This study proposed a deep learning-based model to estimate stream water-use rate (WUR) using precipitation (P) and potential evapotranspiration (PET). Correlations were explored to identify relationships among accumulated meteorological variables for various time durations (three-, four-, five-, and six-month cumulative) and WUR, which revealed that three-month cumulative meteorological variables and WUR were highly correlated. A deep belief network (DBN) based on iterating parameter tuning was developed to estimate WUR using P, PET, and antecedent stream water-use rate (DWUR). The training and validation periods were 2011–2016, and 2017–2019, respectively. The results showed that the PET-DWUR based model provided better performances in Nash–Sutcliff efficiency (NSE), root mean square error (RMSE), and determination coefficient (R²) than the P-PET-DWUR and P-DWUR models. The framework in this study can provide a forecast model for deficiencies of stream water use coupled with a weather forecast model.

This study explored return periods of the streamflow deficit characteristics by threshold levels with different time resolutions (daily, weekly, and monthly) using univariate and Copula functions, reflecting hydrological regimes like rapid recoveries from drought and anthropogenic dam operation by monthly basis in South Korea. As a result, the daily-based threshold had a large variability in severities and durations. This is because threshold levels include a large variability of daily streamflow with a small time interval. Joint return periods of duration under extreme droughts were similar to those from univariate frequency analysis. The severities in these extreme droughts were also around the average level, but the duration was relatively long, leading to an increase in the joint return period. This result helps to effectively manage water resource preparing for the assessment on the drought risk with a long duration.

The relationships between a variety of hydro-meteorological variables and irrigation water use rates (WUR) were investigated in this study. Standardized Precipitation Index (SPI), Potential Evapotranspiration (PET), and Normalized Difference Vegetation Index (NDVI) were explored to identify the relationship with the WUR. The Yeongsan river basin, the agricultural land of which is mostly occupied by well-irrigated paddy, was used for the pilot study. Four different temporal scales of SPI-3, 6, 9, and 12 were tested, and PET was calculated using the Thornthwaite method. To calculate NDVI, the surface spectral reflectance data, which was acquired by Moderate Resolution Imaging Spectroradiometer (MODIS) equipped on the Terra satellite, were used. As a result, there was a statistically significant relationship between SPI9 and the WUR during drought periods in which negative values of SPI9 were obtained. The WUR was strongly correlated with both PET and NDVI. Compared with SPI, the variability of WUR in this study area was more sensitively affected by PET and NDVI, which can cause a potential lack of agricultural water supply. The finding of this study implies that SPI9, PET, and NDVI are the critical factors for predicting water withdrawal during drought conditions so that they can be used for irrigational water use management. Although a part of the findings of this study has been discussed by a few previous studies, this study is novel in that it quantifies the relationship between these factors using actual field observations of streamflow withdrawal for irrigation.

This study evaluates the effects of climatic conditions and drought phase on occurrence frequency (OF) of forest fire in South Korea, and suggested a deep learning-based estimator for the occurrence frequency of fire. There was a high correlation between frequency and relative humidity (RH, R = −0.663) and wind speed (WS, R = 0.532). Using these correlations, we proposed a deep learning model that can estimate the OF. Among the three deep learning models (RH-WS-AMOF, RH-AMOF, and WS-AMOF) by combining RH and WS with average monthly OF (AMOF) during 1997–2019, the RH-WS-AMOF model showed the best performance. R² and NSE were 0.838 and 0.828, respectively. The higher temperatures and drought lead to increase the potential for forest fire. Standardized precipitation evapotranspiration index was introduced for exploring the link between meteorological drought and forest fire. We confirmed that SPEI can improve the performance of the DBN based on OF estimator in spring. The framework of this study can provide a predictive model for forest fire OF combined with a weather forecast model.

The occurrence frequency of forest fires (OF) can be estimated using drought features because droughts are affected by climatic conditions. Previous studies have improved OF estimation performance by applying the meteorological drought index to climatic conditions. It is anticipated that the temperature will rise in South Korea in the future and that drought will become severe on account of climate change. The future OF is expected to change accordingly. This study used the standard precipitation index, relative humidity, and wind speed as predictor variables for a deep-learning-based model to estimate the OF. Climate change scenarios under shared socioeconomic pathways were used to estimate future OF. As a result, it was projected that the OF in the summer season will increase in the future (2071–2100). In particular, there will be a 15% increase in July compared to the current climate. A decrease in relative humidity and increase in wind speed will also affect the OF. Finally, drought severity was found to be the most influential factor on the OF among the four drought characteristics (severity, duration, intensity, and inter-arrival), considering inter-model variability across all global climate models.