Neural Networking to Predict Sap Flow Using AI-Synthesized Relative Meteorological Data

The global water scarcity has created the impulse to understand the hydrological cycle in the environment and react. Transpiration is one of the dominant elements in water cycling in the Soil-Plant Atmosphere Continuum (SPAC). The main objective of the project is to set up and establish a machine learning model for predicting the transpiration and to get a high-level understanding on what processes are controlling the transpiration in the area chosen. The environmental parameters include relative humidity, solar radiation and temperature. The complex relation between these environmental factors and sap flow is studied. The data used here is synthesized from the preexisting dataset by GAN an AI framework. The synthesized data quality was also checked to ensure the reliability of the generated data. And those data were analyzed in Tableau by creating dashboards. The period of sap flow and other feature data taken is three months, out of which seventy percent is used to train and rest thirty percent is used to test the model. Initially the in-situ sap flow and meteorological data of the study area are given as input and subsequently assessing different models into the training set in order to understand what data improves the model. Machine learning algorithms including Logistic Regression (LR), Support Vector Regressor (SVR), Random Forest (RF) and neural network models Keras NN and Long Short-Term Memory (LSTM) are used and out of this it is concluded that the LSTM model performed the best among the test sets, comparing to other machine learning and neural network models. The RMSE obtained was 0.085 and The R squared was 0.92 which means the predicted values are near to accurate. This model provides a scientific reference to study the hydrological flux in the trees. And synthesizing such data helps greatly in the data acquisition cost in real time as it involves highly tedious process