The forward simulation of radar reflectivity requires details of clouds and precipitation from general circulation models (GCMs). But such details are represented as sub-grid processes that involve parameterizations and assumptions about the spatial coverage and thus depend on the GCM. In this research, we propose the use of a statistical method to generate sub-grid precipitation for generic use. In addition, the proposed method can be used to provide uncertainty estimates on the signals. The sub-grid variability is obtained from simulation with a global storm-resolving model called NICAM (non-hydrostatic icosahedral atmospheric model). The proposed method first generates precipitation probabilities for the possible scenarios and then sub-grid precipitation rates are generated from the generalized gamma distribution for the given cloud fraction and grid-scale precipitation rates. Compared to the standard method (which neglects the probabilities) that overestimates the precipitation fraction, our method well reproduces the NICAM dataset profiles of both the precipitation fraction and the radar-based cloud fraction. The in-cloud signal frequencies are also reproduced, although less accurately over a tropical region. Inclusion of sub-grid variability in precipitation rates was particularly important for the tropical region to obtain agreement of the precipitation fraction. Application of the two methods to a GCM shows it to have a robust bias for low-level liquid clouds. The proposed method can be used to identify uncertainty in the signals associated with sub-grid variability in the precipitation processes, indicating an effective way to use a global storm-resolving model to evaluate conventional GCMs.

The increasing frequency of devastating floods from heavy rainfall associated with climate change has made river stage prediction more important. For steep, forest-covered mountainous watersheds, deep learning models may improve prediction of river stages from rainfall. Here we use the framework of multilayer perceptron (MLP) neural networks to develop such a river stage model. The MLP is constructed for the Shimanto river, which lies in southwestern Japan under a mild, rain-heavy climate. Our input for stage estimation, as well as prediction, is long-term rainfall time series. With a one-year time series of rainfall, the model estimates the stage with 50 cm RMSE for about 10 m of stage peaks as well as accurately simulate stage-time fluctuations. Furthermore, the forecast model can predict the stage without rainfall forecasts up to three hours ahead. To estimate the base flow stages as well as flood peaks with high precision we find the rainfall time series should be at least one year. This indicates that the use of a long rainfall time series enables one to model the contributions of ground water and evaporation. Given that the delay between the arrival time of rainfall at a rain-gauge to the outlet change is well simulated, the physical concepts of runoff appear to be soundly embedded in the MLP.