the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 01 Dec 2025
Plastic film residues on cropland: monitoring soil contamination through optical remote sensing
Abstract. Plastic films have been improving agricultural production and covering an increasing surface area of cropland in the last decades. Yet their use has been connected to the generation of plastic residues, potentially acting as a main secondary microplastic source in agricultural soils. Monitoring the generation of plastic film residues is crucial for identifying good management practices and assessing the risk of plastic use in agriculture. Remote sensing has been qualified as a valuable tool for monitoring macroplastic mainly on waters, while its use on agricultural soils is mostly unexplored. Our study combined proximal and remote sensing techniques to lay the foundations of UAV (Unmanned Aerial Vehicle) use for monitoring macroplastic film residues on cropland.
Through proximal and UAV acquisitions of five-bands multispectral data (i.e., blue, green, red, red edge, near infrared), we highlighted the potential of off-the-shelf miniaturised sensors and identified possible workflows for detecting macroplastic film residues. Our findings highlight a greater efficacy of spatial resolution over spectral resolution, encouraging the use of high-resolution RGB cameras over multispectral cameras. Through proximal acquisitions of hyperspectral data, we built spectral libraries and located absorption peaks for the most commonly used plastic films. We highlighted that these absorption peaks unambiguously identify plastic films on cropland and offer the potential to distinguish plastic types, encouraging the development of sensors tailored for plastic detection.
Competing interests: At least one of the (co-)authors is a member of the editorial board of SOIL. The peer-review process was guided by an independent editor, and the authors also have no other competing interests to declare.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(1775 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1775 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-3804', Anonymous Referee #1, 29 Dec 2025
-
AC1: 'Reply on RC1', Alessandro Fabrizi, 14 Jan 2026
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-3804/egusphere-2025-3804-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Alessandro Fabrizi, 14 Jan 2026
-
RC2: 'Comment on egusphere-2025-3804', Anonymous Referee #2, 30 Dec 2025
General comments:
The authors deal with a very important issue of how plastics in agricultural fields can be identified with remote sensing techniques. Identifying plastic with UAVs would be very valuable, especially if it can be achieved with RGB or multispectral sensors. The study is important and addresses an arising concern of plastic pollution in agricultural fields. The authors first set up a controlled experiment to examine if plastics can be identified with proximal hyperspectral and multispectral sensors. They use pristine, crumpled, dirty, and crumpled and dirty samples. This is interesting as plastics left in the field are likely to become dirty, crumpled or torn, and thus the importance of measuring the reflectance of such samples. Next, they examine if plastics can also be identified in field conditions using multispectral UAV images.
However, although the experimental setup can achieve the required goal, there are two major concerns with this study. First, the field dataset is relatively small, raising questions as to how robust and reliable the results are. Secondly, a lot of important information is not provided in the methods and results sections. The missing information in the methods make it challenging to understand exactly what was done and get a clear overview of the results. These comments therefore focus mainly on the methods and results sections, as in my opinion, they need to be substantially improved before this paper can be considered for publication.
More specifically, the study uses a small number of plastic film samples in the field study, especially considering that they were used to train random forest models. Twenty-one black films, 19 transparent films, and 13 white films were placed on the field, and for validation purposes, only 4.2, 3.8 and 2.6 number of points were used for validation for the black, transparent and white plastic films (according to Figure 7), respectively. The concern is that the training and validation of the classification models are not reliable enough.
Additionally, valuable information is lacking in the study. Specific examples follow below, but for example, in the methods section, information is lacking on the soil used in this study, as little or no chemical or physical data is provided (e.g. organic matter, pH, etc.). Soil properties can greatly affect spectral results as thus are important in this study.
The description of the plastics used in this study it is not detailed enough. Which type of plastics was used, beyond the plastic color? This information is vital, especially as one of the aims of the study was to 'build spectral libraries for the most common plastic films used in agriculture' (line 180).
Additionally, some of the methods are not described sufficiently- for example the process of calculating the producers’ accuracy in the field study. How many points were used for validation? How were validation points created? This is especially important with the small dataset that was used in this study.In the results section, important results are not displayed, such as the spectra of the crumpled, dirty, and crumpled + dirty plastics films. Images or results from the multispectral images acquired in the controlled outdoor experiment and of the field in the field scale study are not shown as would be expected.
Please find below specific details in the methods and results sections.
Specific comments:
the comments provided below are examples information that is missing, or unclear, in the methods and the results sections. The suggestion is to review the paper again and provide missing information.
Methods:
Section 2.1.2 (lines 109-132):
Line 113-114: how was it ensured that all the plastic films were equally ‘dirty’? was the entire plastic film covered by soil? From Figure 1, it seems like there is a relatively thick layer of soil that mostly obscures some of the plastic films, but not others. It is hard to understand if that was actually so from the description. which soil was used to make the films dirty? the choice of soil could greatly affect the spectra (clayey vs sandy soils for example).
Lines 115-118: the description of the soils used in this study should be more detailed with at least some chemical and physical data (organic matter etc.). This is important for spectral studies, as soil parameters strongly affect reflectance.
Line 117: I’m curious to know why such a fine sieve was used in this study. Isn’t a 2 mm sieve more common in similar soil studies?
Lines 110-111: Which plastics were used? What was their chemical composition (e.g., LDPE)? It is known that different plastics will have different reflectance, and therefore providing this basic information is critical. Additionally, what do all the treatment names symbolize- BIO, B_1, BW_2 etc.? I suggest giving the samples more informative names. it is unclear what it means that black and white films were double sided and used on both sides.
Table 1: this table is confusion- From ‘application mode’ column and onwards- is this information relevant to the plastics used in the experiment? For example, were B_1, B_2 and BIO used for one growing season in fruit or vegetable fields prior to the experiment, or is this their common use? Sample names are not consistent, with some ending with numbers (B_1 and B_2) and some ending with letters. What do the numbers and letters signify?
More comments in the methods section:
Lines 138-140: What caused the ‘rapidly changing light conditions’ in samples BW_S and BW_L and not in other samples?
Line 147: why was it decided to present the spectra of the pristine films only? In field conditions, it is expected to encounter crumpled or dirty plastic films, therefore, viewing their spectra is very interesting.
Lines 174-175: Soil in the field study is also not sufficiently described. Were chemical or physical analyses conducted? Once again, this can greatly affect reflectance results and the ability to identify the plastics.
Lines 185-190: How was elevation calculated? From the UAV images? If so, what was the flight overlap and what method was used to calculate elevation?
Line 196: which GIS software was used? How were the random points generated for spreading the plastic films?
Lines 198-203: The number of samples in the field study are very few. Especially when training a model and then creating validation points, this number of samples seem highly insufficient.
Lines: 204-209: were Ground Control Points (GCPs) placed in the field for generating the orthomosaics?
Lines 214-215: This seems problematic as adjacent pixels are spatially highly autocorrelated. Was this done for the plastic film training points as well?
Lines 234-240: How many points were actually used in the validation process? How were they created? It is important to understand exactly what was done in this section because of the small amount of samples. this section should be as detailed as possible, perhaps creating a workflow chart to describe all the steps that were done in the field study, including the validation process.
Results:
Lines 245-246: the closeness of the transparent films to the soil spectra seems to depend on the spectral region, and there are clear changes across the spectral region. For example, in the left panel of Figure 3, between 500 till around 750 nm, T_H is closest to the soil reflectance, but then at around 1450-1600 nm, T_S is closest, and also at around 2,000 nm. Also, in the right panel there are changes across the spectrum, with ‘T_L’, which is the thickest film, sometimes being ‘closest’ to the soil spectrum. The relationship between the spectral region and the reflectance is interesting and important, as spectral indices can be derived from them.
Lines 256-258: The claim is that HI_1215 and HI_1732 allow to distinguish between white and transparent plastic films. However, from Figure 4, this does no seem to be so clear-cut. In panel (a) transparent film T_S and white film WB_L have index values that are quite close, with large variation that assumingly does not lead to any significant differences. The same can probably be said for T_H and WB_S, and in panel (b), for T_H and WB_S. Please provide significant letters in the graph.
Please also indicate if the graph is showing pristine or treated plastic films, as these is not clear, and the claim made in Lines 256-256 cannot therefore be checked. Results of all the treatments should be shown to be able to observe the treatment effect,Line 273-275: Figures showing the reflectance of all the treatments (pristine, crumpled, crumpled + dirty, dirty) film are missing. It would be easier to visually see such results, and also see how the treatments affected the reflectance. It would be expected that the dirty films would have a reflectance more similar to the soil reflectance if they were completely or mostly covered by soil.
Lines 276-282: Showing the reflectance of all the treatments, as suggested above, would allow a better comparison of the differences between the treatments. It would be expected for the black films to have a higher reflectance across the entire spectrum. It would be very important to see if the clear absorption peaks also exist in the dirty films. The peaks would indicate how spectrally visible the plastic is even when covered with soil.
Figure 6: from the graph it seems that only white films can be detected with UAV multispectral sensors. Black films and shadow overlap, while transparent and soil overlap. The question is how ‘white’ does the white plastic have to be- if it got covered with soil particles, or mixed in the soil and torn, which is what will likely happen in field conditions, would it still have a distinguishable reflectance? Providing reflectance of the dirty, and crumpled + dirty treatments, as previously suggested, could provide an answer to this question.
Figure 7: for the black, transparent and white plastic films, there were only 2.6-4.2 number of points used for validation. This is problematic, as previously discussed. Figure 7b- what does the factor in the Y-axis refer to?
Figure 8: Providing a classified map of the entire field would be beneficial and is expected in such a study that used a UAV to map plastic contamination in the field scale.
Technical corrections:
Figure 6: It is difficult to see the values of the different treatments on the graph, as some treatment values on top of the other. For example, the soil reflectance values cannot be seen at all in the figure. Please change the figure. A suggestion is to place the dots next to each other, instead of on top of each other.
Figure 8: please add the squares that indicate where the plastics are also on panels (d)-(f). Indicate which plastic is in which panel (white, transparent, black) so classification results can be compared.
Citation: https://doi.org/10.5194/egusphere-2025-3804-RC2 -
AC2: 'Reply on RC2', Alessandro Fabrizi, 14 Jan 2026
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-3804/egusphere-2025-3804-AC2-supplement.pdf
-
AC2: 'Reply on RC2', Alessandro Fabrizi, 14 Jan 2026
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-3804', Anonymous Referee #1, 29 Dec 2025
Main points
Fabrizi et al. present an experiment investigating the potential monitoring of plastic film residues from agricultural activities generated at various stages (e.g., after use, management). The team uses advanced sensors to obtain optical properties of the subset sample of the plastic films in a controlled experiment outdoors and in the field. Datasets are made open-access, key part of supporting future open science. Several aspects about the study must be addressed before it is considered for publication especially some additional literature review as some statements can be rephrased based on facts, citations added to support text, polymer type, terminology used and presentation of the findings to put them into context of current research advances.
Critical Review Points
- Does the paper address relevant scientific questions within the scope of SOIL?
Yes it is within the scope of the journal.
- Does the paper present novel concepts, ideas, tools, or data?
Yes, although certain aspects have to revised to better illustrate the novelty or the text needs to be rephrased see specific comments below.
- Does the paper address soils within a multidisciplinary context?
Yes, it combines use of remote sensing tools to potentially monitor the distribution of plastic films used in agriculture that can contribute to environmental risks as explained in Line 23-42
- Is the paper of broad international interest?
Yes, it is of broad interest considering the issue of plastic pollution and also use of plastics to improve farming.
- Fabrizi et al. should also highlight the benefits of using plastics in the Introduction section as more text is about the unwanted effects. It is equally important to acknowledge how plastic based farming has helped improve output in the agriculture sector.
- Are clear objectives and/or hypotheses put forward?
The objectives are well defined in line 79-82. It is important that the presentation of the findings including discussion is improved
- Objective 1-the supplied dataset should be provide in a more detailed structure with the more metadata for ease of reuse.
- Objective 2-the spectral properties should be well explained and relevant plots to support the text should be presented
- Objective 3-the use of UAVs should be better explained with regards to spatial resolution depends on the camera system used and height of flight, the goals of using UAV for knowing about plastic also rely on several aspects hence the need for spectral information for example can the team explain or confirm how easy it would be to differentiate a green cloth from a green plastic and green leaf if partly covered by soil would RGB be sufficient? Assuming the piece is of same size and imaged at same resolution and same surface texture
- Spectra can be used to evaluate the above question.
- Are the scientific methods valid and clear outlined to be reproduced?
The methods are uptodate but the text should be improved to give sufficient details allowing duplication, justification or understanding of the steps applied. More information on the methods is crucial considering Fabrizi et al. suggest this is one of the first few studies on the topic covering the elements discussed.
- Do the authors give proper credit to related and relevant work and clearly indicate their own original contribution?
In general the references are ok, additional literature review is need to better position the current study with prior studies
Suggest literature review
- For examples of a relevant papers see some commercial materials and ASD used in farming associated experiments
https://doi.org/10.1016/j.solener.2020.11.058
- The authors can compare the measurements in the provided study or at least provide more details about the materials as explained in the above paper. The methods also on the ASD should be improved in terms of details provided.
- Recommended papers for the authors to review include in the introduction and discussion to better put the current manuscript into context
Case studies - including use of hyperspectral tools, proximal measurements of LDPE materials, use of satellite and airborne sensors, algorithms development, data for comparisons
- https://doi.org/10.1016/j.rsase.2025.101802
- https://doi.org/10.1016/j.envc.2025.101398
State-of-the-art reviews
- https://doi.org/10.1002/ldr.4497
- https://doi.org/10.34133/remotesensing.0395
- The data in the above studies should also be used to better demonstrate how the Fabrizi et al. open-access library is benefitting the community
- How does it bring diversity to what is already available?
- Can this be used for developing algorithms? How?see statement line 19-21
- Does the title clearly reflect the contents of the paper and is it informative?
Given the sample size used the title should reflect this is a ‘case study’ with a subset of materials used.
- Does the abstract provide a concise and complete summary, including quantitative results?
The quantitative aspects are missing in the abstract especially how many samples were done to build the open-access library.
Line 11-the sentence should be revised as remote sensing has been used extensively in agricultural plastic-based mapping see the suggested references in point 12 including the cited works in the references above.
Line 16-it must be discussed more and presented in the results to support the statement. How is the efficacy shown? See question raised in point 5 objective 3 comment.
Line 18-the data report is based on reflectance and yet the Fabrizi et al. call this absorption peaks? Is this not commonly termed ‘absorption feature’? revise throughout the text as suggested
Line 19-this is absorption features are relevant in hyperspectral data make it clear because with the camera system is this evident with the five-bands?it is not evident in the Figure 6
- Is the overall presentation well structured?
- Is the paper written concisely and to the point?
It is fairly written with the specific recommendations for improvement being on the figures and supplementary material.
- Is the language fluent, precise, and grammatically correct?
Proof reading of the text is recommended to mitigate some of the spelling mistakes and phrases that were unclear.
If automated proofing tools were used it must be declared?
Line 21-the cited works report on microplastics right?out of curiosity are these figures accounting for the ocean that cover 70% of the Earth surface and the amounts being reported at sea?
Line 29- ‘Despite….monitoring plans’ provide references and rephrase as the main message is unclear
Line 33- are the agricultural soils a source or is it the activity?
Line 38-the references style should be revised check the formatting.
Line 43-define LDPE at first use.
Line 44-can the multiple years be enumerated
Line 48-49 reference to the term ‘earlier studies’ are missing provide citations
Line 49- missing source for the sentence
Line 53-the term broad also include satellite remote sensing?
Line 55-plastics in marine have received much interest but not necessary started the topic or efforts to use UAVs see the suggested literature in point 12
Line 60 also see the case studies provided in point 12
Line 72-is it about the RGB camera or other aspects such as flight altitude, size of the samples, environmental conditions?
-how does RGB camera increase the spatial resolution?is this also accounting for the sensor capability in terms of megapixels or lenses used? Same question would be applicable to the Micasense camera used
Line 81-objective (ii) is repeated, is the goal to define or to propose workflows?
Paragraph starting Line 91-was the ASD run in reflectance or digital number mode?
- Is the 75 number referring to scans for each individual spectral measurements and how was that consistent considering a fixed integration time of 34 ms was use?
- How and when was optimization done?
- What type of Spectralon was used for reflectance any specs? Was the calibration file of the white reflectance applied in the determination of the relative reflectance?
- How big was the white target and distance from ASD foreoptic tip at the 4 minute cycle?
- Did the measurements include a background measurement to account for light from the bottom material
- How was shadow effect avoided in the experiment?
- The step in the data was not corrected for in the ASD at around 1000nm and correction steps are reported in the cited works check for this
Line 110-what does it mean ‘in case of ….’ Rephrase
Line 115-what is a field soil? Is this about a sample or..
Section 2.1.3 -examples should be given to further demonstrate what is written it can be in the supplementary section or here. How effective were the steps done, was it really noise or something?A flowchart might also be useful
Line 147-why only the mean is presented?
Line 162-with absorption features this is typically the convex and hull locations
Equations used are they from the original publications or other values were applied in the current manuscript?
Line 182-replace calibration with ‘referencing’, provide details about the panel used name and properties
Line 184-does placing mean the same as geo-referencing?
Line 195 define GNSS and GIS at first use,
Line 195 provide the images of the samples to support and visualize the text, how many recievers were used?
Line 210-312 random points does it mean training data was balanced for each class in like 211?how many pixels were plastics exactly?
Line 229 what are the numbers in the brackets (3.1….), why the specific index values as features can the results be shown of the products?
Line 231-233-is this part of step 4 or summary of the 4 steps?
Line 244-245 about Figure 3 can be confirmed by a simplified spectral unmixing test of the soil and plastic at simulated pixel amounts say 0.1, 10, 50, 100%. The spectra of the background material is needed also to check for a baseline correction if needed.
Line 261-based on the analyses and Figure 4 plot is the ND_1715 really useful? Or what is the benefit of using it in monitoring plastics?
Figure 4 can be supported by a separability metric value. Figure can be made bigger to allow visual check to see if there are overlaps such as in the (c)
Line 273-what treatments?
Discussion
Line 323-polymer type details are missing see the experiment from Jones et al 2021 point 12
Line 326-the differences is shape can be better visible if all spectra are scaled or normalized
Line 329-336 where is this shown in the manuscript considering the assumption referencing to the white target was done every four minutes so the lighting is relative or as stated Lambertian and the plastics are also unique. The available data should be used to provide supporting proof to the statements here, was this the same situation in the crumbled data hints are in Figure 5 also?
Line 344 to 351-one suggestion to confirm the statements would be spectral unmixing and should be explored here even with a simplified assumptions of the mixtures or the ones from the experiments.
Line 353-a plot can be used to confirm or demonstrate the sentence as difference are only presented in Figure 5
Line 355- sentence unclear ‘It must be accounted for, though,…..’
Line 367-explain how this was feasible and the link to glint issue
Line 369-the data is available why not evaluate the brightness thresholding approaches or at least provide the raw data for potential use in future studies?
Line 380-it is possible to explore the statements or claims here for the very fine pixel satellites for example 30cm/pixel from Pelican or WorldView?
Line 383-is relevant for machine learning classification but also depends on the size of the objects of interest?
Line 386-391-the discussion should also account for the labelling dataset size, supervised or unsupervised approaches
Line 394-contradicts a bit with the figure 3 as the spectra from Silt loam had higher values and yet in Figure 5 differences are high for sandy soils. In any case silt loam seems a better background—a figure is missing here to also support this sentence as well as paragraph to make a clear distinction between the films bright or dark ones.
Line 400-define the spectrum ranges referred to as visible, near infrared and be clear what exactly it means no any ‘unique features’? there is a peak in the green for the WB_S and WB_L also there is an absorption feature in the near infrared of T_L
Proposed algorithms can be tested see the suggested experiments in point 12-a spectral matching can be applied using the created libraries.
A clear message about the use of UAVs should also combine the limitations that can be complemented by satellites.
Line 436-should be revised as there are sensors offering open-access to hyperspectral remote sensing information of the environment see the suggested citations using drone, aircraft and satellite tools.
- Are the figures and tables useful and all necessary?
Figure 1 should be made bigger to fit margins of the text, labelling for the setup should be improved larger text, where is the white reference, distances can be added
Line 112-a figure should be included to show the ‘different conditions’ and the soil sample can be included as a foto
Line 121- the sentence must be highlighted in the figure 1 or a figure be provided
Table 1-polymer type is missing, is the header use specific or it is better ‘Common Use’other give reference for the use defined. How was the thickness obtained or was provided by manufacturer explain.
Figure 2 can have the coordinates northing and easting added, figure made bigger, RGB image of the drone survey is missing only a side view is given in (b).
Figure 3 should be made bigger and improved better with subplots of each material or groups as in the legend black films, transparent, white films.
- The scales must be specific to the material to allow shapes to be visualized
- What are the locations of the absorption features this should be highlighted
- Why is the sandy plastics less reflective? Discuss this more
Figure 5,the plots of the actual reflectance are required here as the difference does provide minimal details although the format of the figure is good. Either have the actuall measured values presented or in the supplementary material because what is the meaning of the difference? Was this maybe due to background or rather experimental setup? This is not evident in the column 1 with clean samples and the lower values also what does it mean negative difference?
Figure 6-add joining line for visualizing the multispectral data and subplots could improve the figure
Figure 8 too small
- Are mathematical formulae, symbols, abbreviations, and units correctly defined and used according to the author guidelines?
Yes, just cross checking of the values in original studies with the stated manuscript values.
- Should any parts of the paper (text, formulae, figures, tables) be clarified, reduced, combined, or eliminated?
Methods and related material require additional material and detail including improved figures.
Confusion matrix and important feature summary for the Random Forest classifier are missing.
- Are the number and quality of references appropriate?
Overall citations are ok but the recommended citations should be reviewed or justification of excluding them provided.
Sentences in the introduction also require supporting citations for completeness
- Is the amount and quality of supplementary material appropriate and of added value?
It is challenging to have a fit all format especially with the community using various programming languages. However, Fabrizi et al. is highly commended for making the dataset open access.
There are a few issues with the provided dataset that should be improved especially the presentation and metadata structure.
The excel sheet can be combined with the pdf instead of having to check five separate files. The excel sheet was challenging to open in excel and required time to decipher the format
Check the spellings of the headers. see "Wavelenght"
- Suggestion is to have the metadata + source information provided as a single structured excel sheet so that users can easily open and format as they see fit? Examples format can be found on PANGAEA or 4TU
- Column-wise arrangement of the data with the headers well defined
- Name of sensors, date time, location, geometric properties, experimental setting well explained
- Pictures in the metadata can be used to better explain the binary options 3 + 4
- The justification of only sharing the descriptive data should be provided otherwise simply provide the raw data or provide at least the plots to better show how the raw data appeared
- Was the mean and median similar?
- ‘Reflectance measurements have been pre-processed with a Savitzky-Golay smoothing of second order and 11 nm of window size’ why was this applied can the before-after be presented to assess the usefulness of the step
- The wavelength ordering seems strange with the first value in column 2 being 1000?
- Why only share the data 405 to 2395 nm? see line 91-93 range of the ASD
- What are the specs of the ‘corrected with reflectance panels.’
- Revised à The measurements have 1 nm ‘resampled’ of bandwidth
Citation: https://doi.org/10.5194/egusphere-2025-3804-RC1 -
AC1: 'Reply on RC1', Alessandro Fabrizi, 14 Jan 2026
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-3804/egusphere-2025-3804-AC1-supplement.pdf
-
RC2: 'Comment on egusphere-2025-3804', Anonymous Referee #2, 30 Dec 2025
General comments:
The authors deal with a very important issue of how plastics in agricultural fields can be identified with remote sensing techniques. Identifying plastic with UAVs would be very valuable, especially if it can be achieved with RGB or multispectral sensors. The study is important and addresses an arising concern of plastic pollution in agricultural fields. The authors first set up a controlled experiment to examine if plastics can be identified with proximal hyperspectral and multispectral sensors. They use pristine, crumpled, dirty, and crumpled and dirty samples. This is interesting as plastics left in the field are likely to become dirty, crumpled or torn, and thus the importance of measuring the reflectance of such samples. Next, they examine if plastics can also be identified in field conditions using multispectral UAV images.
However, although the experimental setup can achieve the required goal, there are two major concerns with this study. First, the field dataset is relatively small, raising questions as to how robust and reliable the results are. Secondly, a lot of important information is not provided in the methods and results sections. The missing information in the methods make it challenging to understand exactly what was done and get a clear overview of the results. These comments therefore focus mainly on the methods and results sections, as in my opinion, they need to be substantially improved before this paper can be considered for publication.
More specifically, the study uses a small number of plastic film samples in the field study, especially considering that they were used to train random forest models. Twenty-one black films, 19 transparent films, and 13 white films were placed on the field, and for validation purposes, only 4.2, 3.8 and 2.6 number of points were used for validation for the black, transparent and white plastic films (according to Figure 7), respectively. The concern is that the training and validation of the classification models are not reliable enough.
Additionally, valuable information is lacking in the study. Specific examples follow below, but for example, in the methods section, information is lacking on the soil used in this study, as little or no chemical or physical data is provided (e.g. organic matter, pH, etc.). Soil properties can greatly affect spectral results as thus are important in this study.
The description of the plastics used in this study it is not detailed enough. Which type of plastics was used, beyond the plastic color? This information is vital, especially as one of the aims of the study was to 'build spectral libraries for the most common plastic films used in agriculture' (line 180).
Additionally, some of the methods are not described sufficiently- for example the process of calculating the producers’ accuracy in the field study. How many points were used for validation? How were validation points created? This is especially important with the small dataset that was used in this study.In the results section, important results are not displayed, such as the spectra of the crumpled, dirty, and crumpled + dirty plastics films. Images or results from the multispectral images acquired in the controlled outdoor experiment and of the field in the field scale study are not shown as would be expected.
Please find below specific details in the methods and results sections.
Specific comments:
the comments provided below are examples information that is missing, or unclear, in the methods and the results sections. The suggestion is to review the paper again and provide missing information.
Methods:
Section 2.1.2 (lines 109-132):
Line 113-114: how was it ensured that all the plastic films were equally ‘dirty’? was the entire plastic film covered by soil? From Figure 1, it seems like there is a relatively thick layer of soil that mostly obscures some of the plastic films, but not others. It is hard to understand if that was actually so from the description. which soil was used to make the films dirty? the choice of soil could greatly affect the spectra (clayey vs sandy soils for example).
Lines 115-118: the description of the soils used in this study should be more detailed with at least some chemical and physical data (organic matter etc.). This is important for spectral studies, as soil parameters strongly affect reflectance.
Line 117: I’m curious to know why such a fine sieve was used in this study. Isn’t a 2 mm sieve more common in similar soil studies?
Lines 110-111: Which plastics were used? What was their chemical composition (e.g., LDPE)? It is known that different plastics will have different reflectance, and therefore providing this basic information is critical. Additionally, what do all the treatment names symbolize- BIO, B_1, BW_2 etc.? I suggest giving the samples more informative names. it is unclear what it means that black and white films were double sided and used on both sides.
Table 1: this table is confusion- From ‘application mode’ column and onwards- is this information relevant to the plastics used in the experiment? For example, were B_1, B_2 and BIO used for one growing season in fruit or vegetable fields prior to the experiment, or is this their common use? Sample names are not consistent, with some ending with numbers (B_1 and B_2) and some ending with letters. What do the numbers and letters signify?
More comments in the methods section:
Lines 138-140: What caused the ‘rapidly changing light conditions’ in samples BW_S and BW_L and not in other samples?
Line 147: why was it decided to present the spectra of the pristine films only? In field conditions, it is expected to encounter crumpled or dirty plastic films, therefore, viewing their spectra is very interesting.
Lines 174-175: Soil in the field study is also not sufficiently described. Were chemical or physical analyses conducted? Once again, this can greatly affect reflectance results and the ability to identify the plastics.
Lines 185-190: How was elevation calculated? From the UAV images? If so, what was the flight overlap and what method was used to calculate elevation?
Line 196: which GIS software was used? How were the random points generated for spreading the plastic films?
Lines 198-203: The number of samples in the field study are very few. Especially when training a model and then creating validation points, this number of samples seem highly insufficient.
Lines: 204-209: were Ground Control Points (GCPs) placed in the field for generating the orthomosaics?
Lines 214-215: This seems problematic as adjacent pixels are spatially highly autocorrelated. Was this done for the plastic film training points as well?
Lines 234-240: How many points were actually used in the validation process? How were they created? It is important to understand exactly what was done in this section because of the small amount of samples. this section should be as detailed as possible, perhaps creating a workflow chart to describe all the steps that were done in the field study, including the validation process.
Results:
Lines 245-246: the closeness of the transparent films to the soil spectra seems to depend on the spectral region, and there are clear changes across the spectral region. For example, in the left panel of Figure 3, between 500 till around 750 nm, T_H is closest to the soil reflectance, but then at around 1450-1600 nm, T_S is closest, and also at around 2,000 nm. Also, in the right panel there are changes across the spectrum, with ‘T_L’, which is the thickest film, sometimes being ‘closest’ to the soil spectrum. The relationship between the spectral region and the reflectance is interesting and important, as spectral indices can be derived from them.
Lines 256-258: The claim is that HI_1215 and HI_1732 allow to distinguish between white and transparent plastic films. However, from Figure 4, this does no seem to be so clear-cut. In panel (a) transparent film T_S and white film WB_L have index values that are quite close, with large variation that assumingly does not lead to any significant differences. The same can probably be said for T_H and WB_S, and in panel (b), for T_H and WB_S. Please provide significant letters in the graph.
Please also indicate if the graph is showing pristine or treated plastic films, as these is not clear, and the claim made in Lines 256-256 cannot therefore be checked. Results of all the treatments should be shown to be able to observe the treatment effect,Line 273-275: Figures showing the reflectance of all the treatments (pristine, crumpled, crumpled + dirty, dirty) film are missing. It would be easier to visually see such results, and also see how the treatments affected the reflectance. It would be expected that the dirty films would have a reflectance more similar to the soil reflectance if they were completely or mostly covered by soil.
Lines 276-282: Showing the reflectance of all the treatments, as suggested above, would allow a better comparison of the differences between the treatments. It would be expected for the black films to have a higher reflectance across the entire spectrum. It would be very important to see if the clear absorption peaks also exist in the dirty films. The peaks would indicate how spectrally visible the plastic is even when covered with soil.
Figure 6: from the graph it seems that only white films can be detected with UAV multispectral sensors. Black films and shadow overlap, while transparent and soil overlap. The question is how ‘white’ does the white plastic have to be- if it got covered with soil particles, or mixed in the soil and torn, which is what will likely happen in field conditions, would it still have a distinguishable reflectance? Providing reflectance of the dirty, and crumpled + dirty treatments, as previously suggested, could provide an answer to this question.
Figure 7: for the black, transparent and white plastic films, there were only 2.6-4.2 number of points used for validation. This is problematic, as previously discussed. Figure 7b- what does the factor in the Y-axis refer to?
Figure 8: Providing a classified map of the entire field would be beneficial and is expected in such a study that used a UAV to map plastic contamination in the field scale.
Technical corrections:
Figure 6: It is difficult to see the values of the different treatments on the graph, as some treatment values on top of the other. For example, the soil reflectance values cannot be seen at all in the figure. Please change the figure. A suggestion is to place the dots next to each other, instead of on top of each other.
Figure 8: please add the squares that indicate where the plastics are also on panels (d)-(f). Indicate which plastic is in which panel (white, transparent, black) so classification results can be compared.
Citation: https://doi.org/10.5194/egusphere-2025-3804-RC2 -
AC2: 'Reply on RC2', Alessandro Fabrizi, 14 Jan 2026
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-3804/egusphere-2025-3804-AC2-supplement.pdf
-
AC2: 'Reply on RC2', Alessandro Fabrizi, 14 Jan 2026
Peer review completion
Journal article(s) based on this preprint
Data sets
Hyperspectral and multispectral reflectance of agricultural plastic films Alessandro Fabrizi et al. https://doi.org/10.5281/zenodo.14336253
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,900 | 1,039 | 142 | 3,081 | 136 | 155 |
- HTML: 1,900
- PDF: 1,039
- XML: 142
- Total: 3,081
- BibTeX: 136
- EndNote: 155
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Alessandro Fabrizi
Peter Fiener
Kristof Van Oost
Florian Wilken
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(1775 KB) - Metadata XML
Main points
Fabrizi et al. present an experiment investigating the potential monitoring of plastic film residues from agricultural activities generated at various stages (e.g., after use, management). The team uses advanced sensors to obtain optical properties of the subset sample of the plastic films in a controlled experiment outdoors and in the field. Datasets are made open-access, key part of supporting future open science. Several aspects about the study must be addressed before it is considered for publication especially some additional literature review as some statements can be rephrased based on facts, citations added to support text, polymer type, terminology used and presentation of the findings to put them into context of current research advances.
Critical Review Points
Yes it is within the scope of the journal.
Yes, although certain aspects have to revised to better illustrate the novelty or the text needs to be rephrased see specific comments below.
Yes, it combines use of remote sensing tools to potentially monitor the distribution of plastic films used in agriculture that can contribute to environmental risks as explained in Line 23-42
Yes, it is of broad interest considering the issue of plastic pollution and also use of plastics to improve farming.
The objectives are well defined in line 79-82. It is important that the presentation of the findings including discussion is improved
The methods are uptodate but the text should be improved to give sufficient details allowing duplication, justification or understanding of the steps applied. More information on the methods is crucial considering Fabrizi et al. suggest this is one of the first few studies on the topic covering the elements discussed.
In general the references are ok, additional literature review is need to better position the current study with prior studies
Suggest literature review
https://doi.org/10.1016/j.solener.2020.11.058
Case studies - including use of hyperspectral tools, proximal measurements of LDPE materials, use of satellite and airborne sensors, algorithms development, data for comparisons
State-of-the-art reviews
Given the sample size used the title should reflect this is a ‘case study’ with a subset of materials used.
The quantitative aspects are missing in the abstract especially how many samples were done to build the open-access library.
Line 11-the sentence should be revised as remote sensing has been used extensively in agricultural plastic-based mapping see the suggested references in point 12 including the cited works in the references above.
Line 16-it must be discussed more and presented in the results to support the statement. How is the efficacy shown? See question raised in point 5 objective 3 comment.
Line 18-the data report is based on reflectance and yet the Fabrizi et al. call this absorption peaks? Is this not commonly termed ‘absorption feature’? revise throughout the text as suggested
Line 19-this is absorption features are relevant in hyperspectral data make it clear because with the camera system is this evident with the five-bands?it is not evident in the Figure 6
It is fairly written with the specific recommendations for improvement being on the figures and supplementary material.
Proof reading of the text is recommended to mitigate some of the spelling mistakes and phrases that were unclear.
If automated proofing tools were used it must be declared?
Line 21-the cited works report on microplastics right?out of curiosity are these figures accounting for the ocean that cover 70% of the Earth surface and the amounts being reported at sea?
Line 29- ‘Despite….monitoring plans’ provide references and rephrase as the main message is unclear
Line 33- are the agricultural soils a source or is it the activity?
Line 38-the references style should be revised check the formatting.
Line 43-define LDPE at first use.
Line 44-can the multiple years be enumerated
Line 48-49 reference to the term ‘earlier studies’ are missing provide citations
Line 49- missing source for the sentence
Line 53-the term broad also include satellite remote sensing?
Line 55-plastics in marine have received much interest but not necessary started the topic or efforts to use UAVs see the suggested literature in point 12
Line 60 also see the case studies provided in point 12
Line 72-is it about the RGB camera or other aspects such as flight altitude, size of the samples, environmental conditions?
-how does RGB camera increase the spatial resolution?is this also accounting for the sensor capability in terms of megapixels or lenses used? Same question would be applicable to the Micasense camera used
Line 81-objective (ii) is repeated, is the goal to define or to propose workflows?
Paragraph starting Line 91-was the ASD run in reflectance or digital number mode?
Line 110-what does it mean ‘in case of ….’ Rephrase
Line 115-what is a field soil? Is this about a sample or..
Section 2.1.3 -examples should be given to further demonstrate what is written it can be in the supplementary section or here. How effective were the steps done, was it really noise or something?A flowchart might also be useful
Line 147-why only the mean is presented?
Line 162-with absorption features this is typically the convex and hull locations
Equations used are they from the original publications or other values were applied in the current manuscript?
Line 182-replace calibration with ‘referencing’, provide details about the panel used name and properties
Line 184-does placing mean the same as geo-referencing?
Line 195 define GNSS and GIS at first use,
Line 195 provide the images of the samples to support and visualize the text, how many recievers were used?
Line 210-312 random points does it mean training data was balanced for each class in like 211?how many pixels were plastics exactly?
Line 229 what are the numbers in the brackets (3.1….), why the specific index values as features can the results be shown of the products?
Line 231-233-is this part of step 4 or summary of the 4 steps?
Line 244-245 about Figure 3 can be confirmed by a simplified spectral unmixing test of the soil and plastic at simulated pixel amounts say 0.1, 10, 50, 100%. The spectra of the background material is needed also to check for a baseline correction if needed.
Line 261-based on the analyses and Figure 4 plot is the ND_1715 really useful? Or what is the benefit of using it in monitoring plastics?
Figure 4 can be supported by a separability metric value. Figure can be made bigger to allow visual check to see if there are overlaps such as in the (c)
Line 273-what treatments?
Discussion
Line 323-polymer type details are missing see the experiment from Jones et al 2021 point 12
Line 326-the differences is shape can be better visible if all spectra are scaled or normalized
Line 329-336 where is this shown in the manuscript considering the assumption referencing to the white target was done every four minutes so the lighting is relative or as stated Lambertian and the plastics are also unique. The available data should be used to provide supporting proof to the statements here, was this the same situation in the crumbled data hints are in Figure 5 also?
Line 344 to 351-one suggestion to confirm the statements would be spectral unmixing and should be explored here even with a simplified assumptions of the mixtures or the ones from the experiments.
Line 353-a plot can be used to confirm or demonstrate the sentence as difference are only presented in Figure 5
Line 355- sentence unclear ‘It must be accounted for, though,…..’
Line 367-explain how this was feasible and the link to glint issue
Line 369-the data is available why not evaluate the brightness thresholding approaches or at least provide the raw data for potential use in future studies?
Line 380-it is possible to explore the statements or claims here for the very fine pixel satellites for example 30cm/pixel from Pelican or WorldView?
Line 383-is relevant for machine learning classification but also depends on the size of the objects of interest?
Line 386-391-the discussion should also account for the labelling dataset size, supervised or unsupervised approaches
Line 394-contradicts a bit with the figure 3 as the spectra from Silt loam had higher values and yet in Figure 5 differences are high for sandy soils. In any case silt loam seems a better background—a figure is missing here to also support this sentence as well as paragraph to make a clear distinction between the films bright or dark ones.
Line 400-define the spectrum ranges referred to as visible, near infrared and be clear what exactly it means no any ‘unique features’? there is a peak in the green for the WB_S and WB_L also there is an absorption feature in the near infrared of T_L
Proposed algorithms can be tested see the suggested experiments in point 12-a spectral matching can be applied using the created libraries.
A clear message about the use of UAVs should also combine the limitations that can be complemented by satellites.
Line 436-should be revised as there are sensors offering open-access to hyperspectral remote sensing information of the environment see the suggested citations using drone, aircraft and satellite tools.
Figure 1 should be made bigger to fit margins of the text, labelling for the setup should be improved larger text, where is the white reference, distances can be added
Line 112-a figure should be included to show the ‘different conditions’ and the soil sample can be included as a foto
Line 121- the sentence must be highlighted in the figure 1 or a figure be provided
Table 1-polymer type is missing, is the header use specific or it is better ‘Common Use’other give reference for the use defined. How was the thickness obtained or was provided by manufacturer explain.
Figure 2 can have the coordinates northing and easting added, figure made bigger, RGB image of the drone survey is missing only a side view is given in (b).
Figure 3 should be made bigger and improved better with subplots of each material or groups as in the legend black films, transparent, white films.
Figure 5,the plots of the actual reflectance are required here as the difference does provide minimal details although the format of the figure is good. Either have the actuall measured values presented or in the supplementary material because what is the meaning of the difference? Was this maybe due to background or rather experimental setup? This is not evident in the column 1 with clean samples and the lower values also what does it mean negative difference?
Figure 6-add joining line for visualizing the multispectral data and subplots could improve the figure
Figure 8 too small
Yes, just cross checking of the values in original studies with the stated manuscript values.
Methods and related material require additional material and detail including improved figures.
Confusion matrix and important feature summary for the Random Forest classifier are missing.
Overall citations are ok but the recommended citations should be reviewed or justification of excluding them provided.
Sentences in the introduction also require supporting citations for completeness
It is challenging to have a fit all format especially with the community using various programming languages. However, Fabrizi et al. is highly commended for making the dataset open access.
There are a few issues with the provided dataset that should be improved especially the presentation and metadata structure.
The excel sheet can be combined with the pdf instead of having to check five separate files. The excel sheet was challenging to open in excel and required time to decipher the format
Check the spellings of the headers. see "Wavelenght"