MEASUREMENT AND MODELLING OF WHOLE-PLANT WATER USE OF COCOA GROWN IN SOUTHWEST, NIGERIA

Abstract

Cocoa (Theobroma cacao) is principally produced by small-holder farmers under conventional management in Nigeria. Water is a key factor in the management of cocoa under this conventional system, however information on water use characteristics of cocoa trees in Nigeria is scarce. Hence, the aim of the study is to measure the water use of rain-fed cocoa and parameterize models of canopy conductance suitable for inclusion in physically-based models for predicting water use of cocoa plants. Sap flow density was monitored in three (3) cocoa trees (Forestaro cultivar group) at eight (8) year old cocoa plantation in Akure, Nigeria, from 7th March 2018 to 10th March 2019, covering wet and dry seasons. A web application system was developed using an R package named Shiny to deal with the problem of pre-processing of both climatic and field data, and computational complexities associated with water use estimation. Vector Autoregressive Models (VAR), a multivariate time series model, and Long Short-Term Memory (LSTM) network, an Artificial Intelligence (AI) model were employed to predict the stomata control of water transport in the cocoa trees. The prediction power of the VAR model was assessed and visualized using the VARs R package, while for the LSTM model, a Recurrent Neural Network (RNN) algorithm was implemented using Python programming within Google COLAB jupyter notebook. VAR models were evaluated using Adjusted R-squared and Root Mean Squared Error (RMSE), while LSTM was evaluated using RMSE and by comparing the train loss and test loss of the model. The result of the research revealed that temporal dynamics of transpiration is consistent in pattern with temperature, solar radiation and relative humidity. Total transpiration during field measurement was 928.06 mm. The five months (151 days) of dry season accounted for 30.96% (287.35 mm) while the seven months (214 days) of wet season was 69.04% (640.71 mm). The mean volume of water transpired during the dry season was about 1.90 mm day-1, while that of the wet season was 2.99 mm day-1 Transpiration accounted for 75.27% of total precipitation during the period of experiment. The potential or reference evapotranspiration (ETo) as estimated by the Penman-Monteith (PM) formula was consistently lower than the transpiration of cocoa mainly during the wet season. The Web application developed was successful in removing duplicates in the data, identifying outliers, presenting summaries, and estimating cocoa water use characteristics in less than a minute. VAR (with Adjusted R-Squared of 0.1139) was found not to be suitable to model the complex relationship between canopy conductance and climatic variables. However, LSTM with the test loss not dropping below the train loss and with RMSE of 0.026 was observed to perform better in modelling the canopy conductance of cocoa. The overall results indicate that annual rainfall of not less than 1000 mm may be required to meet the water use of rain-fed cocoa in South-West Nigeria and LSTM with a prediction accuracy of 97.4% could be successfully employed to predict cocoa canopy conductance. In addition, the web application developed could be used to deal with computational complexities associated with water use estimation of cocoa.