A predictive irrigation scheduling model based on environmental and crop data :

Abstract

The increasing water scarcity globally is a serious challenge to irrigation development. Irrigation water management requires designing optimal irrigation systems for effective water use which is quite challenging due to limited technical skills and in-situ meteorological data, especially in developing world, for estimation of crop water need. Several approaches and models exist on estimation of crop water needs including CROPWAT, soil water balance model. This study developed a comparative predictive irrigation scheduling models based on environmental and crop data that accurately respond to fluctuations in critical parameters to enhance water use efficiency and crop productivity for given crop, the case of Nakivale Refugee Camp, Uganda. Before developing the model, re-analyzed climatic data was obtained from NASA for period of 20 years ranging from 1997-2021for Nakivale Refugee Camp. Although, the spatial variability of environmental parameters is usually dramatic, whose time series data are rowdy, non-linear, and non-stationary, and hard to predict accurately. Based on the effect of several environmental and crop parameters on irrigation schedule, 5 critical parameters were identified for the model construction thus precipitation, mean temperature, relative humidity, crop coefficient and wind speed. The autoregressive integrated moving average (ARIMA) and Extreme gradient boosting (Xgboost) machine learning (ML) algorithms in python, were employed to construct irrigation schedule by splitting the data into 90% training from 1997-2012 and 10% test data from 2020 to 2021. The accuracy and effectiveness of models was evaluated based on Root mean square error (RMSE), mean absolute error (MAE) and coefficient of R-squared (R 2). The results show that ARIMA can predict irrigation water requirement (IWR) with MAE of 0.4580 although Xgboost gave the best predictions with MAE of 0.08085. The Xgboost algorithm with inbuilt application programming interface (Api) predicted future climate parameters, crop coefficients and IWR for 5 days based on the input crop planting date. Model Validation using k-fold cross-validation analysis gave the best MAE, RMSE, MSE, R2 and NSE values of 0.08085, 0.29062, 0.08446, 0.94814 and 0.97432 respectively. This study verified the application of ARIMA and Xgboost models with Xgboost as the best ML algorithm for prediction of irrigation schedule and future IWR at local level. This can help irrigation practitioners and farmers to reduce time, enhance effective water applications and management for improved yields.