MARIMA for Agriculture water consumption

# R 
library(Metrics)
library(Metrics)
library(scales)
library(leaps)
library(caTools)
library(car)
library(RSEIS)
library(devtools)
library(vars)
library(forecast)
library(TSA)
library(remotes)
library(marima)

#REad and split data
MyData1 <- read.csv(file="data for agriculture water prediction.csv",sep=",")
training_data=MyData1[11:47,2:9]
training_data1=scale(training_data)#training data
validation_data=MyData1[48:57,2:9]
validation_data1=scale(validation_data)#training data
MyDataT=MyData1[11:47,1]
MyDataV=MyData1[48:57,1]
Testdata=MyData[58:59,]

mse3=data.frame()
ar<-(7) 
ma<-(3) 
x=dim(training_data) 
Mod <- define.model(kvar=x[2], ar=ar, ma=ma,rem.var = c(3)) 

arp<-Mod$ar.pattern
map<-Mod$ma.pattern
#short.form(arp)
#short.form(map)

# Now call marima
Model <- marima(training_data1, ar.pattern=arp, ma.pattern=map, penalty=2) #penalty2=AIC
#summary(Model)
ar.model<-Model$ar.estimates
ma.model<-Model$ma.estimates

#forecasting
nstart=1
nstep1=nrow(validation_data)
nstep2=nrow(training_data)
forecast=arma.forecast(training_data1,Model,nstart=1, nstep=nstep2-1) # try with differenced data later
forecast1=arma.forecast(validation_data1,Model,nstart=1, nstep=nstep1-1) # try with differenced data later
par(mfrow=c(1,2))

# for training data
My4sc=(training_data[,8])
vector1= c(min(My4sc),max(My4sc))
ReSclActual=(rescale(training_data1[,8], to=vector1))
ReSclforcast=(rescale(forecast$forecasts[8, ], to=vector1))
mse1=mse(ReSclActual,ReSclforcast)
y=(MyDataT)
plot(y,ReSclforcast,type='l', xlab='Year', ylab='AgriWC', main='AgriWC')
lines(y, ReSclActual, type='p')

#for validation data
My4sc1=(validation_data[,8])
vector2= c(min(My4sc1),max(My4sc1))
ReSclActual1=(rescale(validation_data1[,8], to=vector2))
y1=(MyDataV)
ReSclforcast1=(rescale(forecast1$forecasts[8, ], to=vector2))
plot((y1),(ReSclforcast1),type='l', xlab='Year', ylab='AgriWC', main='AgriWC')
lines(y1, ReSclActual1, type='p')
mse2=mse(ReSclActual1,ReSclforcast1)

print("mse for training data")
print (mse1)