基于临近台站气象数据的参考作物蒸散量估算方法

参考作物蒸散量（ET₀）是评估区域植被耗水进而指导水资源优化管理所需的主要参数之一,但我国大部分地区标准化气象台站稀疏、部分研究点的气象资料通常难以获取,给ET₀的计算带来了很大困难。以地处内陆旱区的内蒙古河套灌区为例,利用该区4个标准气象站1981-2006年的气象资料,讨论了研究点在没有历史气象数据且现有气象数据不完备的情况下,采用临近台站气象数据估算ET₀的可靠性。估算方法分别为估算未知气象数据的FAO56 Penman-Monteith方程（PM56）、基于临近台站气象数据校正的经验公式以及利用临近台站气象数据训练的人工神经网络模型。结果表明：（1）在完全没有气象数据的条件下,可采用临近站点的气象数据估算研究点的ET₀,平均绝对误差（MAE）为0.43~0.52 mm d^-1,均方根误差（RMSE）为0.56~0.63 mm d^-1;估算精度与台站间的距离有关,利用维度信息校正太阳辐射值可提高估算精度。（2）仅有最高和最低气温数据时,估算气象数据的PM56方程计算误差较大,且站点之间表现不稳定,人工神经网络模型的估算精度最高,MAE和RMSE分别为0.14~0.22 mm d^-1和0.17~0.29 mm d^-1;校正后的Hargreaves公式的估算效果次之,MAE和RMSE分别为0.23~0.26 mm d^-1和0.30~0.31 mm d^-1。（3）在已知温度和辐射数据时,利用临近台站气象数据训练的人工神经网络模型依然表现最好,MAE和RMSE分别为0.13~0.19 mm d^-1和0.17~0.25 mm d^-1,其他两种方法误差较大。在内陆干旱条件下,利用研究点的气温数据结合临近台站的历史气象信息可有效估算参考作物蒸散发。

Comparison of different methods for estimating reference evapotranspiration with weather data from nearby stations

Potential evapotranspiration (ET₀) is one of the most critical parameters that are essential for evaluating regional vegetation water use and managing water and soil resources. However, accurate estimation of it is so difficult for many parts of China due to a limited number of weather stations. Weather data from nearby stations are available for most sites, but past relevant works mainly focused on ET₀ predicting methods with local weather data and less work was done to investigate the approaches for estimating ET₀ with data from other stations. Therefore, this study was conducted to test the reliability of estimating ET₀ with weather information from nearby stations. Whether data of four weather stations located in Hetao Irrigation District of western Inner Mongolia were collected. The study area has an arid climate with annual rainfall and pan evaporation of 130-215 and 2100-2300 mm respectively. Three commonly used approaches, namely, FAO56 Penman-Monteith equation (PM56) with estimated weather data, the empirical formula corrected with meteorological data of nearby stations, and the artificial neural network model (ANN) developed using meteorological data from nearby stations, were compared for predicting ET₀ when data are limited in this work. The results showed that: (1) When all of the necessary parameters were not measured at the study sites, weather data from nearby stations could be used directly, the average absolute error (MAE) of the ET₀ calculation was 0.43-0.52 mm d^-1, and the root mean square error (RMSE) was 0.56-0.63 mm d^-1, and the error could be narrowed by correcting the radiation data using the latitude information of the stations; (2) When the maximum and minimum air temperature data were available, PM56 with estimated weather data performed the worst, and the performance of the ANN model is the best with the MAE and RMSE ranging between 0.14-0.22 mm d^-1 and 0.17-0.29 mm d^-1, respectively; and the results of the calibrated Hargreaves formula are intermediate with the MAE and RMSE values of 0.23-0.26 mm d^-1 and 0.30-0.31 mm d^-1, respectively; (3) When knowing the temperature and radiation data, the ANN model trained with meteorological data from nearby stations is still the best, with MAE and RMSE values of 0.13-0.19 mm d^-1 and 0.17-0.25 mm d^-1, respectively, and the other two methods have larger error and the performance is inconsistent among different stations. The results of this study demonstrated that we can estimated ET₀ accurately using air temperature data in combination with weather data of nearby stations under arid conditions, and the reliability of the methods still need to be validated under other climatic conditions.