Light Gradient Boosting Machine: An efficient soft computing model for estimating daily reference evapotranspiration with local and external meteorological data

Abstract

Abstract Accurate estimation of reference evapotranspiration (ETo) is required in many fields, e.g. irrigation scheduling design, agricultural water management, crop growth modeling and drought assessment. Nevertheless, reliable estimation of ETo is difficult when lack of complete or long-term meteorological data at the target station. This study evaluated the efficiency of a new tree-based soft computing model, Light Gradient Boosting Machine (LightGBM), for estimating daily ET0 using limited local (target-station) and external (cross-station) meteorological data from 49 weather stations in humid subtropical region of China, including 16 in Jiangxi Province and other 33 in the region. The performance of LightGBM was compared with the tree-based M5 Model Tree (M5Tree) and Random Forests (RF) as well as four empirical models (Hargreaves-Samani, Tabari, Makkink and Trabert). Eight input combinations of daily meteorological data including maximum temperature (Tmax), minimum temperature (Tmin), relative humidity (Hr), wind speed at 2 m height (U2), extraterrestrial solar radiation (Ra) and global solar radiation (Rs) calculated from sunshine duration (n) for the period 2001–2015 were used to test the models. The results showed that LightGBM was superior to M5Tree and RF in local applications under all input combinations during testing, with average root mean square error (RMSE) of 0.08–0.58 mm d−1, 0.11–0.62 mm d−1and 0.13–0.60 mm d−1, respectively. M5Tree performed slightly better than RF under input combinations 6–8, whereas RF outperformed M5Tree under the other input combinations. However, all three soft computing models produced much better daily ETo estimates than the corresponding empirical models with the same input variables. Rs was the most influential meteorological variable for daily ETo estimation in this region, followed by Tmax and Tmin, Hr and finally U2. In external applications, LightGBM also generally performed better than the RF, M5Tree and empirical models. Soft computing models developed with meteorological data from Station 57894, having the most similar climatic characteristics to the other stations, gave satisfactory ETo estimates for the 15 cross stations in Jiangxi Province, even for the other 33 stations across the humid subtropical region of China. LightGBM was proved to be efficient and exhibit good generalization capability in both local and external applications, which was thus recommended as an alternative model for daily ETo estimation.