Leveraging Soil Geography for Land Use Planning: Assessing and Mapping Soil Ecosystem Services Indicators in Emilia-Romagna, NE Italy

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is an important work for land-use planning using innovative soil geography tools and methodologies. I would just comment on few things that could improve their work: in the presentation of the results, the authors give detail of the SES by soil provinces (pedolandscapes), therefore, I would suggest that the authors present the soil data by soil provinces. It would be more schematic and understandable for the lector the relevance of soil properties on the definition of SES.

Comments for author File: Comments.pdf

Author Response

Authors’ answers to reviewers

 

Reviewer #1

This is an important work for land-use planning using innovative soil geography tools and methodologies. I would just comment on few things that could improve their work: in the presentation of the results, the authors give detail of the SES by soil provinces (pedolandscapes), therefore, I would suggest that the authors present the soil data by soil provinces. It would be more schematic and understandable for the lector the relevance of soil properties on the definition of SES.

We thank the reviewer for her/his positive feedback and her/his suggestions, which we fully accepted and implemented in the revised version of the manuscript

Review report

L = Line, R: reviewer’s comments

Summary

This paper presents updated results from a soil-based ecological services (SES) assessment framework, that was first developed and adopted ten years ago in Emilia-Romagna region (22.510 km²), Italy, by the same authors. The authors cover in the present work a larger area, including in addition to the plains the mountainous area, improved the resolution of the digital soil mapping from 1 km grid to 100 m resolution. Additionally, they incorporate the use of the Normalized Difference Vegetation Index (NDVI), the non-parametric machine learning techniques, the Revised Universal Soil Loss Equation, Quantile Random Forests, and thematic maps into the analysis. As results, the authors obtained a map with eight indicators of soil ecosystem services and a table with the average values of the ecosystem service indicators for the mapping units.

General concept comments

In the paper it is missing an introduction to soil-based ecological services (SES) or the contribution of soil to ecological services. The authors write the SES in the abstract (buffering capacity, carbon sequestration, erosion control, food provision, biomass provision, water regulation, water storage, and habitat for soil biodiversity). However, no direct explanation about the SES in the text. I suggest the authors to describe the SES and not only give a reference of their previous work on the subject. They can write a brief description at the paragraph where they explain that this work is an update of a previous one (L103-118).

As suggested by the reviewer we included a brief description of SES in the introduction (lines 106-111 of the track change document). We also emphasized the information about SES presented in Table 4 (former Table 2)

L279-281, it is written: “Although time invariant, the spatially explicit indicator scores allow ranking of the different soils in terms of their potential ecosystem services supply, based on routinely available soil data”.

R: The SES are mapped by “soil province”, the authors do not present the routinely available soil data” by soil province”. It would be convenient to present soil data from each soil province to understand better the resulted SES mapping units.

As suggested by the reviewer we added two new tables (one for the pedolandscape of the plain and one for those of the Apennines) in the revised version of the manuscript reporting the descriptive statistics (mean values, number of observations, Standard Deviation) of the data sets for each soil province (pedolandscape)  (lines 312 and 314 of the track change document). The following text was added to revised version of the manuscript to present soil data at pedolandscape level:

“As the pedolandscapes describe the regional soil geography, delineating area where soils are similar in terms of climate, topography, parent material and vegetation,  they represent a useful reference to analyse results in terms of SES supply. For this reason, and to highlight the relevance of soil properties on the definition of SES considered in this work, Tables 2 and 3 reports respectively for the plain and the Apennines the descriptive statistics for the basic soil properties observed in the pedolandscapes of the region”. (lines 306-311 of the track change document)

 

Specific comments

The scope of the paper fits perfectly the journal scope. The authors have vast experience in the subject. The manuscript is well written and the improved procedures to assess the SES compared to previous work by the same authors make it relevant for the field. Although 54% of the cited references are 10 years old or older and they have several self-citations, I consider the presented references relevant to the topic.

They incorporate the use of updated digital mapping procedures that improved the accuracy of the assessment, making the analysis more robust. All tables and figures are pertinent and easy to understand.

The main question or topic is not original, but the results of this work could be considered an advance of current knowledge on the subject.

Some small details in the text

L121-125 It is written that Emilia-Romagna administrative region has an extension of 22,509.67 km2 and in L124-125 it is written that the north-eastern part occupied by the Emiliano-Romagnola plain is ca. 12,032 km2, which they write it is 47.8% of the total surface area (22,509.67 km2).

R: This is wrong, 12,032 km2 is 53.5%, not 47.8% of the total surface area.

Corrected as suggested by the reviewer; (lines 129-133 of the track change document): “…the north-eastern one (47.8% of the total surface area) occupied by the Emiliano-Romagnola plain (ca. 12032 km2) which is delimited to the East by the Adriatic Sea, and the south-western one characterized by the presence of the Apennines range (hilly for 24.2% of the area and mountainous for 22.4%).”

L240-241 It is written 24 Agricultural Districts, nine in the alluvial plain, eight in the low Apennines, and eight in the medium and high Apennines

R: 9 + 8 + 8 = 25 Agricultural Districts. There are also 25 Districts in Figure 2.

Corrected as suggested by the reviewer; (line 225 of the track change document)

L303 It is written “LCC classes”.

R: In LCC it is implicit “classes”. The word “classes” should be erased.

Erased as suggested by the reviewer; (line 345 of the track change document)

L543, it is written “(e) pH, (e) Organic C %”.

R: It is written two times (e), should be written (f) Organic C %.

Modified as suggested by the reviewer; (line 589 of the track change document)

L596 It is written Figures 5a-h

R: It should be written 6a-h

Modified as suggested by the reviewer; (lines 642 of the track change document)

L601 It is written Figure 3S

R: it should be written S3

Modified as suggested by the reviewer; (lines 647 of the track change document)

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article presents a methodology for assessing and mapping soil ecosystem services in the Emilia-Romagna region of Italy, using an indicator-based approach founded on digital soil mapping techniques. The authors applied different DSM approaches for plain and mountainous areas, creating maps of eight soil ecosystem services at 100m resolution. The work aims to provide tools supporting spatial planning and soil resource management. It should be stressed that the study indicates the comprehensive methodological approach, especially in using the different DSM techniques adapted to terrain specificity (geostatistical simulations for plains, machine learning algorithms for Apennines), is well-justified and appropriate. Also, the implementation of results in spatial planning according to Emilia-Romagna regional law represents a valuable applied aspect. The study presented includes extensive data utilization of diverse data sources (over 40,000 observation points, soil maps at different scales, remote sensing data), which enhances the credibility. 

Nevertheless, the study avoids the non-homogeneity issue of the dataset used in the analysis. It can be evident in the significant differences in data density between plain areas (67-84% of all data) and mountainous regions, which may affect the comparability. While authors apply different DSM approaches, they do not thoroughly analyze the impact of this heterogeneity on result uncertainty. 

Another weak point of the study regards the QBS indicator limitations. The soil biodiversity indicator based on only 330 observation points (compared to thousands for other properties) significantly limits the credibility of the BIO map. Authors acknowledge that the map has "little applicability at the local scale," which undermines its utility. 

The authors use in the analysis the local pedo-transfer function which may not account for the full variability of soil conditions in such a diverse region as Emilia-Romagna. Local calibration on predominantly plain soils may introduce systematic bias for mountain conditions. No validation against independent mountain datasets was performed for derived properties (BD, Ksat, WCFC). 

The uncertainty aspect in the data analysis is not well represented. Although authors mention uncertainty maps in supplementary material, the main text lacks a thorough analysis of error propagation throughout the entire modeling chain. The results of the modelling can only be applied to the studied region due to 0-1 normalization, which is specific for the studied site. 

The soil water storage is a specific value that depends on the soil depth. This measure should be clearly defined in the methodology section. Probably the section 3.2.5 should be rephrased as the “Water content at field capacity. But it should be stressed that the field capacity depends on the soil type. It means that the assumption of the -33 kPa for the whole soil data set may not be precise.

The modelling issue requires an explanation of the data set division. It seems strange that the random division of the dataset occurs in random forest analysis. I think it can be possible, but only in the case of the division by the dominant land use categories, as is indicated in Figure 2. Authors should explain the cross-validation scheme used in the study (5 or 10 validation folds). It should also be explained what type of RF was used, classification or regression.

The work represents a valuable contribution to the development of methodology for assessing soil ecosystem services at the regional level and presents a practical application of modern DSM techniques. However, due to significant methodological gaps, particularly in uncertainty analysis, limited credibility of certain indicators (BIO), and overly optimistic conclusions regarding methodology universality, the article requires major revision before acceptance for publication.

Author Response

Reviewer #2

The article presents a methodology for assessing and mapping soil ecosystem services in the Emilia-Romagna region of Italy, using an indicator-based approach founded on digital soil mapping techniques. The authors applied different DSM approaches for plain and mountainous areas, creating maps of eight soil ecosystem services at 100m resolution. The work aims to provide tools supporting spatial planning and soil resource management. It should be stressed that the study indicates the comprehensive methodological approach, especially in using the different DSM techniques adapted to terrain specificity (geostatistical simulations for plains, machine learning algorithms for Apennines), is well-justified and appropriate. Also, the implementation of results in spatial planning according to Emilia-Romagna regional law represents a valuable applied aspect. The study presented includes extensive data utilization of diverse data sources (over 40,000 observation points, soil maps at different scales, remote sensing data), which enhances the credibility. 

We thank the reviewer for her/his feedback and for highlight several shortcomings of the presented SES mapping approach. We tried to address her/his observations in the revised version of the manuscript even if for some remarks it was not clear which actions should we take to incorporate the recommendations without fully altering the current structure of the paper, whose mail goal is to emphasize the role of the knowledge of soil geography in supporting the  ranking of the different soils in terms of their potential ecosystem services supply at regional scale, based on routinely available soil data and on the update of a methodological framework published nearly a decade ago (Calzolari, C.; Ungaro, F.; Filippi N.; Guermandi, M.; Malucelli, F.; Marchi, N.; Staffilani, F.; Tarocco P. A methodological framework to assess the multiple contributions of soils to ecosystem services delivery at regional scale. Geoderma 2016, 261, 190-203. https://doi.org/10.1016/j.geoderma.2015.07.013; Citation Indexes 92, Policy Citations 3).

Nevertheless, the study avoids the non-homogeneity issue of the dataset used in the analysis. It can be evident in the significant differences in data density between plain areas (67-84% of all data) and mountainous regions, which may affect the comparability. While authors apply different DSM approaches, they do not thoroughly analyse the impact of this heterogeneity on result uncertainty. 

We tackled the issue of non-homogeneity by tailoring two different Digital Soil Mapping (DSM) approaches to the different data density in the two main landscapes of the region. On the plain where available data exceed 30,000 observations, we resorted to Sequential Gaussian Geostatistical Simulations (SGS) conditional on the mean value of the 1:50,000 soil map, which is an autoregressive approach to spatial variability modelling, fully justified in terms of robustness by the very high number of data points. In the case of the Apennines where data are in the order of 9,000 observations, we resorted to ML learning algorithms (regression Quantile Random Forest) where the spatial modelling is supported by multiple correlations with several environmental covariates (continuous and categorical). Results uncertainty was quantified in terms of standard deviations of models estimates (assessed via SGS for the plain and QRF for the mountains) for each basic soil properties and local heterogeneity is implicitly considered when assessing the standard deviation of estimated values at each node of the estimation grid (SGS) or at each grid cell (regression QRF). The issue of comparability was considered in terms of possible artifacts in the maps along the border of the two part of the regions, but thanks to the integration of the knowledge of soil geography (geometry of 1:50,000 soil map polygons) in the two mapping approaches such artifacts were not detected in the resulting maps of soil properties which are coherent with the current status of soil knowledge and at the same time highlight to different degree the within polygon variability of soil properties.

Indeed, we agree with the reviewer and acknowledge that the assessment of the (spatial modelling) uncertainty propagation for each SES indicator calculations is not present, but this was not the original scope of the work which was intended to provide local stakeholders at different administrative levels  with the information on SES potential supply requested for planning and land policies.  To this goal, we are not presenting in the paper a strictly bio-physical quantitative assessment of soil-based ecosystem services via DSM, for which an assessment of spatial uncertainty would be not only advisable but almost necessary, rather a group of interval normalised indicators of potential soil-based ecosystem service supply, and it is important to highlight that the 0-1 interval normalization scale the indicators values over the observed (local) variability to derive a ranking of the SES potential supply.  As such, there is not the strict necessity “to put uncertainty into numbers”, rather to provide ”a quantified judgement” (D. Spiegelhalter, The Art of Uncertainty, Pelican Book, 2024, pp- 219-221)

Further evidence of the widespread lack of assessments of uncertainty in the modelling of ecosystem services at various scales is given by the fact that none of the integrated assessment tools and platforms most used for their ES evaluation and mapping (e.g., ARIES, InVEST, ESTIMAP) currently provide even a qualitative assessment of the estimation uncertainty. The issue is then relevant and characterizing, propagating and communicating uncertainty are indeed unique challenges that deserve attention by the ES scientific community as it is among the key factors that could increase the uptake of ecosystem service assessments.  Nevertheless, the ongoing process of gradual integration of SES assessment results into planning practices by local stakeholders (i.e., municipality) would suggest that the lack of uncertainty associated to the SES indicators is not the major bottleneck in improving the  uptake and integration into decision-making processes in Emilia-Romagna.   Rather, this is to be found in the poor soil literacy level among a large majority of policy makers and land planners, and even more in the conflicting interests and priorities related to land use and its changes. This includes competing demands for land for development, agriculture, and conservation, often with different stakeholders holding opposing views.

In the revised version of the manuscript we highlighted this shortcoming of the modelling flow in the conclusion; we added further clarifications on the modelling approach presented where requested.  

 

Another weak point of the study regards the QBS indicator limitations. The soil biodiversity indicator based on only 330 observation points (compared to thousands for other properties) significantly limits the credibility of the BIO map. Authors acknowledge that the map has "little applicability at the local scale," which undermines its utility. 

We fully agree with the reviewer’s comment, and we clearly stated in the paper that the utility of this indicator is at this moment in time limited and that further data would be necessary to verify the impact of the different drivers on soil biodiversity and on its spatial patterns at local scale.  

In the paper we acknowledge that “Despite the good validation statistics, given the limited number of QBS-ar data available at regional level, the BIO indicator map is to be considered as a preliminary description at regional level but with little applicability at local scale. However, the machine learning algorithms underlying the DSM have highlighted the relevance of remote sensing vegetation spectral indices as predictors of QBS-ar, providing a provisional map that could be the basis for validating hypotheses on the mechanisms that determine the spatial distribution of the BIO indicator at the regional scale”. (lines 885 – 891 of the track change document)

Notwithstanding these limitations, though, the error statistics for the QRF calibration and validation were satisfactory and the presented map is believed to provide a first picture of the current state of soil biodiversity at regional scale given the available information. It is also worth noting that the map presented in the paper for the supporting SES indicator BIO based on QBS-ar data is the first of this kind at regional level in Italy, and among the few available at EU level as well.

The authors use in the analysis the local pedo-transfer function which may not account for the full variability of soil conditions in such a diverse region as Emilia-Romagna. Local calibration on predominantly plain soils may introduce systematic bias for mountain conditions. No validation against independent mountain datasets was performed for derived properties (BD, Ksat, WCFC). 

Although the regional PTFs were calibrated using soil data from samples which were mostly from soils of the plain and hilly areas of the region, the input data from mountain soils were fully within the PTF calibration ranges of all predictors (i.e. textural fractions, C org) providing estimates which proved to be coherent with field evidence and expected soil hydrological behaviour, as assessed by local stakeholders who are also co-authors of this paper. It is worth noting that one of the reasons for adopting standardized indicator scores for SES assessment and mapping is that it allows for a relative ranking of soil functions  without resorting to biophysical units whose interpretation would require a degree of soil literacy which is unfortunately lacking among land planners and policy makers at all administrative levels.

The uncertainty aspect in the data analysis is not well represented. Although authors mention uncertainty maps in supplementary material, the main text lacks a thorough analysis of error propagation throughout the entire modelling chain. The results of the modelling can only be applied to the studied region due to 0-1 normalization, which is specific for the studied site. 

To address the issue of uncertainty, the following piece of  text was added to the conclusion:

 “…On the other side, a shortcoming in the current modelling methodology is the lack of explicit consideration and assessment of how spatial uncertainties propagate through the SES modelling workflow. This means that DSM models and PTFs produce output in terms of potential SES supply indicators maps, but the reliability and confidence of those indicators, given the inherent uncertainties in inputs and model assumptions, are not quantified. This is a well acknowledged issue in DSM but not often considered in ES modelling [77,78]. For some of the SES indicators presented in this work grid-uncertainty maps, which indicate uncertainty directly could be provided straightforward, as for BIO and BIOMASS resorting to their estimation standard deviations (Figures S4 and S5), while for others the uncertainty propagation assessment is more complex and should resort to different postprocessing techniques depending on input data characteristics. For categorical data, as in the case of PRO, class purity could be assessed resorting to disaggregation techniques, while for continuous data, as for most composite SES indicators presented in this work, uncertainty effects might be additive or multiplicative. It is worth noting that for some authors, though, uncertainty assessment would not be strictly required for indicator-based modelling [79-81], particularly when the ES assessment aims to raise awareness in policy makers and when the focus is on highlighting potential magnitudes and relative rankings as in the indicator-based approach presented in this work...” (lines  1076 – 1094 of the track change document)

We fully agree with the reviewer though that the results of the modelling presented in the paper can only be applied to the Emilia-Romagna region as the 0-1 normalization is site specific.

 

The soil water storage is a specific value that depends on the soil depth. This measure should be clearly defined in the methodology section. Probably the section 3.2.5 should be rephrased as the “Water content at field capacity”. But it should be stressed that the field capacity depends on the soil type. It means that the assumption of the -33 kPa for the whole soil data set may not be precise.

We fully agree with the reviewer that soil water storage is a specific value that depends on soil depth. In the methodology section it is specified that our analysis is focused on topsoil for a reference depth of 0-30 cm: all the SES indicators are referred to this depth interval. We have better clarified this in the caption of Table 3, and we corrected an error in the formula for the calculation of the indicator referred to the possible presence/absence of a shallow water table within 30 cm depth.  As suggested by the reviewer we stressed that water content at field capacity depends on soil characteristics by adding these lines at the beginning of section 3.2.5.:

“This indicator is based on the water content at field capacity for the reference depth interval of 0-30 cm, corrected for the content of coarse fragments and the possible presence of a shallow water table within the reference interval considered. Although the soil water potential at field capacity depends on soil type, the WAS indicator is based on the estimated water content at the reference tension of -33kPa, as recommended in absence of field measurements [75].” (lines 835 – 839 of the track change document)

We modified the heading of section 3.2.5. as suggested by the reviewer (lines 834 of the track change document)

 

The modelling issue requires an explanation of the data set division. It seems strange that the random division of the dataset occurs in random forest analysis. I think it can be possible, but only in the case of the division by the dominant land use categories, as is indicated in Figure 2. Authors should explain the cross-validation scheme used in the study (5 or 10 validation folds). It should also be explained what type of RF was used, classification or regression.

To address and clarify the two modelling issues raised by the reviewer we reports below the screenshot of the R script compiled for the implementation of the DSM approach using regression Quantile Random Forest  to map, in the example below, topsoil (0-30 cm) clay content:

In the script, at line 256  different DSM model are preliminarily tested. Then, in lines 261 to 265 the dataset is divided in two subsets:  training with 75% of the observations and testing with 25% of the observations.

At lines 274-277 the regression random forest is calibrate using the predictors listed in ( ) in lines 274-275. Model setting are specified in line 276-277: input data (df.traina), method (quantile random forest, qrf), control method (cross-validation, cv, with number=10, validation folds).

As suggested by the reviewer we specified the cross-validation scheme used in the study (10 validation folds), and explicitly clarified the type of RF was used, i.e. regression. The revised text is then:

…”The models were run in R environment [70], adapting the DSM workflow developed by FAO [71] resorting to Quantile Random Forests (QRF) [72,73]. QRF provides a univariate or multivariate quantile regression forest and returns its conditional quantile and density values”  (lines 418 – 420 of the track change document).

…“Initially, using the same R scripts, the data sets of each variable were randomly divided into a train and a test set, with respectively 75% of the data used for the SGS implementation and QRF calibration, and 25% of the data retained kept for map validation. The script performs a 10-fold cross-validation splitting the training dataset into subsets and using nine subsets for calibration and one subset for validation. This process is repeated 10 times, with a different subset used at each iteration.”  (lines 433 – 436 of the track change document).

 

The work represents a valuable contribution to the development of methodology for assessing soil ecosystem services at the regional level and presents a practical application of modern DSM techniques. However, due to significant methodological gaps, particularly in uncertainty analysis, limited credibility of certain indicators (BIO), and overly optimistic conclusions regarding methodology universality, the article requires major revision before acceptance for publication.

We eventually thank the reviewer for considering the work a valuable contribution to the development of methodology for assessing soil ecosystem services at the regional level. We believe that we properly addressed the methodological gaps highlight by her/his comments by clarifying and revising the text of the original draft as suggested.

We made more explicit the issues related to uncertainty analysis within the context of SES assessment and provided examples for two indicators of how this could be transferred into results.

We though disagree on considering BIO as an indicator of “limited credibility”, rather an adequate one whose representation needs further testing,  as we clearly stated that its evaluation as to be considered a first attempt to map the status of soil biodiversity as based on the available data of QBS-ar and, as such, it would deserve further improvement in  the future. Nevertheless given the statistical accuracy of the DSM model and the more than satisfactory error indexes for the validation data set, we  still believe it is appropriate and useful to present the results for BIO also in the revised version of the manuscript.

We clearly stated that the SES assessment was tailored on data availability in the study area and rely on a set of methodological tools developed in Emilia-Romagna over decades of soil survey and investigation carried out with local stakeholders. For this reason the results of the modelling presented in the paper can only be applied to the Emilia-Romagna region as the 0-1 normalization is essentially site specific.

The conclusion of the revised version of the paper were cleared of any “optimistic claim of universality” of the methodology and shortcomings related to uncertainty analysis were clearly highlighted; furthermore, we added in the supplementary materials two examples of gridded uncertainty maps for two indicators (Figure S4  and S5).

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you very much for the helpful explanation. The authors' responses to my comments are satisfactory and thorough.