{"id":"35543f63-249a-4446-978d-59e6b77ad326","arxiv_id":"2505.05346","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"GNSS-derived precipitable water vapour at the H.E.S.S. site agrees with a 210 GHz water vapour radiometer to within about 0.34 mm, and to 0.15 mm when a locally calibrated temperature model is used.","lead":"This paper compares water vapour measurements from a GPS-based station at the H.E.S.S. site in Namibia with those from a 210 GHz water vapour radiometer, finding 98% correlation and a mean offset of 0.34 mm. The result matters for picking a site for the planned Africa Millimetre Telescope, where water vapour blocks millimetre-wavelength observations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The WVR reference is calibrated with MERRA-2, which the paper itself reports disagrees with GNSS PWV by 7.45% at this site; until that bias is ruled out, the sub-mm offsets show consistency with a model, not absolute accuracy.","rationale":"The paper performs useful site-testing work: it installs a WVR alongside the GNSS station, applies standard retrieval methods, and reports a high correlation with a small offset. The NGL-based comparison is not itself circular, and the public NGL products and the established GNSS PWV equations (Eqs. 2-5, 16-17) provide some independent grounding. However, the central accuracy claim is anchored to the WVR as an absolute reference, and the WVR PWV scale is set by MERRA-2 in Section 2.1. Because the paper's own companion work reports a 7.45% MERRA-2 versus GNSS difference at this site, the reference scale is not established to the claimed few-tenths-of-a-millimetre level. The local Tm improvement in Section 3.2 and Appendix A1 is calibrated against the same WVR PWV and then validated against it, so the 0.15 mm offset is an in-sample consistency check. These issues do not invalidate the relative agreement, but they prevent the paper from demonstrating absolute accuracy. The conditional verdict remains appropriate.","tokens_in":12116,"tokens_out":7011,"duration_ms":74640,"concrete_test":"Recompute the WVR opacity-to-PWV conversion using an independent atmospheric product (ERA5 reanalysis or a dedicated radiosonde campaign at the H.E.S.S. site) instead of the MERRA-2 fit in Table 1, then re-run the Section 3.1 comparison and Table 2 offsets. If the NGL-based offset changes by more than about 0.3 mm, or the WVR PWV series tracks MERRA-2 PWV more closely than it tracks GNSS PWV, the reported agreement is an artifact of the MERRA-2-calibrated reference rather than absolute accuracy.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing assumption is that the 210 GHz WVR provides an absolute PWV reference. In Section 2.1, WVR opacity is converted to PWV with a quadratic fitted to 24 years of MERRA-2 reanalysis at the H.E.S.S. site (Eq. 1, Table 1), and the tipping-curve retrieval assumes an isothermal atmosphere (Eqs. 12-15). Any bias in the MERRA-2 opacity-PWV relation or in the isothermal approximation enters every WVR PWV value, and hence both offsets in Table 2 and the local Tm model of Eq. 20, which is calibrated against WVR PWV in Appendix A1. The paper itself cites Frans et al. (2025), who found a 92% correlation and a 7.45% difference between MERRA-2 PWV and GNSS PWV at this same site; for the cited H.E.S.S. median PWV of 14.27 mm, that is about 1.06 mm, three times the reported 0.34 mm offset. Section 3 never reconciles this tension. If MERRA-2 is biased at H.E.S.S., the WVR PWV inherits that bias and the 'accuracy' claim reduces to agreement between GNSS and a MERRA-2-calibrated radiometer rather than absolute PWV accuracy. The 0.15 mm result is additionally weakened because the Tm model is derived from the same WVR data used for the comparison, so it is an in-sample calibration check, not independent validation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper validates GNSS-derived precipitable water vapour (PWV) at the H.E.S.S. site against PWV from a 210 GHz water vapour radiometer (WVR). The authors build a quadratic PWV–opacity conversion at 210 GHz from 24 years of MERRA-2 data (Section 2.1, Table 1), compute GNSS PWV using NGL zenith delays, on-site pressure, and either NGL or locally derived weighted-mean temperature (Tm), and compare 15-minute averages over April–August 2024. After removing 4.75% of points via a 3-sigma difference cutoff, they report a 98% correlation and a 0.34 mm offset with NGL Tm, improving to 0.15 mm with a locally derived piecewise Tm–Ts model (Eq. 20). The paper concludes that the GNSS stations at H.E.S.S. and Gamsberg give reliable PWV to within a few tenths of a millimetre.","tokens_in":12512,"tokens_out":2813,"duration_ms":30167,"significance":"If the claimed accuracy holds, the study provides a useful and inexpensive validation of GNSS PWV for site testing of the Africa Millimetre Telescope, and the local Tm–Ts relation (Eq. 20) is a practical product for the region. The paper also demonstrates a transparent reduction chain from raw radiometer voltages to PWV. The strength of the work lies in the direct side-by-side comparison using identical instrumentation and the explicit modelling of the PWV–opacity relation. However, the central 'accuracy' claim is contingent on the WVR being an absolute reference, which is not established: the WVR conversion is anchored to MERRA-2 reanalysis, and the paper itself cites a 7.45% MERRA-2–GNSS PWV difference at this site. The 0.15 mm improvement from the local Tm model is an in-sample calibration result because the Tm model is fitted to the same WVR PWV used in the comparison. These issues do not invalidate the comparison as a consistency check, but they do prevent the current version from supporting the absolute-accuracy statement in the title and abstract.","major_comments":[{"comment":"The WVR is used as the reference truth, but its PWV is obtained by converting zenith opacity with a quadratic fitted to 24 years of MERRA-2 data (Eq. 1, Table 1). The paper itself reports in the Introduction that MERRA-2 PWV differs from GNSS PWV at this site by 7.45% (Frans et al. 2025). This means the WVR is effectively a MERRA-2-calibrated instrument, so the reported 0.34 mm and 0.15 mm offsets measure agreement between GNSS and a MERRA-2-based model, not absolute PWV accuracy. The authors should quantify the sensitivity of the offsets to plausible biases in the MERRA-2 PWV–opacity relation (e.g., propagate the 7.45% difference into WVR PWV and recompute the offsets), or provide an independent calibration (radiosonde, or comparison with another established radiometer). Without this, the title's claim of 'accuracy' is not supported.","section":"Section 2.1 and Section 3.1"},{"comment":"The local Tm model (Eq. 20) is derived by regressing the ZWD–PWV slope, obtained from the 210 GHz WVR PWV and GNSS ZWD, against surface temperature (Figure A1 and Figure 7). The comparison in Figure A2 then uses GNSS PWV computed with this same Tm model against the same WVR PWV. This is an in-sample calibration check: the reduction in offset from 0.34 mm to 0.15 mm is expected by construction and does not constitute independent validation. The authors should validate Eq. 20 with withheld data (e.g., a temporal hold-out split) or against an independent Tm source (e.g., radiosonde or an NWP-based Tm), and report the comparison separately for the fitting and validation subsets.","section":"Appendix A1 and Section 3.2"},{"comment":"The 3-sigma flagging procedure removes 4.75% of the data based on the very difference being analysed, with thresholds of -3.52 mm and 4.21 mm. The paper states that the flagged points are radiometer artifacts, but this is an assumption; no instrument log or independent diagnostic is provided to show these are not real PWV variability. Moreover, no uncertainties are reported for the correlation, offset, or standard deviation (e.g., standard error of the mean offset, confidence intervals), and the effective sample size after 15-minute averaging and temporal correlation is not discussed. The authors should report uncertainties and show that the conclusions are robust to the outlier cutoff (e.g., recompute statistics without any flagging and with different sigma thresholds).","section":"Section 3.1"}],"minor_comments":[{"comment":"The text contains several typographical errors and missing spaces (e.g., 'calculated' and 'calculatethe' in the abstract; 'insitu' for 'in situ'). A careful proofread is needed.","section":"Abstract and throughout"},{"comment":"Figure A1 captions state 'RH < 40 %' for all panels, but Section A1 says measurements with relative humidity greater than 40% were used. This appears to be a typo and should be corrected to 'RH > 40 %'.","section":"Figure A1 captions"},{"comment":"The caption for Figure 2 describes red and blue lines, but the colours are not used consistently in the figure text; also, the units of the polynomial coefficients in Table 1 should be stated explicitly (e.g., A in mm, B in mm, C in mm for the PWV vs tau0 fit).","section":"Figure 2 and Table 1"},{"comment":"The reported 'correlation of 98%' is the Pearson coefficient from the linear fit on the cleaned data; the paper should state this explicitly and also report the coefficient for the raw (uncleaned) data to demonstrate the effect of the flagging.","section":"Section 3.1"},{"comment":"The statement that offsets of 0.34 mm and 0.15 mm are 'essentially negligible' is made without a quantitative criterion for what is acceptable for AMT observations. Please relate the offsets to a PWV accuracy requirement, for example the typical phase-noise or opacity tolerance at 1 mm or 0.8 mm wavelengths.","section":"Section 3.2 and Conclusions"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a practical and timely question for AMT site selection, and the data collection effort is commendable. The main issue is that the central accuracy claim depends on an unvalidated WVR absolute calibration and on an in-sample Tm model. I do not see this as an irreparable flaw, but the authors must reframe the conclusions as a consistency check or supply the missing calibration/validation. The scope fits RASTI well. I would suggest the editor require the authors to address the three major comments before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper is a site-testing validation that transfers established GNSS-PWV recipes to the H.E.S.S. site. The useful part is the NGL-based comparison: 98% correlation and 0.34 mm offset against a 210 GHz WVR. The 0.15 mm offset from the local Tm model is not evidence, because that model is fitted to the same WVR data used for the comparison.\n\nWhat is genuinely new: a local Tm–Ts relation (piecewise, equation 20) for the Khomas Highlands, applicable to the Gamsberg site, and a new GNSS-versus-WVR validation at H.E.S.S. The method follows Sugiyama et al. and Combrink, and it is properly cited. The paper is straightforward and gives a useful cross-check for the earlier Frans et al. GNSS campaign.\n\nSoft spots. The WVR is treated as ground truth, but its PWV comes from a quadratic opacity–PWV relation fitted to 24 years of MERRA-2, plus an isothermal tipping-curve assumption. The paper's own companion work (Frans et al. 2025) reports a 7.45% difference between MERRA-2 PWV and GNSS PWV at precisely this site, and Section 3 never reconciles that. If MERRA-2 is biased at H.E.S.S., the WVR PWV inherits that bias, and the offsets measure agreement between GNSS and a MERRA-2-calibrated radiometer, not absolute accuracy. The 0.15 mm result is circular in the way the stress-test note describes: the Tm model is calibrated against the WVR PWV, so the improved offset is expected. The 3-sigma cut removes 4.75% of the data, and no uncertainties are given for the offsets or the fit parameters. These are real limitations, but they are not fatal to the paper's core purpose.\n\nThe NGL-based comparison is decent evidence of relative agreement: high correlation, slope near 1.06, and sub-mm scatter. With revisions that reframe the accuracy claim, add error propagation, and address the MERRA-2 tension, this is a publishable site-testing paper. The audience is the AMT project and the site-testing community.\n\nRecommendation: send it to peer review. It deserves a serious referee. I would ask the authors to discuss the MERRA-2 bias, provide uncertainties, and explicitly state that the Tm-model comparison is a calibration check rather than an independent validation.","headline":"A competent site-testing cross-check that overstates its absolute accuracy: the 0.15 mm offset is in-sample, and the WVR reference inherits MERRA-2's own known bias at this site.","tokens_in":13082,"tokens_out":2522,"would_cite":false,"duration_ms":27071,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"GNSS-derived precipitable water vapour at the H.E.S.S. site agrees with a 210 GHz radiometer to 0.34 mm, improving to 0.15 mm when a locally calibrated weighted-mean temperature is used.","keywords":["precipitable water vapour","GNSS","210 GHz water vapour radiometer","weighted mean temperature","H.E.S.S. site","Gamsberg","Africa Millimetre Telescope","millimetre site testing"],"falsifier":"Co-locate a radiosonde balloon programme at the H.E.S.S. site across at least one wet and one dry season and compare its PWV profiles with simultaneous GNSS and 210 GHz radiometer PWV; a systematic offset between the radiosonde and both instruments that follows the sign of the MERRA-2-based conversion would show that the GNSS-radiometer agreement is a shared-model effect, not absolute accuracy.","tokens_in":11901,"feed_emoji":"🌫️","tokens_out":11809,"duration_ms":102755,"temperature":0.7,"pith_summary":"This paper asks whether a GNSS station can be trusted to measure precipitable water vapour for millimetre and submillimetre astronomy, where water vapour is the main source of atmospheric opacity. The answer attempted here is a head-to-head test at the H.E.S.S. site in Namibia: PWV inferred from the GNSS station is compared with PWV from a 210 GHz water vapour radiometer over the shared observing period in 2024. The paper reports a 98% correlation and a mean offset of 0.34 mm when the standard NGL interpolated weighted-mean temperature is used, and 0.15 mm when a locally calibrated weighted-mean-temperature model is used instead. A sympathetic reading is that this makes the GNSS-based PWV values previously published for H.E.S.S. and the nearby Gamsberg mountain reliable to a few tenths of a millimetre, which matters for deciding whether either site can host the Africa Millimetre Telescope.","feed_headline":"GNSS water-vapour readings match a radiometer to 0.15 mm","feed_subtitle":"The agreement validates GNSS as a reliable site-testing tool for the planned Africa Millimetre Telescope.","key_machinery":"The load-bearing machinery is a set of conversions between three quantities: radiometer opacity, PWV, and GNSS wet delay. The radiometer obtains zenith opacity $\\tau_0$ at 210 GHz by a tipping-curve method that assumes an isothermal atmosphere; a quadratic fit to 24 years of MERRA-2 data at the site, $PWV = A\\tau_0^2 + B\\tau_0 + C$ with $A = -2.601052$ mm, $B = 22.0007$ mm, $C = -0.3455$ mm, turns that opacity into PWV. The GNSS side uses the standard relation $PWV = H(T_m)\\,ZWD$, with the zenith wet delay obtained from NGL products and on-site pressure, and the new ingredient is a piecewise local weighted-mean-temperature model $T_m = 7.03\\,T_s - 1742.64$ K for $T_s \\le 290$ K and $T_m = -2.28\\,T_s + 957.11$ K for $T_s > 290$ K. Calibrating that model requires pairing radiometer PWV with GNSS ZWD and surface temperature only when relative humidity is above 40%, and a 3-$\\sigma$ filter on the difference between the two instruments removes 4.75% of the radiometer points as unphysical scatter.","core_discovery":"On the paper's own terms, the discovery is that PWV from a GNSS station can be made essentially interchangeable with PWV from a 210 GHz radiometer: a correlation of 0.98, a standard deviation of 1.29 mm, and a mean offset of 0.34 mm with the NGL interpolated weighted-mean temperature, improving to 0.15 mm with a piecewise $T_m(T_s)$ model derived on site. The radiometer's 210 GHz opacity is converted to PWV through a quadratic fit built from 24 years of MERRA-2 data at the site, and the GNSS PWV is converted back to opacity through the inverse fit, so the comparison is carried by two model-based conversions meeting in the middle. Because the two GNSS stations are identical and processed in the same way, the paper asserts that the calibration transfers to the Gamsberg station, meaning both candidate sites for the Africa Millimetre Telescope now have PWV measurements with a stated agreement of a few tenths of a millimetre.","pith_inferences":["Because the radiometer's PWV scale and the NGL temperature products both descend from reanalysis data, the close GNSS-WVR agreement is best read as mutual consistency; an independent radiosonde campaign would certify absolute PWV.","The break in the $T_m(T_s)$ relation near 290 K suggests two distinct air-mass or seasonal regimes; testing the same piecewise form at Gamsberg and other southern African stations would show whether the calibration transfers or must be locally retuned.","The paper's dual PWV-opacity fits provide a direct way to translate GNSS PWV into 210 GHz opacity, which could eventually feed real-time opacity forecasts used for scheduling millimetre and submillimetre observations and for correcting interferometric phases."],"forward_implications":["The GNSS PWV values previously reported for H.E.S.S. and Gamsberg (median 14.27 mm and 9.25 mm) can be treated as accurate to within a few tenths of a millimetre, strengthening the earlier conclusion that Gamsberg is the drier site.","A GNSS station plus an on-site weather station can serve as a continuous, low-cost PWV monitor for the Africa Millimetre Telescope, reducing the need to keep a radiometer at the site.","With the local $T_m(T_s)$ model, GNSS PWV can be computed using only GNSS data, on-site pressure, and surface temperature, without relying on interpolated NGL temperature products.","The 0.15-0.34 mm offsets are smaller than the 0.64 mm offset reported for the Atacama comparison, suggesting the same GNSS method can meet or beat the accuracy achieved at established millimetre sites."],"supporting_citations":[{"why":"Supplies the GNSS PWV dataset and site characterisation for H.E.S.S. and Gamsberg that the present comparison validates.","marker":"Frans et al. 2025"},{"why":"Provides the T_m(T_s) modelling procedure and the Atacama 0.64 mm offset that serves as the comparison benchmark.","marker":"Sugiyama et al. 2024"},{"why":"Defines the tipping-curve method the 210 GHz WVR uses to turn sky-brightness measurements into zenith opacity.","marker":"Hiriart et al. 1997"},{"why":"Provides the NGL GNSS products (ZTD and interpolated T_m) from which the GNSS PWV is calculated.","marker":"Blewitt et al. 2018"},{"why":"Supplies the physical relation and constants connecting zenith wet delay to PWV through the weighted-mean temperature.","marker":"Askne & Nordius 1987"},{"why":"Earlier validation of GNSS PWV against VLBI and a water vapour radiometer that motivates using GNSS as a PWV sensor.","marker":"Combrink 2006"},{"why":"Establishes the linear T_m-T_s relationship that the piecewise local model adapts to the H.E.S.S. site.","marker":"Bevis et al. 1992"}],"fun_headline_variants":["GNSS water vapour matches 210-GHz radiometer to 0.15 mm","On-site weather data tighten GNSS water vapour to 0.15 mm","GNSS water vapour: 98% match with 210-GHz radiometer","Radiometer-accurate GNSS water vapour: offsets drop to 0.15 mm","Telescope-site GNSS water vapour now radiometer-true to 0.15 mm"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire comparison treats the 210 GHz radiometer as the truth reference, but the radiometer's PWV values are themselves produced by a quadratic model fitted to 24 years of MERRA-2 weather-reanalysis data and a tipping-curve method that assumes an isothermal atmosphere, so any bias in that model is inherited by both instruments and the agreement is between two model-dependent estimates rather than a measurement of absolute accuracy.","fun_headline_variants_meta":{"raw":{"variants":["GNSS water vapour matches 210-GHz radiometer to 0.15 mm","On-site weather data tighten GNSS water vapour to 0.15 mm","GNSS water vapour: 98% match with 210-GHz radiometer","Radiometer-accurate GNSS water vapour: offsets drop to 0.15 mm","Telescope-site GNSS water vapour now radiometer-true to 0.15 mm"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001339,"raw_usage":{"total_tokens":5510,"prompt_tokens":1080,"completion_tokens":4430,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":696,"completion_tokens_details":{"reasoning_tokens":4318}},"tokens_in":696,"tokens_out":4430,"duration_ms":32463,"temperature":1.0,"reasoning_tokens":4318,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:07:25.111504+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Co-locate a radiosonde balloon programme at the H.E.S.S. site across at least one wet and one dry season and compare its PWV profiles with simultaneous GNSS and 210 GHz radiometer PWV; a systematic offset between the radiosonde and both instruments that follows the sign of the MERRA-2-based conversion would show that the GNSS-radiometer agreement is a shared-model effect, not absolute accuracy.","supporting_citations":[{"cited_title":"A comparative analysis of GNSS-inferred precipitable water vapour at the potential sites for the Africa Millimetre Telescope , \\/ , 537 (2), 1357--1368","cited_arxiv_id":null,"evidence_quote":"Supplies the GNSS PWV dataset and site characterisation for H.E.S.S. and Gamsberg that the present comparison validates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the T_m(T_s) modelling procedure and the Atacama 0.64 mm offset that serves as the comparison benchmark."},{"cited_title":"F., Skrutskie , M","cited_arxiv_id":null,"evidence_quote":"Defines the tipping-curve method the 210 GHz WVR uses to turn sky-brightness measurements into zenith opacity."},{"cited_title":"Harnessing the gps data explosion for interdisciplinary science, Eos\\/ , 99 (2), e2020943118","cited_arxiv_id":null,"evidence_quote":"Provides the NGL GNSS products (ZTD and interpolated T_m) from which the GNSS PWV is calculated."},{"cited_title":"& Nordius , H., 1987","cited_arxiv_id":null,"evidence_quote":"Supplies the physical relation and constants connecting zenith wet delay to PWV through the weighted-mean temperature."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier validation of GNSS PWV against VLBI and a water vapour radiometer that motivates using GNSS as a PWV sensor."},{"cited_title":"A., Rocken, C., Anthes, R","cited_arxiv_id":null,"evidence_quote":"Establishes the linear T_m-T_s relationship that the piecewise local model adapts to the H.E.S.S. site."}],"review_version":1}