Symbolic regression model for reliable estimation and projections of evapotranspiration

Abstract

Evapotranspiration (ET) plays a significant role in defining water demand, surface runoff, atmospheric moisture and precipitation. It is well recognized that ET is changing in regional and global scales due to rising temperature induced by global warming. Reliable estimation and future projections of ET with associated uncertainties are extremely important for agricultural and water resources development, planning and management. However, projections of ET using wellestablished empirical ET models suffer from large uncertainty due to their dependency on a large number of climatic variables. The major objective of the present study was to develop empirical ET models for reliable estimation and projection of ET in the context of global warming. Daily temperature, humidity, solar radiation, wind speed and pan evaporation data recorded at ten meteorological stations distributed over peninsular Malaysia was used for the development of four sets of ET models using Gene Expression Programming (GEP) based on a different combination of meteorological variables. The superiority of GEP generated ET models were established by comparing their performance with the most suitable ET model selected using compromise programming approach from the pool of existing ET models. A trend conserving perturbation approach was used to downscale the Global Climate Model (GCM) projected temperatures, which were then used for projection of future changes in ET using GEP generated temperature-based ET models for four Representative Concentration Pathways (RCPs) scenarios namely, RCP 2.6, 4.5, 6.0 and 8.5. The results revealed the Penman-Monteith as the most suitable method of estimation of ET followed by radiation-based Priestley and Taylor and the mass transfer-based Dalton and Meyer methods. Among the temperature-based methods, Ivanov was found the best. Comparison of GEP-based ET models with the existing most suitable empirical model in peninsular Malaysia showed better performance of GEP models in term of all standard statistics. The Nash Sutcliffe efficiency coefficients of GEP models were found more than 0.93 for all the GEP models during validation, which was higher than that obtained using existing empirical models. Downscaling of temperature revealed a continuous increase in minimum, maximum and average temperatures over the present century under all RCPs. The minimum temperature was projected to increase in the range 2.47-3.30°C, the maximum temperature in the range of 2.79-3.24°C, and the mean temperature in the range of 2.56-3.20°C during 2070-2099. The minimum temperature was found to increase more compared to maximum temperature in most of the stations. The ET in peninsular Malaysia was projected to change in the range of -4.35% to 7.06% under RCP2.6, - 1.99% to 16.76% under RCP4.5, -1.66% to 22.14% under RCP6.0 and -0.91% to 39.7% under RCP8.5 during 2010-2099. Relatively more increase in ET was projected in the North compared to other parts of peninsular Malaysia. The rise in ET was found to follow the trend in temperature in most of the stations. The results also revealed high uncertainty in the projections of ET. The uncertainty in the rise of ET was found to increase with time and for higher RCPs. It can be expected that the methodology proposed in the present study can be useful in the reduction of uncertainty in the projection of ET which in turn can help in cost-effective adaptation and mitigation planning.