 29 of the Copyright, Designs and Patents Act 1988 allows the making of a single copy
solely for the purpose of non-commercial research or private study within the limits of fair dealing. The
publisher or other rights-holder may allow further reproduction and re-use of this version - refer to the White
Rose Research Online record for this item. Where records identify the publisher as the copyright holder,
users can verify any specific terms of use on the publisher’s website.
Takedown
If you consider content in White Rose Research Online to be in breach of UK law, please notify us by
emailing [email protected] including the URL of the record and the reason for the withdrawal request.
[email protected]
https://eprints.whiterose.ac.uk/
Manuscript Click here to download Manuscript manuscript.tex
Click here to view linked References
TheoreticalandAppliedClimatologymanuscriptNo.
(willbeinsertedbytheeditor)
1
2
3
4
5
Effects of diurnal temperature range and drought on wheat yield in Spain
6
7
8
9 S.Hernandez-Barrera · C.Rodriguez-Puebla · A.J.Challinor
10
11
12
13
14
15
16
17 Received:/Accepted:
18
19
20
Abstract This study aims to provide new insight on the wheat yield historical response to cli-
21 1
22
mateprocessesthroughoutSpainbyusingstatisticalmethods.Ourdataincludesobservedwheat
23 2
24
yield, pseudo-observations E-OBS for the period 1979 to 2014, and outputs of general circula-
25 3
26
27 4 tionmodelsinPhase5oftheCoupledModelsInter-comparisonProject(CMIP5)fortheperiod
28
29 5 1901 to 2099. In investigating the relationship between climate and wheat variability, we have
30
31 appliedtheapproachknownasthePartialLeast-Squareregression,whichcapturestherelevant
6
32
33 climatedriversaccountingforvariationsinwheatyield.Wefoundthatdroughtoccurringinau-
7
34
35 tumn and spring and the diurnal range of temperature experienced during the winter are major
8
36
37
processes to characterize wheat yield variability in Spain. These observable climate processes
9
38
39
areusedforanempiricalmodelthatisutilizedinassessingthewheatyieldtrendsinSpainunder
40 10
41
different climate conditions. To isolate the trend within the wheat time series, we implemented
42 11
43
S.Hernandez-Barrera·C.Rodriguez-Puebla
44
45 DepartmentofFundamentalPhysics,UniversityofSalamanca,PlazadelaMerceds/n,37008Salamanca,Spain
46
E-mail:[email protected]
47
48
49 A.J.Challinor
50
SchoolofEarthandEnvironment,UniversityofLeeds,LeedsLS29JT,UK
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
2 S.Hernandez-Barreraetal.
1
2
the adaptive approach known as Ensemble Empirical Mode Decomposition. Wheat yields in
3 12
4
the twenty-first-century are experiencing a downward trend that we claim is a consequence of
5 13
6
widespread drought over the Iberian Peninsula and an increase in the diurnal range of temper-
7 14
8
9 15 ature. These results are important to inform about wheat vulnerability in this region to coming
10
11 16 changesandtodevelopadaptationstrategies.
12
13
Keywords Climate Change impact · Empirical wheat yield model · Partial Least Square
14 17
15
regression·Climatevariability
16 18
17
18
19
20 19 1 Introduction
21
22
The IPCC (2014) report on impacts, adaptation, and vulnerability informs that rising tempera-
23 20
24
25 21 turesandchangesinrainfallmaybenefitagricultureinsomecountriesbutmaydamageinsome
26
27 22 other parts, as consequence of climate variability, weather extremes, and changes of the water
28
29 cycle. The Joint Research Centre (JRC) denoted a reduction around 20% of agricultural pro-
23
30
31 duction in Southern Europe by the end of the twenty-first century, in the PESETA II Project
24
32
33
on impact studies in Europe (Ciscar et al, 2014). They also refer that technical adaptation can
25
34
35
improvetheyieldsalloverEurope,however,modesteffectivenessisexpectedinsouthernSpain
26
36
37
due to excessive aridity. Particularly, in Spain there is currently a national concern about agri-
38 27
39
cultural productions. Wheat is one of the world’s most basic and necessary, its productivity is
40 28
41
as large as olive, citrus and grape farming in Spain (FAO, 2014). Our study aims to address
42 29
43
44 30 thefollowingquestions:whatclimatevariablesareessentialtoexplainingwheatyieldchanges?
45
46 31 What future trends will wheat production experience considering our findings regarding these
47
48 variables?
32
49
50 Some of the motivations to perform this study are: diversity of results on climate change
33
51
52 and crop impacts; variety in crop methodologies; and the need to evaluate the impacts of cli-
34
53
54
55
56
57
58
59
60
61
62
63
64
65
EffectsofdiurnaltemperaturerangeanddroughtonwheatyieldinSpain 3
1
2
mate change on crops variability at the regional level. The methods to evaluate the impact of
3 35
4
climate change on crop productions can be gather into process-based and statistical models.
5 36
6
White et al (2011) reviewed methodologies for simulating impacts of climate change on crop
7 37
8
9 38 productions using process-based crop models, which succeed locally. However, Palosuo et al
10
11 39 (2011)noticedthatprocess-basedcropmodelsforwinterwheatsimulationreproducepoorlythe
12
13 corresponding observations, since agricultural management input data are seldom available for
40
14
15 larger areas. Otherwise, Angulo et al (2013) discussed the regionally applicability of process-
41
16
17 basedcropmodels.Rosenzweigetal(2013)indicatedthatwheatsimulationismoresensitiveto
42
18
19
thecropmodelthantoglobalclimatemodelsimulationandCarter(2013)recommendedmulti-
43
20
21
modelyieldprojectionsforimpactstudies.Someauthors(RotterandHohn,2015;Assengetal,
22 44
23
2013) performed inter-comparisons of process-based crop models by analyzing the uncertainty
24 45
25
of wheat simulation under climate change and considering differences in model structures. A
26 46
27
28 47 meta-analysesfromnumerousstudiesindicatedthatprojectedresponseofcroptoclimatevari-
29
30 48 ability and change can vary according to the methodology (Challinor et al, 2014). However,
31
32 process-based models are useful for determining the causes of yield variations while to repro-
49
33
34 duce historical yield variations statistical models are appropriated (Watson et al, 2015). Thus
50
35
36 statisticalapproachesareattractingattentionforassessingclimatechangeimpactsoncroppro-
51
37
38
ductionforlargerareas(LobellandBurke,2010;Lobell,2013).
52
39
40
Regardingwheatyield,Lobelletal(2011a)studiedtheimpactofclimatetrendonglobalcrop
41 53
42
productionandMooreandLobell(2014)pointoutthebenefitsofadaptationtocompensatethe
43 54
44
negative effect of rising temperature on the crops in Europe. The impacts of climate change on
45 55
46
47 56 winterwheatarethoughttobenegativeacrossEurope(Olesenetal,2011).Trnkaetal(2011b)
48
49 57 calculatedandprojectedagroclimateindices,reporteddecreasesinpotentialproductivityinthe
50
51 caseofNorthandSouthMediterraneanzonesduetoincreasesintheproportionofdrydaysand
58
52
53 increaseinheatwaves.
59
54
55
56
57
58
59
60
61
62
63
64
65
4 S.Hernandez-Barreraetal.
1
2
Themajorityofagro-climaticinvestigationsfocussedonanalysingtherelationshipsbetween
3 60
4
crop yield, temperature, and precipitation; Challinor et al (2014) summarized the responses of
5 61
6
variouscropstochangesintemperature,precipitationandeffectivenessofadaptation.Currently,
7 62
8
9 63 extreme indices of the apparent impacts upon ecosystems (Lobell, 2007; Lobell et al, 2011b;
10
11 64 Ruiz-Ramosetal,2011;Trnkaetal,2014;Eitzingeretal,2013)havegarneredmuchattention.
12
13 Otherstudiesdevelopanalysesregardingtherelationshipbetweencropproductionsandtelecon-
65
14
15 nections(Atkinsonetal,2005;Chenetal,2015;GonsamoandChen,2015;Hansenetal,2001;
66
16
17 Iizumi et al, 2014; Podesta et al, 2002; Royce et al, 2011; Bannayan et al, 2011; Dalla Marta
67
18
19
etal,2011;Jarlanetal,2014;Tianetal,2015).
68
20
21
In Spain, the effects of climate variations on wheat and barley yields in the Ebro valley
22 69
23
have been estimated by Vicente-Serrano et al (2006) using drought indices and remote sensing
24 70
25
data. Iglesias and Quiroga (2007) researched the risks entailed by climate variability for cereal
26 71
27
28 72 production at five sites in Spain; Ruiz-Ramos et al (2011) projected the effects of maximum
29
30 73 temperatureoncerealyieldsbyusingregionalclimatemodels.Studiesbasedonteleconnections
31
32 and crop productions in Spain were conducted by Capa-Morocho et al (2014); Gimeno et al
74
33
34 (2002); Rodriguez-Puebla et al (2007). However, the responses of regional crops to climate
75
35
36 changes are very much uncertain, as indicated by Rotter (2014), hence multiple impact models
76
37
38
shouldbeconsideredforprojectingfuturecropproductivity(Challinoretal,2014).
77
39
40
Most of the statistical studies are based on regression of the historical crop yield, precipita-
41 78
42
tion and temperatures. We aim to identify relationships between wheat variability in Spain and
43 79
44
climate processes such as drought and extreme temperature indices, updating previous work
45 80
46
47 81 (Rodriguez-Puebla et al, 2007) and introducing new approaches: namely, the Partial Least-
48
49 82 Squares(PLS)regressionforascertainingthemodesofclimatevariablesassociatedwithwheat
50
51 yield variability, Ensemble Empirical Mode Decomposition (EEMD) for identifying the trends
83
52
53 andscalesofwheatyieldvariability,andtheMultivariateRegressionmodelforempiricallyes-
84
54
55
56
57
58
59
60
61
62
63
64
65
EffectsofdiurnaltemperaturerangeanddroughtonwheatyieldinSpain 5
1
2
timating wheat yield variability, considering the relative effects of different climate variables
3 85
4
that affect soil moisture content as temperature and precipitation. Hence we have not consid-
5 86
6
ered changes in soil water storage capacity andCO variations. The empirical statistical model
7 87 2
8
9 88 of wheat yield variability in Spain is applied to estimate wheat productivity in the twentieth
10
11 89 and twenty-first centuries, using the output data of twelve GCMs of CMIP5. We analysed the
12
13 changesinwheatyieldsforindividualmodelsandthecorrespondingMulti-modelforhistorical
90
14
15 andrepresentativeconcentrationpathway8.5(RCP8.5)experiments(Tayloretal,2012).
91
16
17 The paper is organized in the following way: the data and methods used are indicated in
92
18
19
Section2.Resultsregardingtheanalysisofclimateimpactuponwheatyield,thederivedstatis-
93
20
21
tical model, and the identification of trends under different climate conditions are presented in
22 94
23
Section3.DiscussionandmainfindingsaresummarizedinSections4and5,respectively.
24 95
25
26
27
28 96 2 DataandMethods
29
30
31 97 2.1 Dataandstudyarea
32
33
Data regarding wheat production or yield over Spain is collected by the Spanish Agriculture,
34 98
35
36 99 Food, and Environment Department (MAGRAMA, 2015). Wheat yield refers to the weight of
37
38 100 production divided by the area of cultivation (T/ha). We used data from different provinces for
39
40 the period 1979 to 2014. Regarding climate data in Spain (35-45N and 10W-5E), we used the
101
41
42 daily pseudo-observations E-OBS (V11.0) dataset 0.25-degree resolution of precipitation (Pr),
102
43
44 mean(Tmed),maximum(Tmax),andminimum(Tmin)temperatures(Haylocketal,2008)for
103
45
46
theperiodofSeptember1978toAugust2014.Althoughthereareotherdatasetsbasedondenser
104
47
48
observationalnetworks,Spain02(Herreraetal,2012),stationdensityisnotasrelevantforpur-
49 105
50
poses of this research as we are primarily interested in climate variations that affect the aggre-
51 106
52
gatedwheatyieldinSpain.Furthermore,theSpain02datasetwasnotavailableuntil2014,while
53 107
54
55
56
57
58
59
60
61
62
63
64
65
6 S.Hernandez-Barreraetal.
1
2
the E-OBS data are frequently updated and extensively used and tested. From the daily tem-
3 108
4
peratureswederivedthedailydiurnaltemperaturerange(DTR),thenthemonthlyandseasonal
5 109
6
DTR.Fromthedailyprecipitationwederivedtheaccumulatedmonthlyandseasonalprecipita-
7 110
8
9 111 tion,thenwederivedtheStandardizedPrecipitationIndex(SPI)(WMO,2012;Vicente-Serrano
10
11 112 et al, 2010) on a time scale of one month to reflect the response of wheat yield to rapid-onset
12
13 droughtevents(Otkinetal,2015)oragriculturaldrought(Lorenzo-Lacruzetal,2013).TheSPI
113
14
15 consists of the transformation of precipitation into a standardized normal distribution, obtained
114
16
17 withthescriptofNcarCommandLanguage(NCL)(UCAR/NCAR,2015).
115
18
19
Ourmodelindirectlytakesintoaccounttheeffectofsoilmoistureeffectoncrops,byconsid-
116
20
21
ering both variables: precipitation, characterized with the SPI index, and temperature using the
22 117
23
DTRindex.Acomparisonofdroughtindiceseffect(Begueriaetal,2014)onwheatyieldwould
24 118
25
beachallengeforfurtherresearchsincethechoiceoftheformulatocomputeevapotranspiration
26 119
27
28 120 iscurrentlyunderdebate(Dai,2011;Trenberthetal,2014).
29
30 121 We used a second dataset of climate variables of Pr, Tmed, Tmax and Tmin correspond-
31
32 ing to the CMIP5 models (Taylor et al, 2012) indicated in the supplementary material (Table
122
33
34 S1). In this study, we considered the historical experiment corresponding to the period of time
123
35
36 fromSeptember1901toDecember2005,forcedbyobservedatmosphericcompositionchanges,
124
37
38
reflecting both anthropogenic and natural sources, and the future projection of the RCP8.5 ex-
125
39
40
perimentfromJanuary2006toAugust2099,whichcorrespondstothepathwaywiththehighest
41 126
42
greenhouse gas emissions and a radiative forcing of 8.5 W/m2 in 2100 (Riahi et al, 2011). One
43 127
44
realization or ensemble run of the individual models is taken into account in order to give all
45 128
46
47 129 modelsthesameweight.TheDTRandSPImodelledarederivedasexplainedaboveinthecase
48
49 130 ofpseudo-observations.Forthiscomparison,wehavere-griddedthedatatothesameresolution
50
51 asE-OBSusingthebilinearinterpolationincludedintheClimateDataOperator(CDO)software
131
52
53 (Schulzweida,2015).ThemodelperformanceoftheGCMsselectedhasbeenevaluatedthrough
132
54
55
56
57
58
59
60
61
62
63
64
65
EffectsofdiurnaltemperaturerangeanddroughtonwheatyieldinSpain 7
1
2
comparisonsofsomepatternstatistics(Taylor,2001)andclimographsagainsttheobservations,
3 133
4
includedinthesupplementarymaterial.
5 134
6
7
8
9 135 2.2 EmpiricalModeDecomposition
10
11
Muchoftheyieldincreaseislikelyduetoimprovedcropmanagement,accordingtoresults
12 136
13
14 137 of Moore and Lobell (2015), since the contribution of the long-term temperature and precipi-
15
16 138 tation trends to wheat yield trend is quite small during the observational period (Xiao and Tao,
17
18 2014). In addition, recent study (Asseng et al, 2013) indicate the controversial benefits from
139
19
20 enhancedCO . Therefore, de-trending the wheat time series is recommended before exploring
140 2
21
22 the relationships between climate variability and wheat yield. Ensemble Empirical Mode De-
141
23
24
composition (EEMD) is an adaptive approach to deconstructing a time series without linear or
142
25
26
stationary assumptions (Chen et al, 2013; Huang et al, 1998; Moghtaderi et al, 2013; Wu et al,
143
27
28
2007). This approach acts as a high-pass filter and is used in decomposing wheat yield time
29 144
30
series.EMDisasiftingprocesstodecomposeatimeseriesx(t):
31 145
32
33
34 k
35 x(t)= ∑ci(t)+r(t) (1)
36 i=1
37
38 146 Here,ci(t)areintrinsicmodefunctions(IMFs)andr(t)istheresidual.IMFsdependonthe
39
signalandsatisfytwoconditions(Huangetal,1998):thenumberofextremeandthenumberof
40 147
41
zero crossing vary by at most one, and the local mean of each IMF is zero. The decomposition
42 148
43
44 149 procedure is as follows: 1) locate all maxima and minima of the x(t) and connect all maxima
45
46 150 (minima)withacubicspline;2)computethedifferencebetweenthetimeseriesandthemeanof
47
48 upperandlowerenvelopestoyieldanewtimeseriesh(t);3)forthetimeseriesh(t),repeatsteps
151
49
50 1) and 2) until upper and lower envelopes are symmetric with respect to the zero mean under
152
51
52 153 thespecifiedcriteriainordertoobtaintheIMF,ci(t);4)subtractci(t)fromoriginaltimeseries
53
54
55
56
57
58
59
60
61
62
63
64
65
8 S.Hernandez-Barreraetal.
1
2 x(t) to yield a residual r(t) and treat r(t) as the original time series and repeat steps 1-3 until
3 154
4
theresidualbecomesamonotonicfunctionorafunctionwithonlyoneextreme;thiscompletes
5 155
6
the sifting process (Chen et al, 2013). For better signal separation, a Monte Carlo approach
7 156
8
9 157 recommended,inwhichzero-meanGaussianwhitenoiseisaddedtoeachEMDprocessandthe
10
11 158 modified method is designed as Ensemble Empirical Mode Decomposition (EEMD) (Franzke,
12
13 2010;Wuetal,2011).
159
14
15 The utility of the EEMD approach in separating the trend from natural variability in ana-
160
16
17 lyzing phenological responses to warming is demonstrated in the paper by Guan (2014).The
161
18
19
robustness of EEMD has been applied in ascertaining surface air temperature trends (Cappar-
162
20
21
elli et al, 2013; Ji et al, 2014), and trends in sea surface temperature (Feng et al, 2014). In our
22 163
23
case, we use EEMD as a high-pass filter by retaining all the IMFs except the residual or trend
24 164
25
componentoftheobservedwheattimeseries;therefore,otherimprovedtechniques(Colominas
26 165
27
28 166 et al, 2014) for analysing the intrinsic mode functions were not implemented. This method is
29
30 167 also used to represent the trend component of the wheat yield simulation from CMIP5 models.
31
32 TheestimationutilizedtheMatlabEMD/EEMDpackageofFlandrinetal(2004).
168
33
34
35
36
37 169 2.3 PartialLeastSquaresRegression
38
39
TheinfluenceofclimatevariablesonwheatproductionisinvestigatedthroughuseofthePLS
40 170
41
regression.Thisprocedureisapowerfulmethodfordescribingcovariancebetweenvariablesby
42 171
43
44 172 means of latent variables. This process entails dimension reduction and regression adjustment.
45
46 173 The method was developed by Wold et al (2001) in order to solve the problem of co-linearity
47
48 in linear regression. It has been applied with great success in chemometrics and is now being
174
49
50 applied in climatology (Gonzalez-Reviriego et al, 2015; Smoliak et al, 2015, 2010; Wallace
175
51
52 et al, 2012). PLS regression seeks to predict variables (Y) based on independent variables (X)
176
53
54
55
56
57
58
59
60
61
62
63
64
65
EffectsofdiurnaltemperaturerangeanddroughtonwheatyieldinSpain 9
1
2
-that are correlated- by finding a few new uncorrelated variables, in addition to denominated
3 177
4
latent variables. Imposing the constraint of orthogonality upon the latent variables serves to
5 178
6
mitigatetheproblemofmulti-linearityandreducesthenumberofindependentvariablesneeded
7 179
8
9 180 to describe variations in the dependent data (Y); but PLS also chooses the optimum subset of
10
11 181 predictors, which is not guaranteed when the Principal Regression Method is applied (Abdi,
12
13 2010).Therefore,PLSfindscomponentsfromX thatbestpredictY.
182
14
15 Inourstudy,PLSregressionisappliedintwodifferentways.Thefirststepbeginstoassess
183
16
17 the modes of a climate field in conjunction with the observed wheat yield variability corre-
184
18
19
sponding to the observational period (1979-2014). The modes include spatial patterns and PLS
185
20
21
componentsortimeseriescongruentwiththewheattimeseries.Weobtainedtailoredtimeseries
22 186
23
of climate variation components that explain changes in wheat yield. In this case, the observed
24 187
25
climate variables will be referred to as independent variables, or fields that vary in time and
26 188
27
28 189 space dimensions X(T,M), (M =lat×lon), and the detrended spatially averaged wheat yield
29
30 190 inSpainisthedependentvariable,whichvarieswithinthetimedimensionY(T).Theoutcomes
31
32 includesomeorthogonallatentspatialvectorsZ(M)andtemporaluncorrelatedPLScomponents
191
33
34 B(T). Figure 1a shows a schematic diagram of the PLS approach. The procedure is applied to
192
35
36 different climate fields such as Tmax, Tmin, Tmean, SPI, and DTR. The PLS component B,
193
37
38
correspondingtodifferentclimatefields,willbeconsideredinpredictingthedependentvariable
194
39
40 Y byapplyingaforwardandbackwardstepwiseregressionprocedure(Wilks,2006)thatselects
41 195
42
the climate indicators B to be included in the empirical agro-climate model. The uncertainty
43 196
44
of the model was assessed through the use of cross-validation or by repeating the appropriate
45 197
46
47 198 procedure upon data subsets to select robust variables and provide the confidence interval for
48
49 199 the estimation. The quality of the model is given by the Pearson correlation coefficient with
50
51 its error, which is obtained by repeating the correlation for many samples using a bootstrap re-
200
52
53
54
55
56
57
58
59
60
61
62
63
64
65