forecast customization system (focus): a multimodel...

Research ArticleForecast Customization System (FOCUS) A MultimodelEnsemble-Based Seasonal Climate Forecasting Tool for theHomogeneous Climate Zones of Myanmar

Itesh Dash 1 Masahiko Nagai 12 and Indrajit Pal1

1Disaster Preparedness Mitigation and Management Asian Institute of Technology Khlong Nueng Pathum ani ailand2Graduate School of Sciences and Technology for Innovation Yamaguchi University Yamaguchi Japan

Correspondence should be addressed to Itesh Dash iteshdashgmailcom

Received 12 April 2019 Revised 5 September 2019 Accepted 28 November 2019 Published 18 December 2019

Academic Editor Pedro Jimenez-Guerrero

Copyright copy 2019 Itesh Dash et al -is is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

A Multi-Model Ensemble (MME) based seasonal rainfall forecast customization tool called FOCUS was developed for Myanmarin order to provide improved seasonal rainfall forecast to the country -e tool was developed using hindcast data from 7 GlobalClimate Models (GCMs) and observed rainfall data from 49 meteorological surface observatories for the period of 1982 to 2011from the Department of Meteorology and Hydrology Based on the homogeneity in terms of the rainfall received annually thecountry was divided into six climatological zones -ree different operational MME techniques namely (a) arithmetic mean(AM-MME) (b) weighted average (WA-MME) and (c) supervised principal component regression (PCR-MME) were used andbuilt-in to the tool developed For this study all 7 GCMs were initialized with forecast data of May month to predict the rainfallduring June to September (JJAS) period which is the predominant rainfall season for Myanmar -e predictability of raw GCMsbias-corrected GCMs and the MMEs was evaluated using RMSE correlation coefficients and standard deviations -eprobabilistic forecasts for the terciles were also evaluated using the relative operating characteristics (ROC) scores to quantify theuncertainty in the GCMs-e results suggested that MME forecasts have shown improved performance (RMSE 129) comparedto the raw individual models (ECMWF which is comparatively better among the selected models) with RMSE 44 and bias-corrected RMSE 43 over Myanmar Specifically WA-MME (CC 064) and PCR-MME (CC 068) methods have shownsignificant improvement in the high rainfall (delta) zone compared with WA-MME (CC 057) and PCR-MME (CC 056)techniques for the southern zone-e PCRmethod suggests higher predictability skill for the upper tercile (ROC 078) and lowertercile categories (ROC 085) for the delta region and is less skillful over lower rainfall zones like dry zones with ROC 06 and063 for upper and lower terciles respectively-emodel is thus suggested to perform relatively well over the higher rainfall (Wet)zones compared to the lower (Dry) zone during the JJAS period

1 Introduction

Rainfall in Myanmar is highly variable over space and timelargely because of a varied topography and multiple envi-ronmental influences It is directly impacted by the IndianSouth Asian monsoon systems as well as convective rainfallfrom the Bay of Bengal [1 2] -e strength of seasonal rainfallin the country to some extent is influenced by the large-scaleclimate drivers such as El Nintildeo Southern Oscillation (ENSO)and Indian Ocean Dipole (IOD) [1 3ndash5] According to theDepartment of Meteorology and Hydrology (DMH) it is

observed that ENSOrsquos warm phase (El Nintildeo) has resulted indeficient rainfall and higher temperatures while La Nintildea thecold phase tends to have opposite impacts in the country [6]Presence of such a teleconnection between the large-scalephenomena and the local climate of Myanmar is expected toenhance seasonal prediction

DMHrsquos operational seasonal forecasting is based on theanalogue method [7] According to this method rainfallpatterns associated with historical ENSO phases (El Nintildeoand La Nintildea) is likely to re-occur during similar ENSOphases in future -e prediction of the present year would

HindawiAdvances in MeteorologyVolume 2019 Article ID 4957127 15 pageshttpsdoiorg10115520194957127

depend on the years with similar ENSO phases in the pastIt was recommended that an improved seasonal forecastingsystem is required despite the traditional method followedby DMH during a user interaction forum conducted byDMH every year [8] Furthermore capacity self-assessmentexercise conducted by RIMES [9] identified the need fordevelopment of a standard platform in order to assistDMH in generating seasonal climate forecast and assist inanalyzing different global models Moreover need forempirical studies focusing on rainfall variability andforecasting for operational applications in agriculture andwater resource sector was emphasized in Asia [10ndash12] andAfrica [13]

Global climate models (GCMs) are useful tools forpredicting seasonal climate however there is large un-certainty in its prediction mainly because of the as-sumptions made in the initial atmospheric state [14] Tosimulate and capture these uncertainties in the pre-dictions GCMs are processed with different initial con-ditions to generate multiple forecast members called theensembles [15 16] Multimodel ensemble (MME) is aprocess where ensemble members of one GCM are sta-tistically assembled with another GCM or a set of GCMs[17ndash21] -e MME approach has increasingly demon-strated better prediction skills over the tropical Asianregion in long-range forecasting when compared to in-dividual model performances [20 22ndash24] For instancethe MME system developed over India for monthlysea-sonal prediction in real time during the South Asianmonsoon season exhibited satisfactory predictions [25]Similarly North American MMEs (NMME) showed lowersystematic error and higher forecast skills compared to theindividual members over Southeast Asian region [26]

MME schemes can be developed using various statisticalmethods (1) by simply taking mean of all ensembles withassigning equal weight to individual ensemble members [20]or (2) by assigning higher weightage to the statisticallysignificant members of GCMs according to their perfor-mance over the hindcast period [20 24 27ndash29] or usingcomplex neural network algorithms It is however a well-established concept that MMEs would be useful schemes forgenerating improved seasonal outlook But so far no at-tempts were made to develop a long-range prediction systemfor Myanmar using such type of advanced techniques Toolssuch as the Climate Prediction Tool (CPT) [30] ClimateInformation Toolkit (CLIK) [31] and Seasonal ClimateOutlook in the Pacific Island Countries (SCOPIC) [32] havethe functionalities to perform statistical analysis with climatedata but are limited in terms of their utility For instanceCLIK is useful for providing predictions at regional scale butnot at the spatial scale suggested in this current study CPThas the capability to do prediction specific to locations butcannot perform MME-based predictions SCOPIC isdesigned to predict seasonal climate only for the PacificIsland countries and not yet applicable for other regions[33]

-e objective of the present work is to overcome theabovementioned limitations and to develop a web-basedgraphical user interface (web-GUI) forecast customization

system tailored for national use -e tool allows generationof monthly and seasonal climate outlooks using the MMEtechniques and assist in evaluating the performance ofoutlooks -e tool is also capable of providing the outlooksfor the defined climatological zones in Myanmar -emethod will be described in the successive section

2 Study Area

21 Zone Classification Myanmar is geographically situ-ated in Southeast Asia between latitudes 09deg 32prime N and 28deg31prime N and longitudes 9deg 10prime E and 101deg 11prime E Myanmar isclimatologically divided into six major zones (Figures 1(a)and 1(b) [10]) (1) central dry zone which has the lowestaverage annual rainfall and intense agricultural practices(2) eastern zone (shan) (3) northern zone which is mostlywith high terrain and forest areas (4) coastal (Rakhine)(5) delta zone (Ayeyarwady region) and (6) southernzone which receives the maximum annual rainfall zones-is also synchronized well with classifications done inthe past in references [1 5 20] -ese classifications arebased on the average annual rainfall in the country(Figure 2) agroecological zoning and seasonal rainfallpatterns respectively

22 Climatology of Myanmar -e annual rainfall cycle inFigure 2 shows distribution of rainfall mostly concentratedover the JJAS period which is mainly due to the influenceof the southwest monsoon (Figure 1(c)) -e monsoononset is marked during May peak during August andwithdrawal towards the end of September [34] -e spatialdistribution of rainfall however varies significantly duringthis period over all the zones -e central dry zone receivesthe lowest amount of seasonal rainfall while the southernregion receives the highest amount [34] As JJAS is majorrainfall season for all the zones of the country this studyfocused on investigating characteristics of rainfall over JJASand also predicting seasonal rainfall for its operationalapplication in agricultural and water resources manage-ment sector

3 Data and Methods

31 GCM Data Hindcast rainfall data from seven GCMs(listed in Table 1) are obtained for the period 1982ndash2011 Allthe GCM hindcast datasets are at monthly temporal scaleFour fully coupled GCMs namely National Center forEnvironment Predictionrsquos Coupled Forecast System Modelversion 2 (NCEP CFSv2) [37] Geophysical Fluid DynamicsLaboratory (GFDL) COLA and the System 4 (ECMWFTechnical Memorandum 2011) European Center for Me-dium-Range Weather Forecast (ECMWF) are used in thestudy -e CFSv2 model has a spectral triangular truncationof 126 waves (T126) horizontally (equivalent to nearly a100 km grid resolution) and a finite differencing verticallywith 64 sigma-pressure hybrid layers -e atmosphericcomponent in this model is the Global Forecast System (GFS2009) while Geophysical Fluid Dynamics LaboratoryModular Ocean Model 4 (GFDL MOM4) is considered as

2 Advances in Meteorology

the oceanic component -e retrospective 9-month forecastshave initial conditions of the 0000 0600 1200 and 1800UTC cycles for every 5th day starting from 0000 UTC 1January of every year COLA and GFDL models are con-sidered from the US National Multimodel Ensemble(NMME) project phase-II [36] -e COLA forecasts aremade with the NCAR CCMv36 [35] a coupled climatemodel with components representing atmosphere ocean

sea ice and land surface connected by a flux coupler -ree2-tier models ECHAM45 CASST ECHAM45 CFSSST andCCMv36 are used -e ECHAM45 CASSTmodel is forcedwith Constructed Analogue (CA) Sea Surface Temperature(SST) [38] as boundary conditions over tropical oceans(30S-30N) and CFSSST is forced with the ClimateForecasting System (CFS) SST data -e rainfall data forthese global climate models are accessed from the

0

5

10

15

20

Jan

Feb

Mar

Apr

Meteorological stations

May

June July

Aug

Sep

Oct

Nov Dec

Rain

fall

(mm

day

)

Month

Rainfall (mmday)0ndash200

200ndash300

300ndash400400ndash500

500ndash600

600ndash700

700ndash800

800ndash1100

Elevation (m)

(a) (b)

(c)

25ndash50

50ndash100

100ndash200

200ndash400

400ndash600

600ndash800

lt25 800ndash1000

1000ndash1500

gt1500State boundaryCountry boundaryClimatological zonesboundary State boundary

Country boundaryClimatological zonesboundary

0 200km 0 200km

Figure 1 (a) Topography of Myanmar (b) Rainfall during June to September overlaid with meteorological surface observatory andclimatological zones (c) Monthly climatology of Myanmar

Advances in Meteorology 3

International Research Institute data library availableonline at httpiridlldeocolumbiaedu [39] Hindcastdata for NCEP CFSv2 are downloaded from httpcfsncepnoaagovcfsv2downloadshtml -e System 4hindcast data from the ECMWF are retrieved from theMeteorological Archival and Retrieval System (MARS)available online at httpappsecmwfintarchive-catalogue

32 Observation Data Observation rainfall data at dailytime-step from 70 surface observatories for the period of1982ndash2011 are obtained from the Department of Meteo-rology and Hydrology (DMH) Myanmar However datafrom only 49 stations are considered (shown in Figure 1(b))for this study based on the following quality checks cli-matological and temporal checks data homogeneity testfactoring human error and percentage of missing data [40]

33 Methods A complete schematic of the method is de-scribed in Figure 3 which involves data acquisition fromdifferent global centers data preparation and processingbias correction and development of MME schemes

generation of probabilistic forecast and finally evaluationof the model skill -ese steps are described in the sub-sequent sections

34 Data Preparation GCM hindcast datasets are main-tained in different formats by different global producingcenters (GPCs) For example IRI data library stores data insequential binary format CFSv2 datasets are in griddedbinary (grib2) format and ECMWF MARS datasets areavailable in either Network Common Data Format(NetCDF) or grib2 At the same time the synoptic obser-vation datasets accessed from DMH are in the simple text(ASCII) format -erefore a data normalization algorithmwas developed using Python programming language tobring all data to a standard format (mat) to handle the datamore efficiently

35 Data Processing -e proposed methods would use thehindcast data to train the model therefore it is essential tocombine the hindcast data with the forecast data for thesame forecast initialization month For example the study

0

5

10

15

20

25

30

35

Jan Feb Mar Apr May June July Aug Sep Oct Nov DecRa

infa

ll in

mm

Months

Monthly climatology

DryShanNorthDelta

CoastalSouthOverall

Figure 2 Annual cycle of rainfall climatology for all six homogeneous zones (bar graph) and overall Myanmar (line graph)

Table 1 Details of GCM datasets used

Model Resolution Model type Ensemblesize Source

Community climate systemmodel (CCMv36) 2813deg times 2789deg 2-tier 24 National center for atmospheric researchBoulder USA [35]

Center for ocean-land-atmosphere (COLA) 1875deg times1864deg Fullycoupled 10 -e center for ocean-land-atmosphere studies

Fairfax USA [36]Geophysical fluid dynamics laboratory(GFDL) 2500deg times 2000deg Fully

coupled 10 Geophysical fluid dynamics laboratoryPrinceton USA [36]

ECHAM 45 CA SST 2813deg times 2789deg 2-tier 24 Max Planck institute for meteorologyDenmark (Li and Goddard 2005)

ECHAM 45 CFS SST 2813deg times 2789deg 2-tier 24 Max Planck institute for meteorologyDenmark (CFS-predicted SST)

European center for medium-range weatherforecasting (ECMWF) 1500deg times1500deg Fully

coupled 41 European center for medium-range weatherforecasting reading UK

Climate foresting system version 2 (CFSv2) 1000deg times1000deg Fullycoupled 24 Climate prediction center


uses the May initial data for the prediction of JJAS-erefore it is required to combine hindcast data of May(Mayhc_1982ndash2011) with forecast data for May (Mayfc_2018)-e model is chosen for the period from 1982 to 2011 tomatch with the observation data availability period-e dataare then interpolated to a preferred resolution of 025deg(sim30 km) using the bilinear interpolation method [41] Asthe target spatial resolution of the seasonal prediction is atthe climate zones the rainfall data for both GCMs andobservation are averaged over these zones Furthermorebias correction methods and different MME schemes areapplied to datasets to generate bias-corrected deterministicforecast and probabilistic seasonal forecast for the definedclimatological zones

36 Model Bias Reduction As global models exhibit largebias in simulating seasonal rainfall the bias needs to beremoved or minimized in order to provide skillfulforecast Several bias correction techniques are availablein which the quantile-to-quantile mapping method iswidely used and proven to be effective for the Indiansummer monsoon period [42] -e method removessystematic bias in the GCM simulations using the inverseof cumulative distribution function (CDF) of observed

values (Fob) at the probability corresponding to the en-semble mean output CDF (Fem) at the particular value-en for Ft the bias-corrected forecast (Fbc) would berepresented as

Fbc Fminus 1ob Fem Ft( 1113857( 1113857 (1)

-is study utilized quantile mapping method to removethe systematic bias in the GCMs before they were used in theMME algorithms

37 Development of MME Schemes MME is a process ofstatistically assembling different global models -ere-fore in the MME process n number of global modelswith t number of years of hindcast runs are statisticallyensembled to construct a prediction for the t + 1 year Forexample the current study used 7 GCMs (n 7) with 30years of hindcast runs (t 30) to provide prediction forthe year 2018 (t + 1) A GCM will be considered only if ithas more than one ensemble member Table 1 lists thetotal number of ensemble members available for eachglobal model In this study three different statisticalensemble MME schemes are used (a) arithmetic meanmultimodel ensemble (AM-MME) (b) weighted averagemultimodel ensemble (WA-MME) and (c) supervised

Obtain GCM data

GRIDDED data (binary grid and netcdf data)

Obtain synopticobservation data

IRI (bin) CFSv2 (grib2) ECMWF (nc)ASCII (csv)

Data format conversion usingPython (mat)

Data interpolation to 025deg spatialresolution

Multimodel ensemble (MME)development

Combine hindcast with forecastdata (Y1982ndash2011 + YF)

Systematic bias removal usingquantile mapping

Generation of probabilisticseasonal forecast

Model skillevaluation

ROC score (areaunder the curve)

RMSE CCand SD

Mean areal rainfall over the homogenousregions

Simple arithmetic mean (AM-MME)

Weighted average (WA-MME)

Supervised principal component

Figure 3 Simplified methodology for the model development and forecast customization and generation of MME-based seasonalprobabilistic forecast along with model skill evaluation


principal component regression multimodel ensemble(PCR-MME) -e MME schemes collectively makeuse of all the members to generate the final ensembleforecast

AM-MME is a simple averaging scheme of all indi-vidual model ensembles [20 43] All individual membersof models are assigned with equal weight with the as-sumption that all models considered in this MME schemepredict the seasonal rainfall with uniform skills All modelforecast data are normalized by removing the mean(average calculated for the period 1982ndash2011) from thetime series and the observed interannual trend is added toderive forecast time series -e AM-MME forecast con-structed with bias-corrected forecast data can be repre-sented as

St O +1N

1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (2)

where St MME prediction at time t Fi t ith model forecastat time t Fi climatology of ith model forecastO climatology of observations σFi interannual variationof ith model forecast σ0 interannual variation of obser-vations and N no of models

In the WA-MME scheme a regression coefficient foreach ensemble is obtained for the training phase (t) by usingthe singular value decomposition (SVD) technique -eregression coefficient assigns a weight to each ensemblebased on the training data which is then used in computing arobust weighted average forecast [44] for the time t+ 1 -eWA-MME forecast is constructed with bias-corrected datausing the following equation

St O + ai 1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (3)

where ai regression coefficient obtained by a minimizationprocedure during the training period between modelrsquosforecasts Firsquos and observation O Other variables are thesame as in the AM-MME scheme

-e supervised principal component regression (SPCR)method is primarily used to eliminate presence of anysignificant correlation among individual models [45] It is adimension reductiontransformation technique to minimizethe number of independent variables that describe themaximum variance of all variables -e prediction modelconsidered in this scheme is based on the concept ofprincipal component analysis (PCA) where the principalcomponents (PCs) are calculated after the eigenvector de-composition of a correlation matrix In this method theprincipal components are considered for the regressionprocess [25] -e PCs are selected based on their correlationwith the observation (predictand) unlike the traditional PCRtechnique where they are chosen according to their vari-ances PCs selection based on correlation would be veryuseful for choosing meaning predictors -e SPCR methodensures that predictors with higher correlation are selectedfor regression and forecast generation

38 FOCUS e GUI -e graphical user interface (GUIsee Figure 4) is developed using a combination of Pythonprogramming language for the backend operations such asprocessing data performing statistical analysis and de-veloping statistical methods to generate forecast products-e front end was designed using the Microsoft netframework as a web-based platform -e tool can beaccessed from the following link http20315916146ForecastWebLoginaspx Web data retrieval packageldquowgetrdquo is used at the backend to automatically downloadrequired global forecast dataset from the respective web-sites FOCUS tool has built-in functionalities for dataprocessing combining and interpolation bias correctionand generating ensemble probabilistic forecasts -e toolalso utilized the superensemble technique to generatecombined and reconstructed products with ensemble ofMME forecasts [22] Additionally the tool can performmodel forecast skill evaluation in terms of ROC score andforecast reliability

39 Generation of Probabilistic Forecast One of the bestways to express uncertainty in a consistent and verifiableway is through probability forecasts [14] A probabilityforecast specifies how likely a defined event is to occur [46]In the study GCM ensemble members are used for esti-mation of the probability through the sampling methodand identifying the possible range of forecasts De-terministic forecasts produced from the MMEs are used togenerate probabilistic forecast based on the observed cli-matology meaning with equal (sim33) chance of occur-rence for each tercile category Probability of an event canbe defined with an event Ω as occurrence of X (rainfall) inan interval (x1 x2)

If F (x | β) is the distribution of the predictand Xconditional on a given value of β then the probability thatX lies in an interval (x1 x2) conditional on β is representedas

Px (Ω | β) Prob Xε x1 x2( 11138571113868111386811138681113868 β1113960 1113961 (4)

With Gaussian noise ε the conditional probability can beexpressed as

Px Ω | β σε( 1113857 FN

X2 minus βσε

1113888 1113889 minus FN

X1 minus βσε

1113888 1113889 (5)

where FN is the distribution function of the standard normaldistribution -e probability depends both on the value of βand the standard deviation of ε

As mentioned earlier probabilistic predictions aregenerated for three tercile categories (i) below normal (ii)near normal and (iii) above normal in reference to theobserved climatology and with the notion that each categoryhas equal chance of manifestation Finally deterministicforecast is used as the mean of the forecast distributionwhereas the spread is calculated by the correlation method[29 47] and the corresponding conditional probabilities ofthe events are given by


Px B | β σε( 1113857 FN

minus β minus Xa

σε1113888 1113889

Px A | β σε( 1113857 FN

β minus Xa

σε1113888 1113889

Px N | β σε( 1113857 1 minus Px B | β σε( 1113857 minus Px A | β σε( 1113857

(6)

and FN again is the distribution function of the standardnormal distribution and xa and xb are the boundaries

310 Module for MME Performance Evaluation Severalstandard techniques such as box and whisker plots relative

operating characteristics (ROC) plots and Taylor diagramsare available to evaluate prediction skills of models Box andwhisker plot [48 49] is used to interpret the distribution andvariability ROC is used for evaluating the skill of theprobabilistic forecast performance [46]

311 ROCCurve ROC curves are two-dimensional measureof classification performance and feature the underlyingdistribution of forecasts [50] ROC curves are graphs con-structed with hit rates (Hr) and false alarm rates (Fr) for thethree different tercile categories ROC area skill score(ROCASS) is a validation index about the probabilityforecasts with no value of information ie Hr Fr anddefined by

Figure 4 Screen capture of the Forecast Customization System (FOCUS) GUI developed using Python programming language (MME1 andMME2 refers to the AM-AMME and WA-MME schemes respectively) showing the ROC score generation for the tercile categories


ROCASS equiv 2(ROCA minus 05) (0leROCASSle 1) (7)

ROCASS is the unit for quantifying the forecast where ascore zero to 05 represents no forecast skill a score betweengt05 to 1 indicates a more skillful forecast and any scoresim05 or less suggests no skill [50]

312 Taylor Diagram Taylor diagram [51] provides a con-cise statistical summary of how well patterns match eachother in terms of their correlation coefficient their root-mean-square difference (RMSE) and the ratio of theirvariances -ese plots are used to devise skill scores thatappropriately weight among the various measures of patterncorrespondence

Mathematically the three statistics displayed on a Taylordiagram are related by the following formula

Eprime2

σ2r + σ2t minus 2σrσt ρ (8)

where Eprime centered RMS difference of observation and theprediction ρ correlation coefficient and σrσt variancesof the observation and the prediction

4 Results and Discussion

41 Performance of the Raw GCMs -e ensemble averagedhindcast skill of seven models for the JJAS season overMyanmar for the period 1982 to 2011 is initially diagnosedbased on their RMSE and correlation coefficient as shown inFigure 5 It is seen that all the GCMs exhibit large error forsimulation of rainfall with relatively less correlation with theobservation CFSv2 (039) and ECMWF (025) show bettercorrelation with lesser errors 717 and 444 respectivelyECHAM45 models both constructed analogue SST andCFS-forecasted SST depicted larger RMS errors similar tothe findings of Singh et al [52] for the Indian summermonsoon prediction CCMv36 has better inverse correla-tion (minus 03) but with a very large RMS error (103) It isevident that none of the models can be utilized directly forthe seasonal prediction and requires appropriate errorcorrection and downscaling method to improve the per-formance of these models over Myanmar

42 Bias-Corrected Model and MME Performance overMyanmar -e bias-corrected results for the seven modelsoverMyanmar shows reasonable improvement in RMS errorand better agreement with the observation (Figure 5(b))especially ECHAM45 models which improved from minus 063to 035 (CASST) and minus 067 to 035 (CFSSST) and with RMSerror reduced from 1401 to 68 for both CASSTand CFSSSTECMWF and CFSv2 have improved correlation from 025 to046 and 039 to 050 respectively with no significant im-provement to the RMS error At the same time CCMv36GFDL and COLA exhibited negative impact of the biascorrections and degraded further with increase in RMSerror -ough visible improvement in specific model per-formances over the country is noticed this is still not ad-equate to operationally use them as none of the models areconsistent

Figure 5(c) and Table 2 show the results of the threeMME techniques for Myanmar which indicates significantimprovement with the correlation coefficient going as highas 064 for both WA-MME (MME2) and PCR method whilethe AM-MME (MME1) was slightly less with 05 At thesame time the RMS error reduced to 139 for MME1 and129 for MME2 and PCR respectively -e MMEs per-forming well over Myanmar provides the impetus to gen-erate the climate information for the different climate zonesand examine its performance

43 MME Performance over Climate Zones

431 Quantifying the Observation and Model VariabilityFigure 6 shows the variability of the observed rainfall in-dividual model outputs that are bias corrected over the sixclimate zones In general the individual models are not ableto capture the variability in the observation whereas theMMEs captured the variability better than the individualmodels Few models such as ECMWF and CFSv2 performbetter in shan region and dry zones (Figures 6(a) and 6(c))as the rainfall variability in the region itself is minimumwhen compared to the coastal mountain and southernregions (Figures 6(b) 6(d) and 6(e)) -e way coupledmodels are designed and parameterized the performancevaries from region to region and from season to season Forinstance the predictability of CFSv2 and GFDL models overIndian region during JJAS months is much better whencompared to other models such as ECMWF and CFSSST-ough the predictability skills of ECMWF are lower for theJJAS season it performs well over the Indian region duringthe winter season [53] In this study CFSv2 performs wellover the shan region and dry zones but GFDL predictabilityskills are low Further investigation on MME schemes overthe study region indicated that the AM-MME scheme is notable to enhance the overall skill of the forecast mainly be-cause an ensemble member with higher skill gets the sameweight as a member with lower skill [16] However the WA-MME method performs better as weights were calculatedand assigned to each ensemble member -e climatology forthe same is shown in Figure 7

44 Correlation Coefficients and RMSE Taylor diagramswere plotted for the different climate zones to quantify theregionwise skill of the MME methods as shown in Figure 8-e results suggest that the WA-MME and PCR modelsshow enhanced skill over the delta coastal and dry zoneswhile no significant improvement is observed over theeastern and northern zones -e AM-MME scheme per-formed better over the coastal and delta regions most likelybecause the individual ensembles agree with each otherwhen compared to regions where the individual ensemblesare not in agreement and the AM-MME performance ispoor Overall all three MME schemes perform better overdelta region meaning they depict the mean rainfall rea-sonably well -e observed temporal variability for the delta(21) coastal (24) and southern (36) regions is the highestwhile for dry (06) north (15) and east (07) regions


ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC

Figure 5 JJAS performance comparison of the raw models with the bias-corrected (BC) models for the overall Myanmar (a) Raw models(b) Bias-corrected models (c) MMEs

Table 2 Correlation coefficients root mean square error and standard deviation for the JJAS season for the six identified zones

MethodszonesAM-MME WA-MME PCR-MME

CC SD RMSE CC SD RMSE CC SD RMSEEast 032 053 069 036 066 075 minus 015 023 073North minus 003 092 179 011 087 166 011 066 158Dry 002 044 075 046 05 059 044 035 057Coastal 013 2 294 035 181 249 015 139 263South 048 28 321 057 365 324 056 158 29Delta 053 165 176 064 202 168 068 114 148

2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day

YearObs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )

Figure 6 JJAS rainfall variability in observed (Obs observed) and various model data (M1-AM-MMEM2-WA-MMEM3-PCRMMEM4-CCMv36 M5-ECHAM-CASST M6-ECHAM-CFSSST M7-CFSv2 M8-COLA M9-GFDL M10-ECMWF) for six zones of Myanmar(a) shan (b) north (c) coastal (d) dry (e) south and (f) delta

0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta

Figure 7 Observed and modeled rainfall during June to September period over the six climatological zones in Myanmar


27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME

00 03 06 09 12Standard deviation

15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued


variability is the lowest Among all the models and methodsWA-MME scheme (Figure 8) captured the observed vari-ation well except the northern zone

45 Measuring the Probabilistic Forecast Skill -e ROCscores shown in Table 3 suggest that probabilistic forecastgenerated with the WA-MME scheme showed better skillsamong all three tercile categories below normal (078)normal (083) and above normal (083) for overall Myan-mar In general all three schemes were able to predict theabove normal rainfall category very well but the pre-dictability skills for the ldquonear normalrdquo rainfall category ispoor especially for AM-MME and PCR-MME Table 3shows the ROC scores of the climate zones and suggeststhat the models are most skillful over the delta region fol-lowed by the southern and coastal regions though it issatisfactory over the dry zone with PCR-MME performingbetter However the skills are very low for the eastern andnorthern regime when compared to other zones-e reasonfor poor skill over the northern mountainous region or theeastern shan state could be mainly due to unavailability ofgood quality and sufficient number of observation pointswhich makes it difficult to define the predictand well forthese regions as Kar et al [47] described similar results overIndian monsoon prediction that the prediction skill is im-proved when a higher quality training dataset is deployed forthe evaluation of the multimodel bias statistics [47] On theother hand it could also be due to failure of the globalmodels to capture the rainfall variability over the high-el-evation region over Myanmar which spreads over thenorthern to eastern zones It is important to notice that the

MME methods are skillful in predicting the lower (belownormal) and upper (above normal) tercile categories betterthan the normal category which is a positive sign as oftenabove and below normal rainfall categories are crucial to beknown for carrying out seasonal preparedness measuresrather than the normal rainfall category

5 Conclusion

Agricultural system is predominantly dependent on skillfulweather forecast with a longer lead time preferably atseasonal scale Critical decision making entails higher risksin the absence of such forecast systems -us the forecastcustomization system (FOCUS) was developed to addressthis issue and it provides an enabling environment to themeteorological service in Myanmar with a standardizedplatform to access and evaluate various global models with astreamlined approach -e tool is developed using free andopen-source scripting language Python and Microsoftrsquosnet framework -ree standard MME methods were de-veloped and integrated into the FOCUS platform withcomponents to interpolate and combine global modelhindcast data with forecast -e MME-based forecast wasthen generated for the defined climate zones for the JJASperiod

To quantify uncertainty the MME outputs were eval-uated for (i) accuracy with standard verification methodsusing RMSE and correlation coefficient and (ii) the pre-dictability skill with ROC scores -e results suggested thatby utilizing the MME methods the performance of forecastwas significantly improved over the country and over theJJAS period in terms of predictability skill Among the

North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )

Figure 8 Correlation coefficient root mean square error and standard deviation for the JJAS period for all six climate zones (a) delta zone(b) southern zone (c) coastal zone (d) dry zone (e) northern zone (f ) eastern shan zone inMyanmar Reference point denotes the standarddeviation for observation for each zone respectively


MMEs the weighted ensemble averaging method(ROC 083) has slight advantage over the simple arithmeticaveraging method (ROC 058) in terms of predictabilityskills for the normal tercile category -e principal com-ponent regression method is performing well over the high-rainfall southern (ROC 07) and delta regions(ROC 085) for prediction of the upper terciles as well asfor the lower terciles with ROC 078 (southern region) andROC 078 (delta region) Overall it is evident that MMEperformance is satisfactory and especially both WA-MMEand PCR-MME could be considered with high reliabilityfor generating seasonal forecast for the high rainfall zones inthe country Again it is worth noticing that the model ishighly reliable for predictions of upper and lower terciles butfailed to accurately predict the normal rainfall category

FOCUS tool uses well-defined methods and has thepotential to be scaled up further for other countries in theregion with use of more advanced statistical and compu-tational techniques However it is necessary for the tool tohave high-quality rainfall observation datasets with adequatespatial and temporal coverage In conclusion the MME-based approach incorporated in a user-friendly interfacewould be a very useful tool for generating skillful seasonalforecast for the tropical region Again an improved seasonalforecast enables effective decision making in all climate-sensitive sectors such as the agriculture and water resources

Data Availability

-e GCM data used to support the findings of this study areavailable from the corresponding author upon requestHowever the ownership of the observation datasets used tosupport the findings are with the Department of Meteo-rology and Hydrology Myanmar

Additional Points

Highlights (i) Forecast customization system (FOCUS) isdeveloped with user-friendly graphical user interface togenerate improved ensemble seasonal forecast and evaluateindividual and ensemble forecast performance of variousglobal seasonal prediction model outputs in a singleplatform to identify an appropriate operational seasonalforecasting scheme for Myanmar (ii) Statistical skills varyspatially however the multimodel ensemble scheme hasbetter predictability skills in simulating the rainfall

variability over different climatological regions of Myan-mar as compared to individual models (iii) Consideringbetter performance of weighted average multimodel andprincipal component analysis ensemble over Myanmarthese schemes could be used by meteorological services ingenerating regular operational seasonal forecast for agri-cultural planning and risk anticipation

Conflicts of Interest

-e authors declare that there are no conflicts of interestregarding the publication of this paper

References

[1] N S Roy and S Kaur ldquoClimatology of monsoon rains ofMyanmar (Burma)rdquo International Journal of Climatologyvol 20 no 8 pp 913ndash928 2000

[2] S S Roy and N S Roy ldquoInfluence of pacific decadal oscil-lation and El Nintildeo Southern oscillation on the summermonsoon precipitation in Myanmarrdquo International Journal ofClimatology vol 31 no 1 pp 14ndash21 2011

[3] R DrsquoArrigo J Palmer C C Ummenhofer N N Kyaw andP Krusic ldquo-ree centuries of Myanmar monsoon climatevariability inferred from teak tree ringsrdquoGeophysical ResearchLetters vol 38 no 24 2011

[4] R DrsquoArrigo and C C Ummenhofer ldquo-e climate ofMyanmar evidence for effects of the pacific decadal oscilla-tionrdquo International Journal of Climatology vol 35 no 4pp 634ndash640 2015

[5] Z M M Sein B A Ogwang V Ongoma F K Ogou andK Batebana ldquoInter-annual variability of summer monsoonrainfall over Myanmar in relation to IOD and ENSOrdquo Journalof Environmental and Agricultural Sciences vol 4 pp 28ndash362015

[6] R R Policarpio and M Sheinkman State of Climate In-formation Products and Services for Agriculture and FoodSecurity in Myanmar Agriculture and Food SecurityCopenhagen Denmark 2015

[7] RIMES ldquo-e 10th monsoon forum briefrdquo Activity reportRegional Integrated Multi-hazard Early Warning SystemKhlong Nueng -ailand 2013

[8] RIMES ldquo-e 11th RIMES monsoon forumrdquo Activity reportRegional Integrated Multi-hazard Early Warning SystemKhlong Nueng -ailand 2013


Table 3 ROC scores for three tercile categories over the six identified climate zones for the three MME schemes

Tercileregions MMEs Shan North Dry Coastal South Delta Myanmar

Below normalAM 06 04 055 048 063 063 078WA 055 055 06 063 07 063 078PCR 07 063 06 055 078 078 075

NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055

Above normalAM 052 033 045 055 063 07 08WA 055 048 07 07 06 07 083PCR 048 04 063 055 07 085 08


[10] T Yi W M Hla and A K Htun ldquoDrought conditions andmanagement strategies in Myanmarrdquo Report of the De-partment of Meteorology and Hydrology vol 9 2013

[11] E Lee T N Chase and B Rajagopalan ldquoHighly improvedpredictive skill in the forecasting of the East Asian summermonsoonrdquo Water Resources Research vol 44 no 10 2008

[12] J Shanmugasundaram and E Lee ldquoOceanic and atmosphericconditions associated with the pentad rainfall over thesoutheastern peninsular India during the North-East IndianMonsoon seasonrdquo Dynamics of Atmospheres and Oceansvol 81 pp 1ndash14 2018

[13] Y He and E Lee ldquoEmpirical relationships of sea surfacetemperature and vegetation activity with summer rainfallvariability over the Sahelrdquo Earth Interactions vol 20 no 6pp 1ndash18 2016

[14] J Slingo and T Palmer ldquoUncertainty in weather and climatepredictionrdquo Philosophical Transactions of the Royal Society AMathematical Physical and Engineering Sciences vol 369no 1956 pp 4751ndash4767 2011

[15] E Kalnay Atmospheric Modeling Data Assimilation andPredictability Cambridge University Press Cambridge UK2003

[16] N Acharya S Chattopadhyay U C Mohanty and K GhoshldquoPrediction of Indian summer monsoon rainfall a weightedmulti-model ensemble to enhance probabilistic forecastskillsrdquoMeteorological Applications vol 21 no 3 pp 724ndash7322014

[17] F Molteni R Buizza C Marsigli A Montani F Nerozzi andT Paccagnella ldquoA strategy for high-resolution ensembleprediction I definition of representative members andglobal-model experimentsrdquo Quarterly Journal of the RoyalMeteorological Society vol 127 no 576 pp 2069ndash2094 2001

[18] R Buizza P L Houtekamer G Pellerin Z Toth Y Zhu andM Wei ldquoA comparison of the ECMWF MSC and NCEPglobal ensemble prediction systemsrdquo Monthly Weather Re-view vol 133 no 5 pp 1076ndash1097 2005

[19] T N Palmer A Alessandri U Andersen et al ldquoDevelopmentof a European multimodel ensemble system for seasonal-to-interannual prediction (DEMETER)rdquo Bulletin of the Ameri-can Meteorological Society vol 85 no 6 pp 853ndash872 2004

[20] R Hagedorn F J Doblas-Reyes and T N Palmer ldquo-erationale behind the success of multi-model ensembles inseasonal forecastingmdashI Basic conceptrdquo Tellus A DynamicMeteorology and Oceanography vol 57 pp 280ndash289 2005

[21] T N Palmer F J Doblas-Reyes A Weisheimer G J ShuttsJ Berner and J M Murphy ldquoTowards the probabilistic earth-system modelrdquo 2008 httpsarxivorgabs08121074

[22] T N Krishnamurti C M Kishtawal Z Zhang et alldquoMultimodel ensemble forecasts for weather and seasonalclimaterdquo Journal of Climate vol 13 no 23 pp 4196ndash42162000

[23] A P Weigel M A Liniger and C Appenzeller ldquo-e discreteBrier and ranked probability skill scoresrdquo Monthly WeatherReview vol 135 no 1 pp 118ndash124 2007

[24] X Zhi H Qi Y Bai and C Lin ldquoA comparison of three kindsof multimodel ensemble forecast techniques based on theTIGGE datardquo Acta Meteorologica Sinica vol 26 no 1pp 41ndash51 2012

[25] U C Mohanty N Acharya A Singh et al ldquoReal-time ex-perimental extended range forecast system for Indian summermonsoon rainfall a case study for monsoon 2011rdquo CurrentScience vol 104 no 7 pp 856ndash870 2013

[26] B A Cash J V Manganello and J L Kinter ldquoEvaluation ofNMME temperature and precipitation bias and forecast skill

for South Asiardquo Climate Dynamics vol 53 pp 7363ndash73802019

[27] B Rajagopalan U Lall and S E Zebiak ldquoCategorical climateforecasts through regularization and optimal combination ofmultiple GCM ensemblesrdquoMonthlyWeather Review vol 130no 7 pp 1792ndash1811 2002

[28] N Acharya S C Kar M A Kulkarni U C Mohanty andL N Sahoo ldquoMulti-model ensemble schemes for predictingnortheast monsoon rainfall over peninsular Indiardquo Journal ofEarth System Science vol 120 no 5 pp 795ndash805 2011

[29] M K Tippett A G Barnston and A W Robertson ldquoEsti-mation of seasonal precipitation tercile-based categoricalprobabilities from ensemblesrdquo Journal of Climate vol 20no 10 pp 2210ndash2228 2007

[30] S J Mason and M K Tippett Climate PredictabilityTool 2016 httpsacademiccommonscolumbiaedudoi107916D8668DCW

[31] APCC CLimate Information ToolKit 2008 httpclikapcc21org

[32] SCOPIC Seasonal Climate Outlook for the Pacific IslandCountries 2005 httpcosppacbomgovauproducts-and-servicesseasonal-climate-outlooks-in-pacific-island-countries

[33] A Cottrill A Charles and Y Kuleshov ldquoAn analysis ofseasonal forecasts from POAMA and SCOPIC in the Pacificregionrdquo in Proceedings of the EGU General Assembly Con-ference Abstracts Vienna Austria April 2013

[34] L L Aung E E Zin P -eing et al Myanmar Climate Report2015 httpswwwmetnopublikasjonermet-report_attachmentdownloadMyanmarClimateReportFINAL11Oct2017pdf

[35] W D Collins J Wang J T Kiehl G J Zhang D I Cooperand W E Eichinger ldquoComparison of tropical ocean-atmo-sphere fluxes with the NCAR community climate modelCCM3rdquo Journal of Climate vol 10 no 12 pp 3047ndash30581997

[36] B P Kirtman D Min J M Infanti et al ldquo-e NorthAmerican multimodel ensemble phase-1 seasonal-to-in-terannual prediction phase-2 toward developing intra-seasonal predictionrdquo Bulletin of the American MeteorologicalSociety vol 95 no 4 pp 585ndash601 2014

[37] S K Saha S Pokhrel K Salunke et al ldquoPotential pre-dictability of Indian summer monsoon rainfall in NCEPCFSv2rdquo Journal of Advances inModeling Earth Systems vol 8no 1 pp 96ndash120 2016

[38] H Van den Dool J Huang and Y Fan ldquoPerformance andanalysis of the constructed analogue method applied to USsoil moisture over 1981ndash2001rdquo Journal of Geophysical Re-search Atmospheres vol 108 no D16 2003

[39] M Blumenthal M Bell J del Corral R Cousin andI Khomyakov ldquoIRI Data Library enhancing accessibility ofclimate knowledgerdquo Earth Perspectives vol 1 no 1 p 192014

[40] World Meteorological Organization Guidelines on QualityManagement Procedures and Practices for Public WeatherServices PWS-11 WMOTD No 1256 Geneva Switzerland2005

[41] G G Dahlquist ldquoA special stability problem for linearmultistep methodsrdquo Bit vol 3 no 1 pp 27ndash43 1963

[42] N Acharya S Chattopadhyay U CMohanty S K Dash andL N Sahoo ldquoOn the bias correction of general circulationmodel output for Indian summer monsoonrdquo MeteorologicalApplications vol 20 no 3 pp 349ndash356 2013

[43] T DelSole J Nattala and M K Tippett ldquoSkill improvementfrom increased ensemble size and model diversityrdquo Geo-physical Research Letters vol 41 no 20 pp 7331ndash7342 2014


[44] W T Yun L Stefanova and T N Krishnamurti ldquoIm-provement of the multimodel superensemble technique forseasonal forecastsrdquo Journal of Climate vol 16 no 22pp 3834ndash3840 2003

[45] B D Fekedulegn J J Colbert and M E Schuckers Copingwith Multicollinearity An Example on Application of PrincipalComponents Regression in Dendroecology US Department ofAgriculture Forest Service Northeastern Research StationNewton Square PA USA 2002

[46] Metoffice nd Probability Forecasts httpresearchmetofficegovukresearchnwpensembleprobabilityhtml

[47] S C Kar N Acharya U C Mohanty and M A KulkarnildquoSkill of monthly rainfall forecasts over India using multi-model ensemble schemesrdquo International Journal of Clima-tology vol 32 no 8 pp 1271ndash1286 2012

[48] R McGill J W Tukey and W A Larsen ldquoVariations of boxplotsrdquo e American Statistician vol 32 no 1 pp 12ndash161978

[49] J W Tukey ldquoAnalyzing data sanctification or detectiveworkrdquo American Psychologist vol 24 p 8391 1969

[50] C Marzban ldquo-e ROC curve and the area under it as per-formance measuresrdquo Weather and Forecasting vol 19 no 6pp 1106ndash1114 2004

[51] K E Taylor ldquoSummarizing multiple aspects of model per-formance in a single diagramrdquo Journal of Geophysical Re-search Atmospheres vol 106 no D7 pp 7183ndash7192 2001

[52] A Singh M A Kulkarni U C Mohanty S C KarA W Robertson and G Mishra ldquoPrediction of Indiansummer monsoon rainfall (ISMR) using canonical correlationanalysis of global circulation model productsrdquoMeteorologicalApplications vol 19 no 2 pp 179ndash188 2012

[53] A Nair G Singh and U C Mohanty ldquoPrediction of monthlysummer monsoon rainfall using global climate modelsthrough artificial neural network techniquerdquo Pure and Ap-plied Geophysics vol 175 no 1 pp 403ndash419 2018


Hindawiwwwhindawicom Volume 2018

Journal of

ChemistryArchaeaHindawiwwwhindawicom Volume 2018

Marine BiologyJournal of


BiodiversityInternational Journal of


EcologyInternational Journal of


Hindawiwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2018

Forestry ResearchInternational Journal of



International Journal of

Geophysics

Environmental and Public Health

Journal of




Microbiology


Public Health Advances in

AgricultureAdvances in


Agronomy




MeteorologyAdvances in

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018Hindawiwwwhindawicom Volume 2018

ChemistryAdvances in

ScienticaHindawiwwwhindawicom Volume 2018


Geological ResearchJournal of

Analytical ChemistryInternational Journal of


Submit your manuscripts atwwwhindawicom

depend on the years with similar ENSO phases in the pastIt was recommended that an improved seasonal forecastingsystem is required despite the traditional method followedby DMH during a user interaction forum conducted byDMH every year [8] Furthermore capacity self-assessmentexercise conducted by RIMES [9] identified the need fordevelopment of a standard platform in order to assistDMH in generating seasonal climate forecast and assist inanalyzing different global models Moreover need forempirical studies focusing on rainfall variability andforecasting for operational applications in agriculture andwater resource sector was emphasized in Asia [10ndash12] andAfrica [13]

Global climate models (GCMs) are useful tools forpredicting seasonal climate however there is large un-certainty in its prediction mainly because of the as-sumptions made in the initial atmospheric state [14] Tosimulate and capture these uncertainties in the pre-dictions GCMs are processed with different initial con-ditions to generate multiple forecast members called theensembles [15 16] Multimodel ensemble (MME) is aprocess where ensemble members of one GCM are sta-tistically assembled with another GCM or a set of GCMs[17ndash21] -e MME approach has increasingly demon-strated better prediction skills over the tropical Asianregion in long-range forecasting when compared to in-dividual model performances [20 22ndash24] For instancethe MME system developed over India for monthlysea-sonal prediction in real time during the South Asianmonsoon season exhibited satisfactory predictions [25]Similarly North American MMEs (NMME) showed lowersystematic error and higher forecast skills compared to theindividual members over Southeast Asian region [26]

MME schemes can be developed using various statisticalmethods (1) by simply taking mean of all ensembles withassigning equal weight to individual ensemble members [20]or (2) by assigning higher weightage to the statisticallysignificant members of GCMs according to their perfor-mance over the hindcast period [20 24 27ndash29] or usingcomplex neural network algorithms It is however a well-established concept that MMEs would be useful schemes forgenerating improved seasonal outlook But so far no at-tempts were made to develop a long-range prediction systemfor Myanmar using such type of advanced techniques Toolssuch as the Climate Prediction Tool (CPT) [30] ClimateInformation Toolkit (CLIK) [31] and Seasonal ClimateOutlook in the Pacific Island Countries (SCOPIC) [32] havethe functionalities to perform statistical analysis with climatedata but are limited in terms of their utility For instanceCLIK is useful for providing predictions at regional scale butnot at the spatial scale suggested in this current study CPThas the capability to do prediction specific to locations butcannot perform MME-based predictions SCOPIC isdesigned to predict seasonal climate only for the PacificIsland countries and not yet applicable for other regions[33]

-e objective of the present work is to overcome theabovementioned limitations and to develop a web-basedgraphical user interface (web-GUI) forecast customization

system tailored for national use -e tool allows generationof monthly and seasonal climate outlooks using the MMEtechniques and assist in evaluating the performance ofoutlooks -e tool is also capable of providing the outlooksfor the defined climatological zones in Myanmar -emethod will be described in the successive section

2 Study Area

21 Zone Classification Myanmar is geographically situ-ated in Southeast Asia between latitudes 09deg 32prime N and 28deg31prime N and longitudes 9deg 10prime E and 101deg 11prime E Myanmar isclimatologically divided into six major zones (Figures 1(a)and 1(b) [10]) (1) central dry zone which has the lowestaverage annual rainfall and intense agricultural practices(2) eastern zone (shan) (3) northern zone which is mostlywith high terrain and forest areas (4) coastal (Rakhine)(5) delta zone (Ayeyarwady region) and (6) southernzone which receives the maximum annual rainfall zones-is also synchronized well with classifications done inthe past in references [1 5 20] -ese classifications arebased on the average annual rainfall in the country(Figure 2) agroecological zoning and seasonal rainfallpatterns respectively

22 Climatology of Myanmar -e annual rainfall cycle inFigure 2 shows distribution of rainfall mostly concentratedover the JJAS period which is mainly due to the influenceof the southwest monsoon (Figure 1(c)) -e monsoononset is marked during May peak during August andwithdrawal towards the end of September [34] -e spatialdistribution of rainfall however varies significantly duringthis period over all the zones -e central dry zone receivesthe lowest amount of seasonal rainfall while the southernregion receives the highest amount [34] As JJAS is majorrainfall season for all the zones of the country this studyfocused on investigating characteristics of rainfall over JJASand also predicting seasonal rainfall for its operationalapplication in agricultural and water resources manage-ment sector

3 Data and Methods

31 GCM Data Hindcast rainfall data from seven GCMs(listed in Table 1) are obtained for the period 1982ndash2011 Allthe GCM hindcast datasets are at monthly temporal scaleFour fully coupled GCMs namely National Center forEnvironment Predictionrsquos Coupled Forecast System Modelversion 2 (NCEP CFSv2) [37] Geophysical Fluid DynamicsLaboratory (GFDL) COLA and the System 4 (ECMWFTechnical Memorandum 2011) European Center for Me-dium-Range Weather Forecast (ECMWF) are used in thestudy -e CFSv2 model has a spectral triangular truncationof 126 waves (T126) horizontally (equivalent to nearly a100 km grid resolution) and a finite differencing verticallywith 64 sigma-pressure hybrid layers -e atmosphericcomponent in this model is the Global Forecast System (GFS2009) while Geophysical Fluid Dynamics LaboratoryModular Ocean Model 4 (GFDL MOM4) is considered as




0

5

10

15

20

Jan

Feb

Mar

Apr


May

June July

Aug

Sep

Oct

Nov Dec

Rain

fall

(mm

day

)

Month


200ndash300


500ndash600

600ndash700

700ndash800

800ndash1100

Elevation (m)

(a) (b)

(c)

25ndash50

50ndash100

100ndash200

200ndash400

400ndash600

600ndash800

lt25 800ndash1000

1000ndash1500



0 200km 0 200km









0

5

10

15

20

25

30

35


infa

ll in

mm

Months

Monthly climatology

DryShanNorthDelta

CoastalSouthOverall




















Obtain GCM data












RMSE CCand SD









St O +1N

1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (2)



St O + ai 1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (3)






Px (Ω | β) Prob Xε x1 x2( 11138571113868111386811138681113868 β1113960 1113961 (4)


Px Ω | β σε( 1113857 FN

X2 minus βσε

1113888 1113889 minus FN

X1 minus βσε

1113888 1113889 (5)




Px B | β σε( 1113857 FN

minus β minus Xa

σε1113888 1113889

Px A | β σε( 1113857 FN

β minus Xa

σε1113888 1113889


(6)











Eprime2











ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC





2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy

















0

5

10

15

20

Jan

Feb

Mar

Apr


May

June July

Aug

Sep

Oct

Nov Dec

Rain

fall

(mm

day

)

Month


200ndash300


500ndash600

600ndash700

700ndash800

800ndash1100

Elevation (m)

(a) (b)

(c)

25ndash50

50ndash100

100ndash200

200ndash400

400ndash600

600ndash800

lt25 800ndash1000

1000ndash1500



0 200km 0 200km









0

5

10

15

20

25

30

35


infa

ll in

mm

Months

Monthly climatology

DryShanNorthDelta

CoastalSouthOverall




















Obtain GCM data












RMSE CCand SD









St O +1N

1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (2)



St O + ai 1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (3)






Px (Ω | β) Prob Xε x1 x2( 11138571113868111386811138681113868 β1113960 1113961 (4)


Px Ω | β σε( 1113857 FN

X2 minus βσε

1113888 1113889 minus FN

X1 minus βσε

1113888 1113889 (5)




Px B | β σε( 1113857 FN

minus β minus Xa

σε1113888 1113889

Px A | β σε( 1113857 FN

β minus Xa

σε1113888 1113889


(6)











Eprime2











ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC





2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy





















0

5

10

15

20

25

30

35


infa

ll in

mm

Months

Monthly climatology

DryShanNorthDelta

CoastalSouthOverall




















Obtain GCM data












RMSE CCand SD









St O +1N

1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (2)



St O + ai 1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (3)






Px (Ω | β) Prob Xε x1 x2( 11138571113868111386811138681113868 β1113960 1113961 (4)


Px Ω | β σε( 1113857 FN

X2 minus βσε

1113888 1113889 minus FN

X1 minus βσε

1113888 1113889 (5)




Px B | β σε( 1113857 FN

minus β minus Xa

σε1113888 1113889

Px A | β σε( 1113857 FN

β minus Xa

σε1113888 1113889


(6)











Eprime2











ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC





2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy





















Obtain GCM data












RMSE CCand SD









St O +1N

1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (2)



St O + ai 1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (3)






Px (Ω | β) Prob Xε x1 x2( 11138571113868111386811138681113868 β1113960 1113961 (4)


Px Ω | β σε( 1113857 FN

X2 minus βσε

1113888 1113889 minus FN

X1 minus βσε

1113888 1113889 (5)




Px B | β σε( 1113857 FN

minus β minus Xa

σε1113888 1113889

Px A | β σε( 1113857 FN

β minus Xa

σε1113888 1113889


(6)











Eprime2











ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC





2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy

















St O +1N

1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (2)



St O + ai 1113944

N

i1

Fit minus Fi

σFi

1113888 1113889⎡⎣ ⎤⎦σ0 (3)






Px (Ω | β) Prob Xε x1 x2( 11138571113868111386811138681113868 β1113960 1113961 (4)


Px Ω | β σε( 1113857 FN

X2 minus βσε

1113888 1113889 minus FN

X1 minus βσε

1113888 1113889 (5)




Px B | β σε( 1113857 FN

minus β minus Xa

σε1113888 1113889

Px A | β σε( 1113857 FN

β minus Xa

σε1113888 1113889


(6)











Eprime2











ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC





2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy















Px B | β σε( 1113857 FN

minus β minus Xa

σε1113888 1113889

Px A | β σε( 1113857 FN

β minus Xa

σε1113888 1113889


(6)











Eprime2











ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC





2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy



















Eprime2











ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC





2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy















ndash08

ndash06

ndash04

ndash02

0

02

04

06

08

0

2

4

6

8

10

12

14

16

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

CCM

3v6

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

AM

-MM

E

WA

-MM

E

PCRM

ME

RAW

(a) (b) (c)

BC MME

Corr

elat

ion

RMSE

STD DEVRMSECC





2

4

6

8

10

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(a)

Year

5

10

15

20

25

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(b)

Figure 6 Continued


Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy















Year

2

4

6

8

10

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(c)

5

10

15

20

25

Rain

fall

mm

day


(d)

0

10

20

30

40

Year

Rain

fall

mm

day

Obs M1 M2 M3 M4 M5 M6 M7 M8 M9 M10

(e)

5

10

15

20

25

Rain

fall

mm

day


(f )


0

5

10

15

20

25

30

Obs

erve

d

AM

-MM

E

WA

-MM

E

PCR

CCSM

3

EC-C

ASS

T

EC-C

FSSS

T

CFSv

2

COLA

GFD

L

ECM

WF

Rain

fall

in m

md

ay

ShanNorthDry

CoastalSouthDelta



27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy















27

00

Delta

01 02 03 0405

06

07Correlation

08

09095

099

24

21

18

152

32

1

1

12

09 3

12

06

03

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


15 18 21 24 27

(a)

South00 01 02 03 04

0506

07Correlation

08

09095

099

4

6

3

2

3

1

2

48

42

36

30

24

18

12

06

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


30 36 42 48

(b)

Coastal00 01 02 03 04

0506

07Correlation

08

09095

099

4

3

2

2

1

36

32

28

24

20

16

12

08

04

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


20 24 28 32 36

3

12

(c)

Dry00 01 02 03 04

0506

07Correlation

08

09095

099

1

1

0

0

09

08

07

06

05

04

03

02

01

00

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


05 06 07 08 09

3

1 2

(d)

Figure 8 Continued





5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy


















5 Conclusion



North00 01 02 03 04

0506

07Correlation

08

09095

099

225

200

175

150

125

100

075

050

025

000

1ReferenceAM-MME 3

2 WA-MMEPCR-MME


125 150 175 200 225

3

1 2

2

2

0

1

(e)

East

09

00 01 02 03 0405

0607

08

09095

099

08

07

06

05

04

03

02

01

00

3


ReferenceAM-MME

WA-MMEPCR-MME

05 06 07 08 09

1

2Correlation

1

1

0

0

312

(f )





Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy

















Data Availability


Additional Points





References













NormalAM 04 033 055 04 048 055 058WA 048 048 055 063 06 04 083PCR 063 04 06 05 063 063 055



















































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy































































Journal of










Volume 2018





Geophysics


Journal of




Microbiology





Agronomy















forecast customization system (focus): a multimodel...

Documents