以支援向量機為基礎於有限氣象資料進行蒸發量補遺

Imputation of Evaporation Data by Using a Support Vector Machine Based Model with Limited Meteorological Data

林軒宇(Hsaun-Yu Lin) ； 劉說安(Yuei-An Liou)

《農業工程學報》 61卷4期 (2015/12) Pp. 1-10

https://doi.org/10.29974/JTAE.201512_61(4).0001

摘要

於水文循環中，蒸發量為相當重要的水文因子，準確的估算蒸發量可作為水資源管理與農業灌溉重要參考資訊，然而蒸發量資料可能會因儀器毀損或記錄錯誤等原因造成資料缺漏或遺失。因此本研究提出以支援向量機為基礎，利用氣象因子建置蒸發推估模式以進行資料補遺。首先，蒐集與蒸發過程相關之氣象因子，並篩選最佳氣象因子組合以建置蒸發推估模式。接著，本研究嘗試於有限氣象資料下，以較常見之氣溫與濕度因子為輸入項建立蒸發推估模式，以驗證提出模式之適用性。最後則以建置完成之模式進行缺漏資料補遺，以評估資料補遺與否之差異。結果顯示，本研究所提出之模式不論於最佳氣象因子組合或是於有限氣象因子條件下，皆可保有相當優異之推估能力。此外，提出之模式於不同輸入項時，皆能保有一致性之補遺成果。因此建議可以本研究提出之方法進行推估模式之建置與缺漏資料之補遺。

關鍵字

支援向量機 ； 蒸發量 ； 資料補遺 ； 氣象因子

並列摘要

Evaporation is a major factor in hydrological cycle. Its estimation can provide a practical reference for water resources management and agricultural irrigation. However, observed evaporation data are sometimes not available due to measurement or recording failure. In this research, an effective model based on support vector machine (SVM) is proposed to estimate missing pan evaporation by using meteorological data as input. First, the meteorological data that affect evaporation are collected, and the optimal input combination is selected by input determination process to construct SVM_(opt) model for evaporation estimation. Then, in order to extend the applicability of the proposed models, SVM_(temp) and SVM_(hum) models, which use commonly measured data in a weather station as input, are also constructed. Additionally, the proposed models are used to estimate missing evaporation data, and the estimation results are evaluated. Results show that the proposed models can estimate evaporation accurately with limited meteorological data, and the proposed models can estimate missing data consistently under different input combinations. The proposed modeling technique is expected to be useful to construct an evaporation estimation model, and the proposed model is recommended as an alternative approach for estimating missing evaporation data.

並列關鍵字

Support vector machine ； Evaporation ； Data imputation ； Meteorological data