Arifin et al look at three different ways of measuring groundwater storage dynamics in the Lower Kutai Basin (LKB) of Indonesia: the GRACE budget residual approach, data from the existing (few) local piezometric sensors, and the possibility to use the sensors from pumping wells. The study is motivated by the fact that Indonesia’s new capital Nusantera is under development in the LKB which will lead to increased pressure on water resources.

The main message of the paper seems to me that currently neither of the three approaches can provide a reliable assessment of groundwater resource variability, and the government should think about rolling out a comprehensive measuring network. In my understanding the region is simply too small to be resolved well in GRACE (as the authors know very well), and there are too few in-situ sensors. The authors suggest that about 30% of their GRACE-derived dGWS ensemble  timeseries are not usable since they provide unphysical results, and they discuss many timeseries with correlations about 0.3 as „weakly correlated“ – I think given that timeseries here are not very long on climate timescales we can say that such low correlation means nearly uncorrelated.

On balance, my judgement is that the topic is very relevant and there seems a pressing need to improve the monitoring system, however the quantitative evidence that is presented is a bit weak and the logic is not straightforward. It is clear that the region is not well suited for a thorough assessment of the GRACE data, or the GRACE residual approach for groundwater, since the basin is too small, the GRACE data are affected by ocean signals, and there are too few in-situ sensors. I am missing a discussion of related papers that discuss bigger inland regions of similar hydrogeophysics but better monitoring network with a similar approach. If I am right about the authors‘ intention, I am missing a much more extended discussion of how a monitoring network could look like, how many piezometers or observation wells would be needed, how could they be distributed to connect to the GRACE data. Without this I feel the paper misses a bit its mark.

Further issues

• Title: I don’t think the authors really succeed in „reconciling GRACE to piezometry“. This should be reflected in the title.
• The authors discuss the challenges of the GRACE residual approach, e.g. we need to know the soil moisture contribution. But we also need to know the contribution by surface water storage variability, water levels in wetlands and rivers, lakes and artificial reservoirs. The authors are aware of this. In my understanding, wetlands cover a significant share of land in Indonesia including the LKB. We need a quantitative discussion of the error that could be introduced here. WaterGAP may not be very good at this – WaterGAP does not simulate level changes in reservoirs, for example. And I suggest that the authors look into ways of quantifying this contribution, e.g. from remote sensing and/or radar altimetry
• Similar for hydrology, in my understanding in Indonesia peatlands are abundant and peatland hydrology may not be well represented in GLDAS or WaterGAP. What is the anticipated error? Or isn't this the case in the LKB?
• Similar, the Makassar and Sulawesi Straits are part of an ocean region that experiences above-average variability and sea level rise. Ocean signals are reduced in the standard GRACE data products but we know that the MPI reference ocean model which is forced by the atmosphere only does not capture all mass signals, and in coastal areas this may introduce a significant error irrespectively what mascon products were used. Put in simple words, the GRACE data may include real ocean mass change at least on the seasonal timescale here that could be misinterpreted as a groundwater signal. This is an error source that is not relevant for the average hydrological inland basin, but for Indonesia it may be very well. I’m missing a quantitative discussion here.
• The authors should also try to estimate the error in the piezometric analysis introduced by not knowing the local yield factor. They could take a range of possible yield factors from the hydrogeophysics maps or from publications and do a best/worst-case assessment.
• The comparison of the GRACE product error with three mascon solutions appears not very robust. I would very much recommend that the authors consider at least one product based on spherical harmonics. It is an unproven claim that mascon solutions are better or more suitable to coastal regions. They are easy to apply but this is not the same as being more appropriate or having less errors.
• I’m missing a map of the aquifer systems in the LKB, in particular are these aquifers extending under the sea? The dashed line in Fig. 1 suggests this. Would coastal groundwater withdrawal then cause storage changes in the marine part of the aquifer? Would that be expected to become visible in GRACE as well? Isn’t this suppressed by the mascon approach? These are just ideas, I am not an expert in this regions. Again, most GRACE groundwater studies need not worry about this, but the region here is particularly challenging.
• There are GRACE-assimilating hydrology model runs from NASA/Goddard and from the University of Bonn, Germany, and these provide the groundwater storage change at resolution between 30 and 50 km. Why don’t the authors look into these data sets or add them to their ensembles?
• The authors mention several times that poor GRACE data or poor corrections in the residual approach lead to arithmetic problems. This is true, but an arithmetic problem is just a symptom that the data are poor. In other words, even if no arithmetic problem occurs this may be just by chance, and we should not trust the data. I think this needs to be made clear in a scientific paper.
• Potential instrumental problems of the piezometric sensors should be discussed. I liked the part on the correlation between these data and the rainfall data, as it tells about the sensitivity. But this instigates trust for the short timescales only. What about the seasonal and interannual timescales, are there biases to be expected?

Overall, the error budgeting needs more detail and quantification. This is, as I said earlier, partly a consequence of the fact that the LKB region is a particularly challenging one for GRACE. That also means if the authors succeed to make their case, this could be a breakthrough in the application of GRACE data, so it is really worth to dig deeper. At the moment, results appear somewhat inconclusive and the message is not too clear. I suggest that the study logic – what is the underlying hypothesis, what exactly do we expect from GRACE at such small scales, why looking at the piezo-rainfall correlation, why looking at ENSO – is explained right at the start.

Best regards, Jürgen Kusche

Issues with scale:  Have the authors thought about GRACEs coarse footprint? The basin, measuring about 23,000 km², is significantly less than GRACEs native resolution of around 90,000 to 300,000 km². Why is there the implication that interpolating GRACE to 0.25°–0.5° would add new detail? The ms explains that the "downscaled" delTWS on 0.25°–0.5° does not significantly change the results, since it correlates strongly with r≈0.85–1 with the initial 3° mascon and whole-Borneo signal. That is, the authors are basically looking at the same broad signal. Can we discuss whether this adds to our knowledge, or could this section be reworded or revised for simplicity? Could explain why these finer grids matter, as they are not independent signals.

Data and uncertainty:  The delGWS estimates are based on the subtraction of soil and surface water models from GRACE TWS and thus inherit all related uncertainties. You are using several GLDAS models and WGHM and reporting their differences. But it would be helpful to discuss the unreliability of those in this situation. For example, you noticed that one of the GLDAS runoff datasets was "implausible," and you discarded it – this says that model results can not be completely trusted. Did you compare GLDAS soil moisture or WGHM surface water with any local observations? If that is the scenario, it would be helpful if this could be mentioned. Your technique for producing 36 GWS "realizations" through combinations of GRACE and model ensembles is resourceful. Nevertheless, the selection to eliminate some 30% of them as "implausible". Have you thought about reducing model biases instead? You are correct to have noted that these arithmetic anomalies mainly appear when delTWS is smaller than delSMS plus delSWS in the wet months, or the reverse during the dry months. Overlooking those months might affect your trends. It would be useful to have a numerical estimate of the effect of the filtering on the results. You show that removing the outliers improves the TWS–GWS correlation. It might be helpful to provide error estimates for delGWS or to note that the "mean plausible" series is just one model of noisy data? Assumptions of the model:  Along these lines, it should be pointed out that GLDAS and WGHM dont account for groundwater pumping.

There appear to be signs that some wells are experiencing decline due to abstraction, but the GWS calculations dont account for this. If water pumping is stopped from the basin (and not simply recharging local surface water), GRACE would find a deficit. But the average GRACE DEL GWS trend indicates a very gentle positive change, even when one well went significantly deeper and had a local decrease. How do you resolve that? The conclusion blames the discrepancy on "withdrawals," but it would be nice to supply some quantification or at least investigate whether the pumping is too localized to be seen in GRACE. Are there other factors to take into account, such as groundwater draining from the basin by rivers or recharge assumptions? 

ENSO correlations: The moderate values of delTWS/delSMS versus ENSO indices seem to make sense and are in line with results from other studies, but care should be taken with the short time series. Did you test for statistical significance or account for autocorrelation? Has it been possible for you to investigate correlating detrended or deseasoned data? It might improve your case if you think about stripping the seasonal cycle before associating it with ENSO. In some way or other, the physical explanation works out: large El Niño events dry the basin, which in turn amplifies the GRACE delGWS drops. But it is really the soil moisture changes that are having an effect on TWS or it seems that delGWS is much less sensitive? It may be beneficial to emphasize how the soils and floodplains contribute significantly to the GRACE signal here. 

Piezometer comparisons: I do like the attempt to incorporate groundwater wells, but please be careful. In the first place, matching GRACE delGWS (in cm over the whole basin) with a small set of point measurements is of course an approximation. Your conclusion shows a "moderate" correlation , which is reasonable. Claiming, however, that the time series "align with groundwater-level" (Abstract) goes too far. For example, GRACE indicates a small upward trend in general, while the deep well has a downward trend. How are the authors confirming that these are actually deep observation wells? Furthermore, head to storage conversion demands a certain yield values, which has not been discussed. Thus, the comparison is qualitatively of the rise/fall type instead of being concerned with quantitative volume. Please the range of specific yield you are assuming, or flagging this as a possible source of uncertainty? without knowing aquifer properties, its not easy to relate a GRACE-derived cm change to an observed m depth change.

Monitoring groundwater: Filtering for the minimum daily level is an able strategy and seems to track recharge. As a reader, I am interested in the strength of that: can pump-off intervals be incorrectly interpreted, or can the lowest daily level nevertheless be influenced by slower pumping rates? Can you indicate how well these loggers are calibrated? A mention of data quality control, would be helpful.

Clarity and flow: The writing is generally easy to follow, although there are spots where its a bit dense. The methods section is very thorough, although it could do with more obvious labelling. It may be a good idea to divide the GRACE discussion from the GLDAS/WGHM. Some of the sentences are also very long – try breaking them up to make them easier to read, especially in the introduction and methods sections.

This paper shows a worthy effort in tackling a difficult problem, however, I am concerned about the 23,000 km² study area. The authors have used a wide range of data and monitoring methods. I would recommend significant revision, moderating any very assertive conclusions, especially about delGWS match with wells etc, and being forthright about the large uncertainties involved. I look forward to reading a revised version.