{"id":"de8501d8-7199-41b0-a221-4b8646b93622","arxiv_id":"2411.17731","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":13,"one_line_summary":"The paper presents an IoT soil monitoring system and fits exponential and neural-network models linking soil resistivity to salinity for a watermelon field in Bangladesh.","lead":"This paper describes an Internet of Things system that measures soil moisture, pH, and temperature in a watermelon field, then compares those readings with portable meters and a soil science laboratory. It also fits exponential equations and a neural network to estimate soil salinity from electrical resistivity measurements.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The resistivity measurements feeding every fitted equation and the ANN are not valid soil resistivity values: the two-probe bowl method has no calibrated geometric factor, and the reported 23.52 kΩ·m for 20% salinity is orders of magnitude higher than physically plausible.","rationale":"The paper's central claim is that soil salinity can be predicted from resistivity after fitting exponential equations and an ANN. All of those fits take the resistivity values produced by the analog-multimeter/bowl method as ground truth. If those values are not true soil resistivity, the model is fitting noise or an artifact. The reported numeric example gives a resistivity of 23.52 kΩ·m for a heavily salinized soil, which is physically implausible and indicates electrode contact resistance. This is a correctness risk, not merely a disagreement with the literature: the geometric factor in Eq. (1) is not applicable to the measurement geometry, and the use of DC through metal probes in moist soil produces polarization effects. The missing raw data table makes it impossible to audit the fitted equations, but even the single shown example is enough to raise the concern. The reader's weakest assumption identified the same measurement-validity issue, so I agree. The recommended verdict remains Reject; no adjustment is needed.","tokens_in":16451,"tokens_out":4963,"duration_ms":46643,"concrete_test":"Measure a standard solution of known resistivity, such as 0.01 M KCl (~7 Ω·m at 25 °C), with the same SANWA YX-360-TRF multimeter, probes, and bowl geometry. Compute ρ from Eq. (1) and compare to the known value; if the computed value is more than 2× off, the setup does not measure true resistivity and the salinity model based on it is not valid. A complementary check is to compare the two-probe bowl readings on a field soil sample with a calibrated four-electrode resistivity meter; discrepancies over 100% would confirm contact resistance dominates.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III.C.1 derives soil resistivity from a single resistance reading via ρ = RA/L, setting L to the probe spacing and A to the bowl's cross-sectional area (πr², r = 0.05 m). This is only valid for a uniform sample with parallel current flow; a two-electrode cell in a bowl produces a non-uniform current distribution and an unknown geometric factor, and the analog multimeter's DC signal adds electrode polarization and contact resistance. The one worked example, 150 kΩ at 20% salinity and 5% moisture, yields 23.52 kΩ·m. A soil with 20% salt is extremely conductive; realistic bulk resistivity for such a sample would be tens of Ω·m or lower. The three-orders-of-magnitude gap is consistent with contact resistance rather than bulk soil resistivity. Since Eqs. (2)-(7) and the ANN are fitted to these resistivity values, the salinity predictions are calibrated to an instrument-dependent quantity, not a soil property. Independent validation of the measurement (for example against a standard solution or a four-electrode setup) is absent.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports an IoT-based soil characterization system for watermelon fields, measuring moisture, temperature, and pH with sensors, uploading data to ThingSpeak, and displaying it via a web page and mobile app. The authors compare these measurements with field meters and laboratory analyses and claim close agreement. The central scientific claim is a new model for estimating soil salinity from soil resistivity: for six moisture levels (5% to 50%), the paper fits exponential curves to resistivity-versus-salinity data and then algebraically inverts these curves to obtain salinity from resistivity; separately, an artificial neural network (ANN) is trained on the same data to predict salinity from moisture, pH, temperature, and resistivity. The conclusion is that soil salinity can be determined from resistivity measurements, which would permit low-cost, real-time salinity assessment.","tokens_in":16858,"tokens_out":1978,"duration_ms":21169,"significance":"If the salinity-resistivity relationship were valid and transferable, the proposed method could be practically valuable for watermelon farmers in coastal Bangladesh, where salinity is a known production constraint. The IoT data-collection and visualization pipeline, and the comparison of sensor, field-meter, and laboratory values, are reasonable engineering contributions. However, the significance of the paper's central claim is nullified by three load-bearing weaknesses: the resistivity measurements are not physically meaningful soil resistivities, the 'prediction' equations are simply inversions of curves fitted to the same data (with an internal coefficient mismatch), and the ANN is trained and evaluated on the same samples used for the curve fits, so the reported high R² values are expected and not evidence of predictive skill. Consequently, the manuscript does not establish a reliable method for salinity estimation.","major_comments":[{"comment":"The soil resistivity values used throughout the paper are not valid soil resistivities. The method measures resistance with a two-probe analog multimeter in a small bowl and computes resistivity via ρ = RA/L (Eq. 1), using the probe spacing as L and the bowl's cross-sectional area as A. This formula is only valid for a uniform sample with parallel current flow; a two-electrode cell in a bowl produces a non-uniform current distribution and an unknown geometric factor, and the DC multimeter adds electrode polarization and contact resistance. The single worked example (150 kΩ for 20% salinity at 5% moisture) gives 23.52 kΩ·m, which is about three orders of magnitude higher than physically plausible for a soil with 20% salt; the discrepancy strongly suggests contact resistance rather than bulk soil resistivity. Since Eqs. (2)-(7) and the ANN are fitted to these values, the entire salinity model is calibrated to an instrument-dependent measurement, not to a soil property. No independent validation against a standard solution, a four-electrode setup, or any reference resistivity method is provided.","section":"§III.C.1"},{"comment":"Equations (8)-(13) are not independent predictive models; they are algebraic inversions of the exponential curves (2)-(7) fitted to the same dataset. For example, Eq. (8) is obtained by solving Y = 26.213e^(-0.007X) for X, and the other equations are analogous. Thus the 'prediction' of salinity from resistivity simply reproduces the fitted curve and cannot serve as validation. Additionally, Eq. (9) uses coefficients A = 1.6843 and B = 0.006, but its source Eq. (3) has A = 1.7002 and B = 0.007; even the inversion is internally inconsistent. The reader cannot reproduce the inversion from the reported fit.","section":"§IV.C"},{"comment":"The fits at 30%, 40%, and 50% moisture have R² values of 0.607, 0.442, and 0.477, respectively. These values indicate that the exponential model explains less than half of the variance for the 40% and 50% cases, so the claimed relationship between resistivity and salinity is effectively absent at higher moisture contents. The text does not discuss this limitation, instead asserting that the equations provide 'a quantitative method for assessing soil salinity based on resistivity'; this is not supported by the reported fit quality for these moisture levels.","section":"§IV.C, Eqs. (5)-(7)"},{"comment":"The ANN is trained, validated, and tested on the same 100-sample dataset that was used to derive the exponential resistivity-salinity curves. Because the inputs include resistivity, which is itself a direct function of salinity through the fitted curves, the high training and test R² values (e.g., 0.99519 at 5% moisture) are expected and do not demonstrate that the model learns a generalizable salinity relationship. No independent test set from separate samples or field measurements is used, so the claim that the ANN can 'predict' soil salinity is unsubstantiated. The paper also omits details on the dataset itself: only one worked example is given, and Table II, which should contain the full resistivity measurements, is not included in the text, so the entire analysis is not reproducible.","section":"§III.C.2 and §IV.C (ANN)"}],"minor_comments":[{"comment":"The text contains an apparent typo: 'data can be con Fig. d as public or private' should read 'configured as public or private'.","section":"§III.B"},{"comment":"The reference list has numbering issues: [18] appears twice (one is the STM32 paper and another is the Singh et al. paper), and the in-text citations are not consistently matched to the reference list.","section":"References"},{"comment":"The axes labels and legends in the resistivity-versus-salinity plots are not readable in the provided figures, and the plotted data points are not shown; only fitted curves appear, preventing the reader from assessing scatter or outliers.","section":"Figures 10-15"},{"comment":"The pH, moisture, and temperature comparison tables list values but do not include uncertainty estimates or statistical measures (e.g., mean absolute error, standard deviation), so the claim that the three methodologies are 'practically the same' is only qualitatively supported.","section":"Tables III-V"}],"recommendation":"reject","confidential_remarks":"The paper is a mix of an IoT deployment and a calibration study. The IoT and mobile-app components are straightforward and not novel, while the salinity-resistivity model contains fundamental methodological flaws: the resistivity measurement is not physically calibrated, the prediction equations are algebraic inverses of fits, and one inversion is numerically inconsistent with its source curve. These are not fixable by local edits; the resistivity measurement would need to be redone with a proper four-electrode or calibrated cell, and the model would need validation on independent data. Given the scope of the journal and the central claim, rejection is appropriate. There is also a reproducibility issue: the complete dataset is not provided."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is two things glued together. The first is a standard IoT soil-characterization system—Arduino, Raspberry Pi, sensors, ThingSpeak, a mobile app—that seems to work and is compared against field meters and lab readings with reasonable agreement. That part is fine as an engineering report, though not novel. The second is a salinity-prediction model built on resistivity, and that part is not in publishable shape.\n\nThe core problem is circular: equations (2)-(7) are exponential fits to the resistivity-salinity data, and equations (8)-(13) are derived by algebraically inverting those same fitted curves. The ANN is trained and tested on the same 100 samples, so its high R² values just reflect curve fitting on the training set. There is no independent validation on fresh soil samples, no raw data table, and no error analysis. The numeric mismatch in Eq. 9—coefficients 1.6843 and 0.006 instead of 1.7002 and 0.007 from Eq. 3—is a concrete internal error that a referee would catch immediately.\n\nThe stress-test note about the resistivity measurement is worth taking seriously, though I think its strongest point is not the absolute magnitude. Yes, 23.52 kΩ·m for a high-salt sample at only 5% moisture is not automatically absurd—dry soil can be quite resistive. The real issue is that ρ = RA/L assumes uniform parallel current flow through a regular conductor. Two steel probes in a bowl of soil produce a non-uniform current distribution, and the analog multimeter's DC signal introduces contact resistance and polarization. The paper never calibrates this against a four-electrode setup or a known solution. So the fitted equations are calibrated to an instrument-dependent quantity, not to a reliable soil property. That makes any field prediction suspect.\n\nWhat the paper does do well: the field visit and farmer interviews are genuine, the three-way comparison of pH, moisture, and temperature is a reasonable validation of the IoT system, and the writing is clear enough. The authors are not hiding their R² values; they report 0.607, 0.442, and 0.477 for the 30%, 40%, and 50% moisture fits, which is honest but also tells you the exponential model breaks down at higher moisture.\n\nMy recommendation: do not send this to peer review in its current form. The salinity model is load-bearing and it rests on an unverified measurement and internally inconsistent fits. If the authors supply the raw data, fix the coefficient error, and validate the resistivity method against a standard technique, the IoT system part could become a legitimate application paper. As it stands, the scientific claim is not supported.","headline":"A routine IoT soil-monitoring build with an unvalidated, internally inconsistent salinity-resistivity model that should be desk-rejected until the authors supply raw data, fix the coefficient mismatch, and test the setup against a standard resistivity method.","tokens_in":17385,"tokens_out":2484,"would_cite":false,"duration_ms":27390,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Soil salinity can be predicted from a simple resistivity measurement","keywords":["Internet of Things","soil salinity","soil resistivity","artificial neural network","watermelon cultivation","soil moisture","soil pH","precision agriculture"],"falsifier":"Take soil samples with known salinity, measure resistivity with the same two-probe bowl setup across the six moisture levels, and compare the salinity predicted by Eqs. (2)-(7) with the known values; if the predictions drift systematically with sample batch, probe placement, or bowl size, or fail outside the 5-50% moisture range, the central claim is falsified. A simpler version is to run one blind sample through the multimeter setup and the ANN and check whether the predicted salinity matches an independent lab conductivity measurement.","tokens_in":16318,"feed_emoji":"🍉","tokens_out":4172,"duration_ms":41613,"temperature":0.7,"pith_summary":"This paper tries to establish that soil salinity can be estimated from soil resistivity, so that a farmer could check salinity with a cheap electrical measurement instead of sending samples to a laboratory. The authors build an IoT system that measures soil pH, moisture, and temperature and uploads readings to the cloud, and they compare those readings with field meters and lab tests to argue the system is accurate. The central new claim is a set of fitted exponential equations, one per moisture level, that turn a resistivity value into a salinity percentage, plus an artificial neural network that does the same mapping from pH, moisture, temperature, and resistivity. If the claim holds, salinity assessment becomes fast and low-cost, which matters for watermelon fields where salt intrusion can ruin yields.","feed_headline":"Soil salinity predicted from a simple resistivity reading","feed_subtitle":"Exponential equations and a neural net turn a two-probe resistance measurement into a salinity estimate for watermelon fields.","key_machinery":"The load-bearing object is the resistivity-to-salinity relationship. Resistivity is defined by $\\rho = RA/L$ from a two-probe multimeter reading $R$ across a known probe spacing $L$ and bowl cross-section $A$. For each moisture level the paper fits an exponential curve $Y = A e^{-BX}$ to the measured points, inverts it logarithmically, and also trains an ANN with one hidden layer, tan-sigmoid activations, and a linear output using Levenberg-Marquardt backpropagation to predict salinity from pH, moisture, temperature, and resistivity. The exponential equations carry the simple field-use claim; the ANN is the paper's evidence that the multi-variable relationship is stable enough to learn.","core_discovery":"On the paper's own terms, the discovery is that soil electrical resistivity and soil salinity are linked by a simple exponential decay for each fixed moisture level: $Y = A e^{-BX}$, where $Y$ is resistivity in kΩ·m, $X$ is salinity in percent, and $A$ and $B$ are fitted constants. Inverting these equations gives $X = \\frac{1}{B}\\ln(A/Y)$, so a single resistance reading from two probes in a soil sample yields an estimated salinity. The paper also trains a feed-forward ANN with inputs pH, moisture, temperature, and resistivity to predict salinity, and reports high regression $R^2$ values and low mean squared errors for moisture levels up to 30%. The IoT system's pH, moisture, and temperature readings are shown to match laboratory and field-meter values closely, supporting the claim that the whole chain from sensor to cloud to mobile app is serviceable.","pith_inferences":["The exponential form hints at a physically plausible picture: dissolved salts add charge carriers, so resistivity falls rapidly as salinity rises, but the rate of fall depends on how much water is available to dissolve the salt, which is why the paper needs a separate equation per moisture level.","A direct extension would be to compare the two-probe resistivity estimates against a standard electrical-conductivity meter on the same samples; agreement would let the method be calibrated without a laboratory.","The ANN results reported for 40% and 50% moisture are weaker than for lower moisture levels, so a field deployment would likely need more data in those ranges before relying on the model there.","Because soil resistivity is temperature-dependent, the fitted equations may need a temperature correction when used at times of day far from the lab conditions under which the data were collected."],"forward_implications":["A farmer with a multimeter and a fixed-volume bowl could estimate soil salinity on site rather than waiting for lab results.","The salinity estimate depends on knowing the moisture level, because the fitted equations differ for 5%, 10%, 20%, 30%, 40%, and 50% moisture.","The IoT system gives real-time pH, moisture, and temperature data through a mobile app, so the salinity estimate could be combined with live moisture readings.","If the ANN generalizes beyond its training samples, salinity prediction can use several soil properties jointly instead of relying on resistivity alone.","The fitted equations are calibrated to the tested watermelon-field soil, so applying them to other soils would require refitting on local samples."],"supporting_citations":[{"why":"Provides the prior observation that soil resistivity varies with salinity, the premise that the fitted equations extend.","marker":"[31]"},{"why":"Compares regression and ANN models for relating soil properties to electrical resistivity, justifying the ANN approach.","marker":"[21]"},{"why":"Reports that low resistivity tracks higher salinity and moisture, the qualitative link the paper inverts.","marker":"[42]"},{"why":"Shows that chemical soil properties can be inferred from electrical measurements, supporting resistivity as a salinity proxy.","marker":"[45]"},{"why":"Supplies the Levenberg-Marquardt backpropagation algorithm used to train the ANN.","marker":"[46]"}],"fun_headline_variants":["Soil salinity from a single resistivity reading","One resistivity reading gives watermelon soil salinity","IoT and AI monitor watermelon soil salinity","Exponential law converts resistivity to soil salinity","Neural net turns soil resistivity into salinity readings"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire salinity model rests on the assumption that the resistance values from an analog multimeter and two probes in a small bowl are accurate and reproducible measures of the soil's true resistivity; the paper gives only one worked example and no raw data table, so if that measurement is noisy or setup-dependent, every fitted equation and the ANN inherit the error.","fun_headline_variants_meta":{"raw":{"variants":["Soil salinity from a single resistivity reading","One resistivity reading gives watermelon soil salinity","IoT and AI monitor watermelon soil salinity","Exponential law converts resistivity to soil salinity","Neural net turns soil resistivity into salinity readings"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000986,"raw_usage":{"total_tokens":4168,"prompt_tokens":916,"completion_tokens":3252,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":532,"completion_tokens_details":{"reasoning_tokens":3202}},"tokens_in":532,"tokens_out":3252,"duration_ms":22608,"temperature":1.0,"reasoning_tokens":3202,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:49:32.707781+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take soil samples with known salinity, measure resistivity with the same two-probe bowl setup across the six moisture levels, and compare the salinity predicted by Eqs. (2)-(7) with the known values; if the predictions drift systematically with sample batch, probe placement, or bowl size, or fail outside the 5-50% moisture range, the central claim is falsified. A simpler version is to run one blind sample through the multimeter setup and the ANN and check whether the predicted salinity matches an independent lab conductivity measurement.","supporting_citations":[{"cited_title":"Comparison between regression and ANN models for relationship of soil properties and electrical resistivity,","cited_arxiv_id":null,"evidence_quote":"Compares regression and ANN models for relating soil properties to electrical resistivity, justifying the ANN approach."},{"cited_title":"A quarter century experience in soil salinity mapping and its contribution to sustainable soil management and food security in Morocco,","cited_arxiv_id":null,"evidence_quote":"Reports that low resistivity tracks higher salinity and moisture, the qualitative link the paper inverts."},{"cited_title":"Study of chemical effects on soil compaction characterizations through electrical conductivity,","cited_arxiv_id":null,"evidence_quote":"Shows that chemical soil properties can be inferred from electrical measurements, supporting resistivity as a salinity proxy."},{"cited_title":"Effects of salinity on the soil microbial community and soil fertility,","cited_arxiv_id":null,"evidence_quote":"Supplies the Levenberg-Marquardt backpropagation algorithm used to train the ANN."}],"review_version":1}