Search

Karst spring water dynamic characteristics and its response to atmospheric precipitation are of great significance for water resources utilization under the background of climate change. This paper selects Longzici spring area, North China, as the study area. Based on a long series of spring water flow and precipitation data, the dynamic characteristics of spring flow were analyzed and the numerical simulation of the groundwater flow model was established. The results show that the groundwater kept the sustained decline over the past decades which is in a negative equilibrium state, with a storage variable of − 2.26 million m3/year. The sensitivity of spring flow to precipitation under different precipitation scenarios shows that the water level changes in the recharge and drainage areas are similar about (3–5 cm) and slightly larger than that in the runoff area(1.5 cm) when minimum rainfall (287.24 mm) happens. When the precipitation is at its maximum (867.66 mm), the water level change in the runoff area can reach 95 cm which is much larger than those in the recharge and discharge areas. The results indicate that Longzici karst spring has a relatively good regulation water resource capacity and the runoff area is more sensitive which plays an important role in response to climate change. [ABSTRACT FROM AUTHOR]

The examination and integration of numerical forecast products are essential for using and developing numerical forecasts and hydrological forecasts. In this paper, the control forecast products from 2010 to 2014 of four model data (China Meteorological Administration (CMA), the National Centers for Environmental Prediction (NCEP), the European Centre for Medium-Range Weather Forecasts (ECMWF), and the United Kingdom Meteorological Office (UKMO)) from The Interactive Grand Global Ensemble (TIGGE) data center were evaluated comprehensively. On this basis, a study of runoff forecasting based on multi-model (multiple regression (MR), random forest (RF), and convolutional neural network-gradient boosting decision tree (CNN-GBDT)) precipitation integration is carried out. The results show that the CMA model performs the worst, while the other models have their advantages and disadvantages in different evaluation indexes. Compared with the single-index optimal model, CMA model had a higher root-mean-square error (RMSE) of 18.4%, and a lower determination coefficient (R2) of 14.7%, respectively. The integration of multiple numerical forecast information is better than that of a single model, and CNN-GBDT method is superior to the multiple regression method and random forest method in improving the precision of rainfall forecast. Compared with the original model, the RMSE decreases by 13.1 ~ 27.9%, PO decreases to 0.538 at heavy rainfall, and the R2 increases by 4 ~ 15.2%, but the degree of improvement decreases gradually with the increase in rainfall order. The method of multi-model ensemble rainfall forecasting based on a machine learning model is feasible and can improve the accuracy of short-term rainfall forecasting. The runoff forecast based on multi-model precipitation integration has been improved, and NSE increases from 0.88 to 0.935, but there is still great uncertainty about flood peaks during the flood season. [ABSTRACT FROM AUTHOR]

Accurate gridded precipitation data with high spatial and temporal scales are required for diverse studies such as climatology, meteorology, and hydrology. Currently, one of the sources of global precipitation estimation is the satellite-based precipitation estimate products. Nonetheless, their spatial resolution is often too coarse for usage in local region and basin scales or for parameterizing of meteorological and hydrological models at regional scales. In the present paper, a reconstruction method of satellite-based monthly precipitation was developed to attain improved pixel-based precipitation data with high spatial resolution on Golestan province in Northern Iran. In this endeavor, we considered the spatially heterogeneous relationships between tropical rainfall measuring mission (TRMM) precipitation and environmental variables utilizing the moving-window regression methods, the geographically weighted regression (GWR) and the mixed geographically weighted regression (MGWR) models. By in situ observations from rain gauges in the study area, the calibration and validation were performed, and the following conclusions were derived: (1) the proposed procedure had the ability to enhance both the spatial resolution and accuracy of satellite-based precipitation estimates; (2) the monthly reconstructed precipitation using the GWR model (CC = 0.69, bias = 0.75) and using the MGWR model (CC = 0.72, bias = 0.64) outperformed the TRMM-3B43V7 data (CC = 0.58, bias = 0.84) against ground observations; (3) this research offered a potential solution for producing gridded precipitation estimates at high spatial resolution. [ABSTRACT FROM AUTHOR]