ESTIMATION OF REFERENCE EVAPOTRANSPIRATION USING ARTIFICIAL NEURAL NETWORK FOR MOHANPUR, NADIA DISTRICT, WEST BENGAL: A CASE STUDY

Abstract

Estimation of evapotranspiration plays a key role in various water management studies including irrigation scheduling and water budgeting. Being an extremely complex and non-linear phenomena, precise estimation of evapotranspiration requires large number of climatological data as well as vast time. In recent past, artificial neural network has emerged as a successful tool to model complex non-linear relationships including evapotranspiration process. The current study investigates the potential of artificial neural network models to estimate reference evapotranspiration (ET0) for Mohanpur area and compares the performance of ANN models with reference ET estimated by FAO-Penman method. Different combinations of six weather parameters namely maximum air temperature, minimum air temperature, maximum relative humidity, minimum relative humidity, wind velocity and actual sunshine hours were used as inputs to train the 12 multilayer feed forward perceptron ANN models selected for the study. The FAO-56 Penman estimated ET0 was used as output for all the models. The models were trained with back propagation learning algorithm. The analysis is carried out in MATLAB software. For each combination of input parameters, the best ANN model was selected with least SEE and highest R 2. The result of the study inferred that ANN performed very well with all the input parameters which were used in reference ET estimation by FAO-Penman method but the ANN models with less input variables also yielded very good estimation of ET0. Therefore, it can be suggested that ANN method can be used for ET0 estimation for the study area with high degree of accuracy in limited data condition also. Evapotranspiration (ET) represents one of the major contributor of hydrologic cycle which combinedly represents two hydrological process namely evaporation and transpiration. This parameter of hydrologic cycle plays a key role in various water resources management studies and therefore, its precise estimation has paramount importance on efficient water resources planning and budgeting. Actual evapotranspiration can directly be measured from the field, but field measurement of ET demands precision and time. Reference evapotranspiration (ET0) is one of the very common terms used to define ET as theoretical evapotranspiration from an extensive surface of green grass of uniform height, actively growing, completely shading the ground, and not short of water [1]. In the past few decades, numerous studies were carried out in developing various empirical and semi-empirical models for precise estimation of ET0 using various climatological data [2, 3, 4, 5, 1]. Some of these models depend on a variety of weather parameters whereas, some models give good results with less climatological data. Among the various reference ET estimation methods, the Penman-Monteith method is widely accepted [6] and it is proved to be the most accurate method of estimating evapotranspiration. But the requirements of large input parameters have limited the use of this method. In recent past, artificial neural network (ANN) approach has been successfully employed in evapotranspiration estimation studies [7, 6, 8, 9]. According to Sudheer et al., (2003) [10], ANN methods lead over conventional methods in their ability to solve the complex nonlinear relationships through flexible mathematical configurations. Evapotranspiration process is a complex and nonlinear phenomenon and depends on several interacting climatological factors, such as temperature, humidity, wind speed, radiation, type, and growth stage of the crop etc. The unique feature of ANN of solving complex nonlinear problems has made this technique a huge success in evapotranspiration estimation and modeling studies. Bai and Sinha (2015) [11] used artificial neural networks (ANNs) model for estimating reference crop evapotranspiration with limited climatic data and compared the performance of ANNs with P-M method for Ambikapur regions. The result depicted a satisfactory performance of ANN in the ET0 estimation as compared to Penman Montieth method and inferred that these ANN models may therefore be adopted for estimating ET0 in the study area with reasonable degree of accuracy. With this background, the present study has aimed to model evapotranspiration using artificial neural network for Mohanpur area, West Bengal.