Weather data-centric prediction of maize non-stressed canopy temperature in semi-arid climates for irrigation management

Authors: NAKABUYE, H - University Of Nebraska; RUDNICK, DARAN - University Of Nebraska; DeJonge, Kendall; Ascough, Katherine; LIANG, WEI-ZHEN - University Of Nebraska; LO, TSZ - Mississippi State University; FRANZ, TRENTON - University Of Nebraska; QIAO, XIN - University Of Nebraska; KATIMBO, ABIA - University Of Nebraska; DUAN, JIAMING - University Of Nebraska

Submitted to: Irrigation Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/25/2023
Publication Date: 5/17/2023
Citation: Nakabuye, H., Rudnick, D., DeJonge, K.C., Ascough, K.A., Liang, W., Lo, T., Franz, T., Qiao, X., Katimbo, A., Duan, J. 2023. Weather data-centric prediction of maize non-stressed canopy temperature in semi-arid climates for irrigation management. Irrigation Science. https://doi.org/10.1007/s00271-023-00863-w.
DOI: https://doi.org/10.1007/s00271-023-00863-w

Interpretive Summary: Maize canopy temperature (Tc) measurements are increasingly being used to estimate water stress for improved irrigation water management. However, most methods of this type require a non-stressed canopy temperature (Tcns) for a basis of comparison, and it may be easier to predict this value from weather data than to maintain an observable non-stressed crop. This study used 36 methods combining models with weather data to determine the best methods to predict Tcns in North Platte, Nebraska and Greeley, Colorado. Results showed that selection of the right weather variables for comparison might be more important than the specific models used, and that air temperature, relative humidity, and solar radiation were the most important weather inputs. The outcomes of this study will make irrigation scheduling based on Tc measurement more feasible in practice, which will be of benefit to farmers with water limitations.

Technical Abstract: Maize canopy temperature (Tc) measurements are increasingly being used to compute crop thermal indices for water stress estimation and improved irrigation water management. Conventionally monitoring of crop thermal response requires the maintenance of a well-water crop from which non-stressed canopy temperature (Tcns) is measured as a reference for thermal index computation. This study alternatively evaluated the performance of 36 model x weather data driven model combinations to predict peak time Tcns in maize grown in semi-arid and arid climates. The data-drive models considered were multilinear regression model (MLR), forward feed neural network model (NN), recurrent neural network model (RNN), multivariate adoptive regression splines (MARS), random forest (RF) and, k- nearest neighbor (KNN). For each of these models the following weather data combinations were tested: average air temperature (Ta), average relative humidity (RH), wind speed (U2), and solar radiation (Rs) (combination 1); RH, U2, and Rs (combination 2), Ta, RH, and Rs (combination 3); Ta, and RH (combination 4); RH, and Rs (combination 5); and Ta, and Rs (combination 6). Ranking the performance of the weather data x model combinations across both climate sites showed that the MARS model with combination 1 was a better predictor of Tcns with R2 value of 0.910, RMSE value of 0.693 in semi-arid conditions and R2 of 0.866 and RMSE value of 0.966 in the arid conditions. Model performance metric ranks and the computation of Shapley Additive Explanations (SHAP) values indicated that in the instances of limited weather parameters, Ta and RH were better predictors of Tcns in semi-arid climates while Ta and Rs showed better prediction of Tcns in arid climates. The performance site specific (localized) and generalized model combinations was also compared and indicated that cross site prediction of Tcns was primarily determined by weather data combinations, rather than model specificity.