Evaluation of Rule and Decision Tree Induction Algorithms for Generating Climate Change Scenarios for Temperature and Pan Evaporation on a Lake Basin

Abstract

Climate change scenarios generated by general circulation models (GCMs) have too coarse a spatial resolution to be useful in planning disaster risk reduction and climate change adaptation strategies at regional to river/lake basin scales. This paper investigates the performances of existing state-of-the-art rule induction and tree algorithms, namely, single conjunctive rule learner, decision table, M5P model tree, decision stump, and REPTree. Downscaling models are developed to obtain projections of mean monthly maximum and minimum temperatures (Tmax and Tmin) as well as pan evaporation to lake-basin scale in an arid region in India using these algorithms. The predictor variables, such as air temperature, zonal wind, meridional wind, and geo-potential height, are extracted from the National Centers for Environmental Prediction (NCEP) reanalysis data set for the period 1948–2000 and from the simulations using third-generation Canadian coupled global climate models for emission scenarios for the period 2001–2100. A simple multiplicative shift was used for correcting predictand values. The performances of various models have been evaluated on several statistical performance parameters such as correlation coefficient, mean absolute error, and root mean square error. The M5P model tree algorithm was found to yield better performance among all other learning techniques explored in the present study. An increasing trend is observed for Tmax and Tmin for emission scenarios, whereas no trend has been observed for pan evaporation in the future.