Journal
JOURNAL OF HYDROLOGY
Volume 588, Issue -, Pages -Publisher
ELSEVIER
DOI: 10.1016/j.jhydrol.2020.125087
Keywords
Reference crop evapotranspiration; CatBoost; Random forests; Generalized regression neural network; Arid and semi-arid regions of Northern China
Funding
- Major Program of National Social Science Foundation of China [17ZDA064]
- Youth Xinjiang Science and Technology Innovation Talents Training Project of China [QN2016YX0347]
- Xinjiang Natural Science Foundation of China
Ask authors/readers for more resources
Establishing a computational model for accurate prediction of reference crop evapotranspiration (ET0) is critical for regional water resources planning and irrigation scheduling design. FAO Penman-Monteith equation is recommended as the standard model to predict ET0. However, its application is restricted by lack of complete meteorological data in many regions. This study evaluated the performance of CatBoost, an algorithm for gradient boosting on decision trees, for estimating daily ET0 using limited meteorological data in arid and semi-arid regions of Northern China. The CatBoost model was further compared with their corresponding generalized regression neural network (GRNN) and random forests (RF) models. Eight input combinations of daily meteorological data including daily maximum air temperature (T-max), daily minimum air temperature (T-min), wind speed at 2 m height (u(2)), relative humidity (RH) and net radiation (R-n) from 15 weather stations during 1996-2015 were used to train and WA the models. Four statistical indicators were used to evaluate the accuracy and performance of the models, including coefficient of determination (R-2), root mean square error (RMSE), mean absolute error (MAE), and mean bias error (MBE). The results showed that all the three models using T-max, T-min, u(2) and R-n could obtain satisfactory ET0 estimates in arid and semi-arid regions of Northern China with incomplete sets of data. For the local models, CatBoost (on average RMSE ranging 0.096-0.821 mm d(-1)) was superior to GRNN (on average RMSE ranging 0.206-0.847 mm d(-1)) and RF (on average RMSE ranging 0.169-0.866 mm d(-1)) under the same meteorological parameters as input. The results of the generalized models were similar to the local models, but the former ones performed worse than the latter ones. Overall, CatBoost is observed to be the best alternative for estimating ET0, which is helpful for irrigation scheduling in arid and semiarid regions of Northern China and maybe elsewhere with similar climates.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available