Performance of neural network model in estimating daily maximum and minimum temperature in Jaffna

Abstract

Serially complete climate data are required as input for climate dependent models such as crop and insect development, soil erosion and evapotranspiration (DeGaetano et al 1995). One of the main problems of using series climate data is missing records. Often daily maximum and minimum temperature are used by climate scientists as proxies for climate change studies. Therefore, the development of methods to estimate missing maximum and minimum temperature observations is important. Sri Lanka has a reasonable infrastructure to measure climate observations with 22 main meteorological stations covering many parts of the island and having climate observations dating back to the 1870’s. However, the weather stations in the northern and eastern parts of the country experienced problems in maintaining continuous weather records during the period from 1984 to 2000 due to the hostilities in the region. Although the recording weather observations were resumed in 2001, no attempts on the reconstruction of missing observations have been reported. The methods of estimating the monthly, seasonal or yearly mean meteorological observations are derived by averaging, interpolation, multiple linear regressions etc (Huth and Nemesova 1995). However the estimation of daily measurements is often difficult due to high variability influenced by spatial and temporal changes. The main objective of this work is to develop a technique to reconstruct the missing daily maximum and minimum temperature in Jaffna using artificial neural networks.