Abstract. In general, there are no
long-term meteorological or hydrological data available for karst river
basins. The lack of rainfall data is a great challenge that hinders the
development of hydrological models. Quantitative precipitation estimates
(QPEs) based on weather satellites offer a potential method by which rainfall
data in karst areas could be obtained. Furthermore, coupling QPEs with a
distributed hydrological model has the potential to improve the precision of
flood predictions in large karst watersheds. Estimating precipitation from
remotely sensed information using an artificial neural network-cloud
classification system (PERSIANN-CCS) is a type of QPE technology based on
satellites that has achieved broad research results worldwide. However, only
a few studies on PERSIANN-CCS QPEs have occurred in large karst basins, and
the accuracy is generally poor in terms of practical applications. This paper
studied the feasibility of coupling a fully physically based distributed
hydrological model, i.e., the Liuxihe model, with PERSIANN-CCS QPEs for
predicting floods in a large river basin, i.e., the Liujiang karst river
basin, which has a watershed area of 58 270 km2, in southern China.
The model structure and function require further refinement to suit the karst
basins. For instance, the sub-basins in this paper are divided into many
karst hydrology response units (KHRUs) to ensure that the model structure is
adequately refined for karst areas. In addition, the convergence of the
underground runoff calculation method within the original Liuxihe model is
changed to suit the karst water-bearing media, and the Muskingum routing
method is used in the model to calculate the underground runoff in this
study. Additionally, the epikarst zone, as a distinctive structure of the
KHRU, is carefully considered in the model. The result of the QPEs shows that
compared with the observed precipitation measured by a rain gauge, the
distribution of precipitation predicted by the PERSIANN-CCS QPEs was very
similar. However, the quantity of precipitation predicted by the PERSIANN-CCS
QPEs was smaller. A post-processing method is proposed to revise the products
of the PERSIANN-CCS QPEs. The karst flood simulation results show that
coupling the post-processed PERSIANN-CCS QPEs with the Liuxihe model has a
better performance relative to the result based on the initial PERSIANN-CCS
QPEs. Moreover, the performance of the coupled model largely improves with
parameter re-optimization via the post-processed PERSIANN-CCS QPEs. The
average values of the six evaluation indices change as follows: the
Nash–Sutcliffe coefficient increases by 14 %, the correlation
coefficient increases by 15 %, the process relative error decreases by
8 %, the peak flow relative error decreases by 18 %, the water
balance coefficient increases by 8 %, and the peak flow time error
displays a 5 h decrease. Among these parameters, the peak flow relative
error shows the greatest improvement; thus, these parameters are of the
greatest concern for flood prediction. The rational flood simulation results
from the coupled model provide a great practical application prospect for
flood prediction in large karst river basins.