Assessment of soft computing techniques for estimating missing temperature data using limited meteorological variables

Abstract

In agriculture, the importance of complete meteorological datasets is a deciding element, supporting farmers in harvesting healthy and plentiful produce. The essential meteorological parameters for agriculture are precipitation and temperature, which are used to plan field operations from seeding to harvesting. This study aimed to modelling the infill missing temperature data. This research examines soft computing models viz. artificial neural networks (ANNs) and multiple linear regressions (MLR) for estimating missing daily maximum and minimum temperature, for which several input composites were constructed and performance was ensured using various performance indices. Intended for that 30-year (1990-2019) of daily meteorological datasets (maximum and minimum temperature, relative humidity, wind speed, and bright sunshine hours) were used to develop the various input composite models. The comparative results of various input composites illustrated that the ANN model outperformed MLR models to estimates maximum and minimum temperature. For estimating maximum and minimum temperature, the optimal input composite ANN model has superior performance indices in terms of R, NSE, RMSE, MAE as 0.86, 70.08, 2.74, 2.14 and 0.88, 76.00, 3.20, 2.38, respectively. The results also demonstrated the temperature can be expressed as non linear function of specified meteorological factors. Also, confirmed that the ANNs’ ability to forecast temperature in semi-arid areas.