|
摘要
|
Crop evapotranspiration, also known as crop water requirement (ETcrop), is a crucial parameter for the government’s promotion of farmland reorganization or the expansion of irrigation services, helping to grasp the water demand in irrigation areas. The reference crop evapotranspiration (ETo) is the primary step in indirectly calculating crop water requirements (ETcrop). The FAO56 Penman-Monteith method is one of the recommended methods by the Food and Agriculture Organization (FAO) for estimating ETo and is widely used internationally; however, this method requires multiple meteorological parameters, making it difficult to apply extensively in practice. Furthermore, considering the impact of climate change on climate and water resources, the fluctuations in crop water requirements over different future periods influence the overall supply and demand of water resources in Taiwan. Therefore, assessing future trends in advance can help the country prepare for changes in weather and water conditions and formulate policies for adjusting industrial water use and regional water resource allocation.In this context, this study aims to establish an optimal model for estimating reference crop evapotranspiration (ETo) based on the rapidity, universality, accuracy, and trend prediction in response to climate change. The model consists of three calculation methods: (1) Machine Learning (MLP), (2) Hargreaves and Samani (HS) empirical formula adjustment, and (3) comparison of HS, MLP, and FAO-recommended parameter substitution methods. RMSE is used as the selection evaluation index. Additionally, through a single-parameter method in machine learning, three suitable GCMs for Taiwan (CanESM5, EC-Earth3, ACCESS-ESM1-5) are selected, and future temperature data from TCCIP is applied to estimate trend changes under four future scenarios. This study classifies Taiwan into four different climate zones based on the Köppen-Geiger climate classification method and selects Chiayi, Hsinchu, Hengchun, and Yongkang weather stations as representatives for each climate zone. Twenty years of daily data (2004-2023) are used for evaluation, with 16 years for model establishment and four years for validation. The results show: (1) The linear regression method (HS_adj2) is superior to the single coefficient adjustment method (HS_adj1) in adjusting the Hargreaves and Samani (HS) model, with RMSE values reduced by 0~21.74% and 0~12.35%, respectively; (2) Cross-analysis and validation indicate that solar radiation data is the most important- 164 meteorological parameter for estimating ETo, with wind speed and humidity being less significant; (3) The optimal model shows that when meteorological observations only include temperature data, the best estimation method for ETo is the MLP-a model, followed by the HS_adj2 model, while the performance ranking of the HS model and PM1 model depends on the target station, but in most cases, the PM1 model outperforms the HS model; (4) Climate change prediction results show that all stations applying the three GCMs and four simulation scenarios exhibit the same trend, indicating that ETo continuously rises from the short term (2021-2040) to the medium term (2041-2060) and to the mid-long term (2061-2080), reaching a peak in the mid-long term or long term (2081-2100). This underscores the increasing demand for agricultural irrigation water in Taiwan over time due to climate change, suggesting that future water resource supply and demand policies should be adjusted in advance to respond accordingly. |