Interactive comment on “Modelling stream flow with a discrete rainfall–runoff model and 37 GHz PDBT microwave observations: the Xiangjiang River basin case study” by Haolu Shang et al

The authors may need to pay more attentions to the overfitting issue. For example, the accuracy of their rainfall-runoff model is improved when they increase the longest durations of antecedent precipitation. The authors interpret this result in terms of the basin characteristics. However, when the authors increase the longest duration of antecedent precipitation, the performance of the linear regression should be always improved since the number of weights (unknown parameters) increases. It is possible that the improvement of the rainfall-runoff model in this paper is not related to any real rainfall-runoff processes. The authors may need to check if their statistical model does not overfit to the data using the cross-validation analysis. In the second and third implementation, the model has better performances with the larger durations of antecedent precipitation in the calibration period (Figures 6 and 7) while their performances are degraded by increasing the duration in the validation period (Figure 12). In the case of second and third models, increasing the duration causes the overfit to the data.