Assessing forecasting performance of daily reference evapotranspiration using public weather forecast and numerical weather prediction

Highlights

• PWF and NWP forecast of ET₀ were evaluated at 31 stations across China.

• The NWP has advantages in predicting R_net and u₂, and disadvantages in T_max and T_min.

• Forecast performance of NWP for ET₀ was better than that of PWF in the MP and TC, while worse in the TM and SM.

• The systematic error was an important error source for ET₀ forecast of both PWF and NWP.

Abstract

Accurate estimation of reference evapotranspiration (ET₀) is important for water resource management and irrigation scheduling design. Limited by data availability of the numerical weather prediction (NWP), the public weather forecasting (PWF) has been widely used to forecast daily ET₀ in China. Recently, two Chinese independently developed NWP products have been open to the public, namely, the Global/Regional Assimilation and Prediction System (GRAPES) and the T639 Global Medium-term Numerical Forecast Model (T639L60). In this paper, the forecast of ET₀ is evaluated based on the recently released NWP outputs and the PWF using the FAO-56 PM equation. The daily ET₀ forecast of the two datasets was compared against the ET₀ that was calculated using weather variables observed from 31 automatic weather stations across a wide range of climate zones in China, including the temperate continental zone (TC), temperate monsoon zone (TM), mountain plateau zone (MP) and subtropical monsoon zone (SM). The results showed that the forecast performance of NWP for maximum and minimum air temperature (T_max, T_min) was worse due to the large errors in the altitude estimation process. The forecast performance of NWP for net radiation (R_net) forecasted was better than that of PWF in the MP, TC and TM, while slightly worse in the SM. The VPD from the NWP was more accurate in TC and TM, while showed little difference with that from PWF in MP and SM. For the wind speed (u₂), the NWP showed better forecast accuracy than PWF in all climate zones. Furthermore, the better forecast performance of ET₀ was obtained by the NWP in the MP and TC. In the TM and SM, the PWF can provide a more accurate ET₀ forecast compared to the NWP. Overall, the NWP can serve as a promising data source to generate acceptable ET₀ forecasts across China. The high spatial resolution of the NWP provides the ability to capture the high resolution of the spatial variability in ET₀, especially in the MP and TC where weather stations are sparse.

Introduction

Evapotranspiration (ET) is an essential component of the regional water budget and the key process linking the hydrologic cycle and biochemical cycles (Fan et al., 2018, Ma et al., 2015). Accurate estimation of ET is important for better understanding the interactions among the soil–vegetationatmosphere systems, irrigation scheduling design and validating hydrological or land surface process models (Hu et al., 2017, Yan et al., 2018, Zhang et al., 2019). Although ET can be directly measured by specialized instruments such as lysimeters and eddy covariance equipment (Allen et al., 2011a, Allen et al., 2011b, Liu et al., 2019), it is difficult to spatially replicate the equipment due to its high cost and technical complexities. Alternatively, ET can be obtained by multiplying reference evapotranspiration (ET₀) with the corresponding crop coefficient (K_c) (Allen et al., 1998). Therefore, the estimation of ET relies on the accurate computation of ET₀.

The ET₀ estimation procedures can be classified as direct and indirect methods, depending on the methodology used and the input data. For the direct methods, the time series method and artificial computational or neural networks (ANNs or CNNs) (Feng et al., 2017a, Landeras et al., 2009, Yassin et al., 2016) are the two primary procedures utilized to forecast ET₀ based on current and historical weather data. In recent years, many studies have addressed ET₀ estimation using machine learning methods, such as extreme learning machines (ELMs) (Abdullah et al., 2015, Kisi and Alizamir, 2018, Gocic et al., 2016), support vector machines (SVMs) (Fan et al., 2018, Ferreira et al., 2019) and support vector regression (SVR) (Granata, 2019). As for the indirect methods, the future weather variables are forecasted and used in empirical or analytical models such as Hargreaves-Samani (Hargreaves and Samani, 1985), the Blaney-Criddle equations (Blaney and Criddle, 1962) and the FAO-56 PM (Allen et al., 1998) models to forecast ET₀. The numerical weather prediction (NWP) model, which provides relevant weather variables such as temperature, vapor pressure, and solar radiation, is a valuable source of information for ET₀ forecasting. Ishak et al., 2010, Silva et al., 2010 used the MM5 model, a mesoscale model developed by the National Center for Atmospheric Research (NCAR) and Pennsylvania State University (PSU), to estimate daily ET₀ in the Brue catchment, southwest England and the Maipo basin in Chile, respectively. The researchers found that the forecasted daily ET₀ was overestimated by 27%-46% in England, while ET₀ was more accurate after the MM5 output was corrected in Chile. In the United States, Tian and Martinez, 2012a, Tian and Martinez, 2012b used the National Centers for Environmental Prediction’s (NCEP’s) Global Forecast System (GFS) reforecast dataset and observed solar radiation data to forecast daily ET₀ up to 5 days at a grid resolution of 2.5°×2.5° and then compared the performances of the two downscaling methods on the ET₀ forecast at points. Martins et al. (2017) used the NCEP’s blended reanalysis products with gridded datasets to compute monthly ET₀ in the Iberian Peninsula. The ACCESS-G system (developed by the Earth System Modeling program of the Centre for Australian Weather and Climate Research) was used to forecast ET₀ for lead times up to 9 days at 40 auto weather stations in Australia (Perera et al., 2014). The results indicated that ACCESS-G was capable of generating skillful forecasts of weather variables and ET₀.

In China, the national operation of the NWP started in the 1980 s. Based on the European medium-term weather forecast model, China established the first generation of global medium-term numerical forecasting systems in the early 1990 s. The system soon developed from T106 to T639, with a higher spatial resolution, lead time and forecast accuracy (Guan et al., 2008). In 2000, the China Meteorological Administration (CMA) established the National Innovative Base for Meteorological Numerical Prediction in the Chinese Academy of Meteorological Sciences (CAMS) to develop a new generation of the national operational NWP system, the Global/Regional Assimilation and Prediction System (GRAPES) (Xue, 2004, Zhang and Shen, 2008). Unfortunately, neither of these products are open to non-meteorological researchers or the public until recently. Alternatively, the public weather forecast (PWF), which contains fewer and qualitative data, including air temperature, weather type and wind scale, is widely used to forecast ET₀ in China. Cai et al. (2007) proposed an analytical method that was able to translate public forecasts into weather variables needed to calculate daily ET₀ through the FAO-56 PM equation. Luo et al. (2014) used the Hargreaves-Samani (HS) model and forecasted temperature to estimate ET₀ for lead times up to 7 days at 4 sites. Soon after, many researchers have adopted different ET₀ equations, such as the Blaney-Criddle (BC), Priestley-Taylor (PT), FAO-56 PM and temperature-based PM equations (Xiong et al., 2016, Yang et al., 2016, Yang et al., 2019a,b; Zhang et al., 2015), to forecast daily ET₀ in different regions of China. A recent study by Yang et al. (2019b) compared the forecast accuracy of daily ET₀ for six ET₀ equations based on the PWF in four main climate zones across China. The results indicated that the FAO-56 PM and temperature-based PM models provided the most accurate ET₀ due to their robust model structures, and the HS and BC models were the second choices.

Despite the wide applications of the PWF in estimating the ET₀ in China, the disadvantages of PWF are also obvious. First, limited by the few qualitative data, the PWF data must be translated by the analytical method to apply in the FAO-56 PM equation, which may cause errors and uncertainties. For the temperature-based models, a calibration procedure is needed among different weather conditions. Previous researchers usually used a long series of historical data (over 10 years) to calibrate the models and a short dataset (1–3 years) to validate the calibrated models (Luo et al., 2014, Xiong et al., 2016, Yang et al., 2019b). However, this method ignores the trends of climate change; the calibrated models may behave well in short validated years, but when the validated dataset is extended, instability may appear for the calibrated models (Feng et al., 2017b). Second, although the number of forecast weather sites has grown in recent years, in northwest China, such as Xinjiang and Tibet, the number of the weather sites is still small considering their large land areas. At most locations in northwest China, forecast weather data are not available. This issue can be solved by interpolating the point meteorological variables or ET₀ to generate high-resolution grid ET₀ (Tomas-Burguera et al., 2018, Strong et al., 2017), yet the interpolation methods and accuracy remain questionable. Recently, China’s NWP products, which have more weather variables and higher resolution, are open and available to the public in the Meteorological Data Network (http://data.cma.cn). Previous studies have focused on the forecast performance of rainfall, dust storms and typhoons in the NWP products (Huang et al., 2013, Wang et al., 2010, Yu et al., 2018), while the quantification of ET₀ forecast performance using outputs from China’s NWP products has not yet been reported as far as we know. Therefore, the objective of this study is (1) to compare the forecast performance of weather variables from the NWP and PWF outputs in different climate zones in China and (2) to assess the accuracy of ET₀ forecast using the two datasets in different climate zones in China.