The authors performed modelling of the transport of Escherichia coli (E. coli) in a tropical headwater catchment located in Lao PDR using a deep learning model and the Hydrological Simulation Program–FORTRAN (HSPF). The deep learning model was built using the long short-term memory (LSTM) technique, whereas the process-based model was constructed using the HSPF. Their results show that the LSTM provided accurate results for surface and subsurface flow, by showing 0.51 and 0.64 of Nash–Sutcliffe Efficiency (NSE), respectively, whereas the NSE values yielded by the HSPF were -0.7 and 0.59 for surface and subsurface flow. The simulated E. coli concentration from LSTM also improved, yielding an NSE of 0.35, whereas the HSPF showed an unacceptable performance, with an NSE value of -3.01. The subject is interesting, important and useful. However, there are still some key points need to be addressed. This reviewer recommends to do some revision taking into account the below comments.

1. Line 62, add g in “determining”.
2. Line65-84, although the authors have performed a good review of literature of process-based models, some latest literatures of water quality should be introduced, such as E. coli (Sowah et al., 2020; Meshesha et al., 2020), and limitations of process-based models (Wang et al., 2020, 2021).

https://www.sciencedirect.com/science/article/pii/S0048969720341917?casa_token=itdrAAzone8AAAAA:boSZBTjS_8FQY3RZ2bJz9zwQjiwpz9QOLuLvqK2iB6_CMu7NFaqmjvYcaCwPxwvLO7yw1XEQKA

https://www.sciencedirect.com/science/article/pii/S002216941931248X

https://www.sciencedirect.com/science/article/pii/S0048969720326097

https://www.mdpi.com/2073-4441/13/4/518

1. Line85-95, it is unclear what advantages DL has over process-based models. Are there any disadvantages of DL, compared to process-based models?
2. Line376-378, There are no Figs. S2a-S2c in Figure S2.
3. In Fig. S11d and S11e, a peak on 2015-08-28 was captured by both models. However, It look no data of E. Coli. This should give some explanation if no observed data.
4. It is unclear where input data sources are from for both LSTM and HSPF. Furthermore, is land use resolution same for both LSTM and HSPF?
5. Fig. 4 and Fig. 5 can be merged to remove a rainfall figure.
6. Line 182, it should be briefly described how the data has been converted to a 6 min frequency.
7. Figs. S6-S11 should be explained and discussed.

It is important and meaningful to improve prediction accuracy in modeling works. Nowadays, application of machine learning including deep learning techniques may be very promising to support conventional modeling approaches. This study constructed LSTM model to simulate surface/subsurface flow and E. coli concentration in a catchment and compared the performance with HSPF model, which is a well-known watershed model. The results are quite interesting and can be useful in scientific and practical fields. I think that this study can be considered as a publication in the journal with minor revision. Some comments are as follows.

In construction of LSTM in a catchment, this study used only meteorological data as an input to predict flow rates. An issue is that how we can consider characteristics of catchment such as land use and soil property in simulation of flow rate. I think that the basin area in this study may be relatively small. What if LSTM model application in a large-scale watershed? Is the meteorological data enough to predict flow rate? I think that this discussion can be very informative to readers in LSTM application to watershed scale.

Line 29, full name is needed for “PDR”

Line 52, what is the meaning of “less dangerous than other pathogens”?

In study site description, basin area is needed.

Line 167, what is the meaning of “rewrote” Did you modify the source code? Rephrase it.

Line 192-193 and 376-378, It is difficult to understand scenario 1 and scenario 2. Scenario 1 is land use change with same E. coli loading (Fig. S1-a and b) and scenario 2 is land use change with variable E. coli loading in terms of land use (Fig. S1-a and c))? It is confusing.

Line 265, among the 10 most sensitive parameters? 10 variables are equally sensitive?

Table 2 and line 313, number of optimal batch size and lookback steps are mismatched between the table and sentence. 128 vs 100 and 50 vs 5 h

Figure 8, what is (a) and (b) in the figure? (a) July 15 and (b) August 1?

Line 362 – 370, where is minmax and logarithmic transformation from? There was not any mention about application of minmax and logarithmic transformation in method. All E. coli simulations was based on logarithmic?

Line 376-378, Referencing of the figure in the sentences may be wrong. Not Fig. S2 but Fig. S1. Check it..