基于ConvLSTM的西北太平洋海表温度中短期预报

尽管海表温度(Sea Surface Temperature,SST)短期变化较小,但这种变化对海洋涡旋、海洋锋以及热带气旋的发生发展仍有着重要的影响,因此短期SST预报意义重大,且对预报精度的要求较高。本文基于ConvLSTM的深度学习模型,利用SST和温度平流双预报因子对西北太平洋划定区域内SST进行7 d的连续预报,将其结果与仅使用SST预报因子ConvLSTM以及混合坐标海洋模型(HYbrid Coordinate Ocean Model,HYCOM)的预报结果分别进行了对比。结果表明,在7 d的预报时效内,温度平流预报因子的加入可使得ConvLSTM模型预报技巧大幅提升,明显优于HYCOM模式。此外,本文将预报时效进一步延长至30 d,对模型在不同季节的预报能力进行了分析,发现ConvLSTM模型在春、秋季(夏、冬季)的预报效果相对较好(差)。

Short-medium-term forecast of SST over western North Pacific based on ConvLSTM

Despite of the small change in short-term variation of Sea Surface Temperature (SST), the change plays an important role in determining the occurrence and development of ocean vortices, ocean fronts and tropical cyclones. Therefore, short-term SST forecast is of great significance and requires high accuracy. In this study, to make a continuous forecast of 7-day SST over a certain area in western North Pacific, a deep learning model based on the ConvLSTM was adopted by using the two features, namely, SST and temperature advection. The forecast results of this two-feature ConvLSTM were compared with not only those of one-feature (i.e., SST) ConvLSTM but also those of HYbrid Coordinate Ocean Model (HYCOM). Results show that, within the 7-day forecast time, the addition of the temperature advection feature can largely improve the forecast skill of ConvLSTM, which even beyond HYCOM. Moreover, this study extended the forecasting time to 30 days, and analyzed the forecast skill of the ConvLSTM model in different seasons. It was found that the ConvLSTM model exhibits relatively high (low)forecast skill in spring and autumn (summer and winter).