Effect of different data quality control on evapotranspiration of winter wheat with Bowen ratio method

Abstract

The Bowen ratio energy balance (BREB) method is widely used to study surface evapotranspiration, but its major drawback is the uncertainty when Bowen ratio (β)→ −1. Various approaches have been employed to address this issue, but their performances were less evaluated via long-term field observations. Using data from three growing cycles, this study investigated the effect of five screening methods (Mth1 to Mth5 for −1 − |ε₁| < β < −1 + |ε₁|, −1.05 < β < −0.95, β < −0.75, −1.3 < β < −0.75 and β < −0.7 or β > 10 or Δe ≤ 0, Δe denotes the measured vapor pressure gradient, and ε₁ is a coefficient depending on sensor resolution and Δe) on winter wheat evapotranspiration in northern China. On diurnal, daily and seasonal basis, the effect was in the order of Mth5 > Mth3 > Mth1 > Mth2 > Mth4, and the seasonal mean daily value of the gap-filled was 0.38, 0.22, 0.11, 0.01, and 0.01 mm d⁻¹ higher than the unfilled ones, yielding a seasonal total of 96.0, 53.5, 26.0, −0.9, and 0.4 mm, or 18.9 %, 11.4 %, 6.5 %, −0.2 %, and 0.1 % higher than the unfilled ones, respectively. These values resulted from the large difference in data rejection ranking as Mth5 > Mth3 > Mth1 > Mth4 > Mth2, seasonal mean daily 10-min rejection rate ranging from 15.4–73.2 %, 10.3–48.9 %, 5.3–44.9 %, 1.6–10.4 %, and 0.5–7.3 %, respectively (averaging 42.4 %, 30.5 %, 23.2 %, 5.7 %, and 2.6 %, respectively). The corresponding daily rejected hours ranged from 6.83–8.88, 3.60–6.11, 1.85–3.49, 0.10–0.39, and 0.07–0.33 h/day, respectively (averaging 7.53, 4.77, 2.90, 0.28, and 0.24 h/day, respectively), resulting in large data gaps for Mth5 (58.8 %), Mth3 (38.2 %), and Mth1 (17.5 %). Nighttime deletion dominated for Mth2 to Mth4, accounting for 61.1 %, 64.4 %, 68.3 %, and 63.2 % of the total deletion, whereas daytime deletion dominated for Mth1, accounting for 58.1 %. A large portion of invalid rejections of Mth1 (40.4 %–77.6 %), Mth3 (54.3 %–90.9 %) and Mth5 (61.8 %–92.7 %) was observed at the selected period, which was probably a consequence of the sensor’s error cancellation effect, questioning the traditional a priori assumption that small vapor gradients within instrumental error should be discarded. Overall, large differences were observed and the simple Mth4 performed better than the more restrictive ones. These findings are expected to guide the selection of post-data processing in the application of BREB method.