Design of Real—Time Sampling Strategies for Submerged Oil Based on Probabilistic Model Predictions
Round 1
Reviewer 1 Report
The paper proposes sampling plans based on a probabilistic prediction which can guide to comprehensive, real time-updated sampling plans during oil responses to locate and track the submerged oil. The paper presents some repetitions in words which makes the text hard to read for example: In the abstract the word submerged oil is repeated enthusiastically as well as the word sampling in the introduction. It could be omitted in some parts.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Review of Ji et al.,” Design of real-time sampling strategies…”
The reviewer wishes to make two points before providing the actual review. First, the editor requires a very short time period for the review of a 37-page article and hence some author arguments may have been missed by the reviewer. My apologies. Second, the reviewer recognizes the authors as well qualified experts in the field and read with interest their previous publication (Ref 8) on the subject.
Unfortunately, the reviewer cannot recommend this manuscript for publication in ‘Marine Science and Engineering”. The article offers little innovative material while ignoring some existing related work. Moreover, they chose to test their formulas not against actual data but rather a model that itself incorporates additional approximations about the submerged oil that make use of it as a surrogate a questionable proposition.
Most children in the United States have at sometime played a game called ‘battleship’. According to the game, straight lines of varying length are drawn on a sheet of gridded paper. The lines may be vertical, horizontal or diagonal and represent the fleet of ships of the player. The opponent cannot see the lines and tries to guess where they are in order to ‘sink’ the ships. It appears to the reviewer that the manuscript closely resembles a modification of this game applied to submerged oil. The opponent begins random guessing or some systematic sweep (their first phase using random or zig-zag) until he hits a line. He then chooses the squares around the hit to determine the orientation of the line (their adaptive systematic) and finally selects squares along the orientation (their SOSim?) The last phrase contains a question mark because SOSim does not include hydrodynamics but rather seems to be some version of persistence.
The problem of this approach can be best illustrated by playing the battleship game on papers of different grid size. The task of the opponent becomes significantly more difficult as the grid size shrinks. The opponent will get no credit for selecting a square without a line even if it is next to a square with a line. This would give an unrepresentative high relative error value according to their equation 1. The Weather Service has long recognized the challenges of assessing forecasts against point source measurements but the authors appear unaware of this literature. For submerged oil, this circumstance corresponds to the width and distribution of the oiled bands. Estimating these bands in any practical sense requires information on the spill source and then running a spill model such as Oscar to forecast the location and nature of the bands. This should provide a superior estimate for the manifold of subsurface oil than does krieging since the latter incorporates none of the fluid dynamics. Hence, the sampling protocols are best set by looking at the spill model forecast, both for surface and subsurface oils. This is the current practice. The relative error in the model can then be determined by such techniques as described in the book, Forecast Verification, by Jolllife and Stephenson and the model parameters and predictions for the next round of sampling adjusted accordingly.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors present very clearly the methods and results of seven different methods of the two-phase sampling system on a real case of the underwater reactions of oil. The authors focused on the oil plum, which was caught in the isopycnic layer at a sea depth of 1300 to 900 meters, where the density of the oil and the sea would balance and thus show a neutral buoyancy state.
Comments / explanations:
- all data refer to the oil plum; e.g., that oil plum which is located in the immediate vicinity of the site of the first release at a depth of 1500 meters. Does this mean that in the water column from 1500 to 1300 meters, the movement of the oil plum is completely vertical?
- individual sampling procedures are time-consuming and the total sampling distance is more than 200 km. What is the time frame for sampling (it would probably be good to add this to the article), and how does the shape of the plum relate to the time-lapse given for each sampling, since we know that the plum is constantly spreading and transporting?
- what quantity of oil is discussed in detail; this information is missing for the reader (5 and 6 May)
- the kriging method may be well known to a certain circle of experts, but I think it is still necessary to add a note here (line 305)
- the same applies to the use of the Hausdorff distance (line 381); you may be able to provide an example using a figure with calculated Hausdorff distances (within a case)…this distance method is part of the OSCAR application, please clarify this
- can OSCAR very accurately model oil concentrations on the order of 1ppb?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
see comments in attachment
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The reviewer appreciates the effort the authors made to revise the manuscript and the time spent on answering the first review. Unfortunately, the author’s comments only confuse the reviewer more as to the purpose of this effort. Consider the following:
1) Kriging is a type of regression that generates an estimation manifold based on a set of random data points. A typical example might be estimating groundwater contamination where the underlying rock strata (e.g. porosity, permeability etc) is known but nothing is known about the contamination distribution itself. It also usually assumes that the space is stationary, both in the statistical sense (joint probability distribution does not change) and in colloquial usage. Measurements over different days can all be compiled into a larger data set.
Both of these conditions typically do not apply to tracking neutrally buoyant oil such as the subsurface plumes from DWH. Given the presumption of the paper that the isopycnal layer for neutrally buoyant oil is mapped (and presumably measured or modeled layer thickness and subsurface currents are known), then a standard hydrodynamical model fitted to the data is going to give a better answer than kriging, particularly a krig result that uses data that have significant time variation as would be required if the search path covers a couple of hundred kilometers, as proposed in their figure 9b. Do the authors contend that the oil concentrations or water flow patterns will not change?
2) The authors assume that the oil can be treated as a passive, conserved, low dispersion contaminant that follows the water and has no sharp concentration gradients. This was certainly not true for DWH. Consider the OSCAR output for DWH on May 9 (Figure 2c). Banding of the oil concentration is obvious. As a full-scale spill model, OSCAR can predict such phenomena but the reviewer is not convinced that SOSIM would do the same.
3) The reviewer still does not see the advantage of SOSIM over a complete spill model that is re-calibrated as new data becomes available, something SINTEF did with their model during DWH. The authors note that their model uses a statistical approach but so can OSCAR by varying its diffusion coefficients or flow estimates. All the major spill models (OILMAP, GNOME among others) can provide statistical answers and presumably would give a more accurate answer than a reduced model like SOSIM (see point 2) that seems at best to interpolate flow velocity readings. This is surprising to the reviewer. At least one of the authors is a world recognized expert in oil transport modeling and quite familiar with building near-field flow models, based on local hydrodynamics, that can be integrated into a spill model. Why did not the authors go this route?
4) Unfortunately, the authors did not utilize the suggested reference from the previous review for constructing their success rate formula (Section 2.3.3). They define (lines 408-409) their success rate as the number (ns) of success predictions [SOSIM predicts oil] divided by the number of data points (I) detecting submerged oil [as defined by OSCAR]. As written in equation 5, this will not work. There is no penalty for SOSIM wrong guesses so one can get a 100% by simply having SOSIM make N predictions of oil everywhere with ns being a subset of N. While there are newer and better metrics, the authors should at minimum adopt the Pierce Skill Score (1884) that subtracts the fraction of misses from the fraction of hits. I think this may be what the authors somewhat intended, given their discussion in line 410, but it is not what equation 5 says. The Pierce Skill Score (PSS) ranges from 1 (model gets everything right) to -1 (model gets everything wrong).
Based on the above, the reviewer cannot in good faith recommend this manuscript for publication. However, given the reputation of the authors and the obvious effort that went into this work, the reviewer does recommend that the editor send the paper out to another reviewer for a second opinion (you may include this review if you wish), a good practice both for the medical field and scientific research. Scientific wisdom requires that one recognize that others may be more wise than you on certain matters.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
