Total precipitable water (TPW), a column of water vapor content in the atmosphere, provides information on the spatial distribution of moisture. The high-resolution TPW, together with atmospheric stability indices such as convective available potential energy (CAPE), is an effective indicator of severe weather phenomena in the pre-convective atmospheric condition. With the advent of high performing imaging instrument onboard geostationary satellites such as Advanced Himawari Imager (AHI) onboard Himawari-8 of Japan and Advanced Meteorological Imager (AMI) onboard GeoKompsat-2A of Korea, it is expected that unprecedented spatiotemporal resolution data (e.g., AMI plans to provide 2 km resolution data at every 2 min over the northeast part of East Asia) will be provided. To derive TPW from such high-resolution data in a timely fashion, an efficient algorithm is highly required. Here, machine learning approaches—random forest (RF), extreme gradient boosting (XGB), and deep neural network (DNN)—are assessed for the TPW retrieved from AHI over the clear sky in Northeast Asia area. For the training dataset, the nine infrared brightness temperatures (BT) of AHI (BT8 to 16 centered at 6.2, 6.9, 7.3, 8.6, 9.6, 10.4, 11.2, 12.4, and 13.3 $\mathsf{\mu }\mathrm{m}$ , respectively), six dual channel differences and observation conditions such as time, latitude, longitude, and satellite zenith angle for two years (September 2016 to August 2018) are used. The corresponding TPW is prepared by integrating the water vapor profiles from InterimEuropean Centre for Medium-Range Weather Forecasts Re-Analysis data (ERA-Interim). The algorithm performances are assessed using the ERA-Interim and radiosonde observations (RAOB) as the reference data. The results show that the DNN model performs better than RF and XGB with a correlation coefficient of 0.96, a mean bias of 0.90 mm, and a root mean square error (RMSE) of 4.65 mm when compared to the ERA-Interim. Similarly, DNN results in a correlation coefficient of 0.95, a mean bias of 1.25 mm, and an RMSE of 5.03 mm when compared to RAOB. Contributing variables to retrieve the TPW in each model and the spatial and temporal analysis of the retrieved TPW are carefully examined and discussed.

Early warning of severe weather caused by intense convective weather systems is challenging. To help such activities, meteorological satellites with high temporal and spatial resolution have been utilized for the monitoring of instability trends along with water vapor variation. The current study proposes a retrieval algorithm based on an artificial neural network (ANN) model to quickly and efficiently derive total precipitable water (TPW) and convective available potential energy (CAPE) from Korea’s second geostationary satellite imagery measurements (GEO-KOMPSAT-2A/Advanced Meteorological Imager (AMI)). To overcome the limitations of the traditional static (ST) learning method such as exhaustive learning, impractical, and not matching in a sequence data, we applied an ANN model with incremental (INC) learning. The INC ANN uses a dynamic dataset that begins with the existing weight information transferred from a previously learned model when new samples emerge. To prevent sudden changes in the distribution of learning data, this method uses a sliding window that moves along the data with a window of a fixed size. Through an empirical test, the update cycle and the window size of the model are set to be one day and ten days, respectively. For the preparation of learning datasets, nine infrared brightness temperatures of AMI, six dual channel differences, temporal and geographic information, and a satellite zenith angle are used as input variables, and the TPW and CAPE from ECMWF model reanalysis (ERA5) data are used as the corresponding target values over the clear-sky conditions in the Northeast Asia region for about one year. Through the accuracy tests with radiosonde observation for one year, the INC NN results demonstrate improved performance (the accuracy of TPW and CAPE decreased by approximately 26% and 26% for bias and about 13% and 12% for RMSE, respectively) when compared to the ST learning. Evaluation results using ERA5 data also reveal more stable error statistics over time and overall reduced error distribution compared with ST ANN.