Abstract:
Sri Lanka is heavily dependent on both rain-fed and irrigated agriculture and thus irrigation
has had a unique contribution towards country`s agro economy from history to this date. The
established patterns of rainfall in different parts of the country have changed and the demand
for agricultural water has to be balanced with the municipal and industrial water demand.
The improved procedures for estimating agricultural water requirements both for irrigation
and rain-fed agriculture have become an important research particularly due to erratic
rainfall patterns and inadequate water resources in dry season. The aim of this study is
therefore to develop time series models to predict weekly reference evapotranspiration (ETo)
for Yala and Maha seasons in Polonnaruwa district using climate data from 2010 to 2015. As
actual evapotranspiration is not available, those values on weekly basis were computed using
Pan Evaporation method based on relative humidity, wind speed and pan evaporation. 85%
of the data computed were used for training and balance of 15% was kept for validation. The
weekly evapotranspiration during Yala varied from 2.23mm (6 – 12 September 2013 ) to
5.37mm (1 – 7 May 2015) with mean of 3.62mm and SD of 0.53 and that during Maha
varied from 0.76mm (21 – 27 December 2012) to 5.56mm (17 – 23 October 2014) with
mean of 2.29mm and SD of 0.85. Both series were able to make stationary by taking one
short-term difference and one long-term difference with the length of 26. The identified best
fitted ARIMA models for Yala and Maha weekly evapotranspiration were SARIMA (1,1,1)
(1,1,1)26. The errors produced by two models were found to be white noise. The percentage
errors in both models for validation data set were within the range of ± 3% and it was found
that the correlations between observed and predicted values for Yala (r=0.90) and for Maha
(r = 0.88) were highly significant (p<0.05). The best fitted model identified for the pooled
weekly series was SARIMA (0,1,2) (0,1,1)52. Though the errors found to be satisfied all the
diagnostic tests, the percentage error was higher in the combined model than the
corresponding values for two separate models. Therefore, it is recommended to use the
developed separate models to forecast ET0 on short-term or long-term basis which will be
useful for the appropriate water management for real time irrigation scheduling in Dry Zone
of Sri Lanka.These models can also be used for estimating irrigation water requirements for
different crops. It is suggested to use Artificial Neural Network (ANN) techniques to
improve the accuracy of the developed models.
