Early adopters of nature-based WW technologies seek methods to find suitable location and minimize the risk of implementing. The goal of our work is precisely to validate a prediction technology in the context of a decentralized WWT, specifically a variety of constructed wetland—so-called METland.
3.1. Multi-Criteria Evaluation of Two Independent Areas
Two different regions located at Mediterranean and oceanic climates were analyzed using MCE. Previous to the application of the MCE method, a correlation analysis was performed between all the factors covered by the study. The results of correlation achieved a very low value (less than 0.3 in all cases); therefore, the factors were non-redundant. The maps corresponding to each of the factors were shown once normalized (0–255) for each study area with a raster grid definition of 25 × 25 m (
Figure 4).
Subsequently to the performance of the MCE for both locations, the suitability maps were obtained (
Figure 5). Each pixel value indicates the portion of the territory suitable for the location of a METland. The higher values reveal the most suitable places for such treatment contrary to the lower values that point out less appropriate locations for implementing it.
Regarding the suitability map at the provincial level (
Figure 5), the main difference between the two provinces was the proportion of suitability areas. In the southwest of Málaga, a vast area of low suitability could be noticed, while in Bizkaia the low suitability areas were smaller. In the overall visual comparison between the two provinces it should be pointed out that Bizkaia had more intermediate values of suitability and Málaga had more abrupt changes. This perception could be due to the different area of each province; Málaga had a larger extension than Bizkaia. Another reason was regarding the frequency histogram of the suitability map because it had a very different distribution in each province. To verify this hypothesis, the distribution of parcels was represented according to their level of adequacy (
Figure 6).
Regarding the detail scale map, it should be noted that the oceanic location had a distribution of rural population widely dispersed in villages throughout the territory. While in Málaga, the population was grouped into larger settlements. Therefore, Málaga had an unequal population distribution between the coast and the inland area, characterized by a higher population density along the coast and medium/large urban settlements in other areas. As is noticed in
Figure 5, the higher suitability values were congregated in areas close to dwellings and river channels. In addition, the maps showed that in both provinces the most suitable areas for METlands were located close to the population establishments or even isolated homes. This distribution proved the importance of treating the WW near the population centers. Moreover, in the mountain areas (in Málaga the southwest area and in Bizkaia the southern and the northeast area) lower suitability values were obtained, possibly due to the decrease of population cores and an increase of slope and unfavorable weather conditions.
In order to perform a more in-depth analysis of the values represented on the map, a histogram of suitability distribution was plotted (
Figure 6). In Málaga, 6609 km
2 (10,574,876 pixels) out of a total of 7307 km
2 were suitable for METlands, representing 90.45% of the total. Whereas, the percentage of pixels suitable for such solutions in Bizkaia was 88.55% (3,090,217 pixels). The percentage of suitable area was higher compared with the study performed by Demesouka et al. (2013) [
43] for nature-based WWT systems. However, it should be noted that Demesouka et al. (2013) [
43] applied very prohibitive restrictions such as 5% of maximum slope, excluding from the analysis all the areas with higher slope and achieving lower rates of suitability.
On one hand, Málaga showed a normal distribution of the suitability, with the majority of pixels in the range of 160 to 200. In comparison, Bizkaia values were more regularly distributed where most of the pixels had adequacy between 135 and 160. On the other hand, Málaga achieved a percentage of suitable parcels slightly higher than Bizkaia, which might be due to the influence of the land uses map (
Figure 4). Specifically, in the standardized map of land uses, Bizkaia presented a mean value of 89.17, while in Málaga it was 139.72, causing the great influence of land use in the final results of each province (
Table 4). To test this hypothesis, the reclassified land use map had been overlaid on the suitability map. Thus, it could be verified that most of the areas of lower suitability mainly correspond to the land uses with a lower value in
Table 1. Finally, in order to assess a detailed study of the data, 1%, 5% and 20% of pixels with higher suitability were analyzed (see
Figure 7).
The result of this application was some Boolean maps that represent a specific percentage of the best suitability areas out of the total. For Bizkaia, the 1% of pixels (30,902 pixels or 19.31 km2) with better suitability correspond to suitability values higher than 230, the 5% (154,510 pixels or 96.57 km2) at values greater than 216 and the 20% (618,043 pixels or 386.28 km2) to values greater than 192. For Málaga, the 1% of pixels (105,748 pixels or 66.09 km2) with higher suitability correspond to adequacy values greater than 219, the 5% (528,743 pixels or 330.46 km2) to values greater than 205 and the 20% (211,4975 pixels or 1321.86 km2) to values greater than 192.
To sum up, a map was obtained with the most suitable areas for the construction of METlands in both provinces (
Figure 5). The results were better than expected and revealed optimal locations within the provinces analyzed. This analysis determined the most suitable areas on a large scale; however, further analysis could be performed for specific areas like specific municipality or areas with isolated houses. It should be noted that, for local analysis, it would be necessary to reduce the pixel size for a more accurate METland location. Regarding the comparison between the two provinces, similar results were obtained and the same methodology could be applied in other areas.
For replicability within other nature-based solutions, factors considered in the analysis should be adapted. The variables that determine the nature-based technology implementation must be listed (environmental, social and economic). Afterwards, experts should decide the importance of each variable in the final location for the system. The methodology proposed would eventually provide a suitability map from the area of study. We expect to help the stakeholder in the selection of new habitats to implement nature-based technologies for WWT, and set some guidance in the variables that specifically could influence operation of constructed wetlands. In this sense, some generic analysis had been conducted for WWT [
41] and nature-based solutions [
42], thus a specific analysis should be address for each technology and situation, taking into account the stakeholders interests. Additionally, some authors have previously analyzed different variables for specific technologies such as stabilization ponds [
44] or restored wetlands [
49,
87].
3.2. Results of the GSA
The results of the GSA for each of the model components (factors and weights) were compiled in
Table 5. These results indicated that the variation in three of the factors contributed decisively in the model results, for all the study area. The determining factors were land uses, distance to population centers and distance to river channels. The meteorological criteria (average temperature and maximum precipitation) assumed a low contribution to the final result. These results obtained by the SA conclude that only a small number of the input factors were found to have a significant influence on the model results. This conclusion was corroborated in similar studies based on GIS-MCE models [
31,
32,
52].
In Bizkaia, the order of importance of the criteria by the GSA was land use (38%), distance to the population centers (20%) and distance to river beds (20%). In Málaga, the order was land use (31%), distance to river beds (23%) and distance to population centers (22%). It could be highlighted that the proportion was not equal in both provinces since in Bizkaia land uses had more influence on the results as was anticipated in the discussion of the suitability maps. First-order sensitivity indices of the criteria of land use, distance to population centers and river beds were responsible for 78% of the output variability of the model. The influence of weights and other criteria was nearly negligible, confirming that the weights established for the variables were robust and the addition of small variations did not influence the final results of the model. Similar results were obtained by Vavatsikos et al. (2020) [
45] with a 5% of variation for the weights and by Gómez-Delgado and Tarantola (2006) [
52] with a 20% variation. Instead, for variations between 50% and 75% of the weights, some studies presented that the weight had a higher influence on the results than other criteria, without reaching the factors that represent the major source of variability [
32,
52].
In addition, the difference between the total effect sensitivity index (STi) and the first-order sensitivity index (Si) is a measure of how much each factor is involved with the interaction with other factors in the model. For significant differences, the value should be greater than 0.2. In this analysis, the differences were never higher than 0.0121. Therefore, the variation in the results was due to the action of the factors individually and not in combination with the others. This circumstance was corroborated through the sum of all STi, which were almost equal to 1 showing that potential interactions present in the model had no influence on the variability of outcome.
From the results obtained it could be deduced that the most influential factors in the final model according to the SA were land uses, distance to the population centers and river beds. These results coincided with the initial factor classification; therefore, the weights were coherent and consistent. The methodology followed in the assessment was validated with these results, clarifying the relation between the input criteria and the final model. In both provinces similar MCE results were obtained, which could be interpreted as meaning that the established procedure was reproducible in other study areas, with similar purposes. Thus, to assess the robustness of the results a GSA should be performed to examine the effects that a change on the input might have on the model results.
3.3. Optimization of Resources Based on the GSA
The GSA procedure implemented in the present study determined that the factors that most influenced the final model were the three mentioned above. Additionally, total indices identified unessential variables for model simplification, detecting those that were not important singularly or in combination with others. Thus, for the optimization of resources, a simplification in the number of factors was possible. Following the procedure implemented for other authors [
52], which was summarized in
Figure 3, the analysis was reproduced only with those three factors. The weights were redistributed and the same restrictions were considered. As a result, similar suitability maps were obtained for both locations, with a maximum variation in the suitability value of 29% (variation of 74 points in the scale of suitability over 247 of the maximum suitability value for Bizkaia). The range of variations produced between the first model and the optimized model was shown in
Figure 6.
As could be noted in the maps, the main variations were associated with areas where the slopes and orientation (
Figure 4) were responsible for most of the suitability value. Málaga presented a higher variation in the southwest as a result of the influence of temperature and precipitation factors in the first model and not in the second. At a local scale (
Figure 8c,f) some patterns were discerned regarding the variation between models, for example, higher variations in the slopes that face south. Those parcels near river beds and populated areas were more affected regarding variations within models in Bizkaia (
Figure 8b,c) in comparison with Málaga (
Figure 8e,f); this is probably due to the different patterns of distribution of houses in the countryside (Bizkaia characterized by scattered single households and Málaga by villages). Once the main characteristics of the variations were analyzed, a specific study was implemented to relate both the most suitable areas from the first model and the higher fluctuation of the suitability values among models. The 10% of most suitable parcels in both provinces obtained the same suitability value with the first and the optimized model (
Figure 9). Thus, the best locations for METland were not deeply influenced by the factors dismissed with the GSA. Indeed, we have produced a map that highlights the 10% of the most suitable parcels for METland (
Figure 8). In the detail view, it could be noted that there is no overlap between these parcels and the areas with higher variability among the models (the original with seven factors versus the optimized with three factors). Thus, the most suitable parcels for METland are not affected by the reduction of factors in the second model. Once more, the disparity in the demographic distribution among provinces could be noted, Bizkaia presents a typical construction of single households disseminated through the area and Málaga shows small communities forming villages or small towns.
Regarding the similarities between the provinces, it could be acknowledged that the 10% of the most suitable parcels did not fluctuate in suitability value, but in the 20% of parcels such value did vary, with 28% in Bizkaia and 5% in Málaga. Our studies revealed that the maximum variation between the two models was 29% of the maximum suitability value.
From the literature review, only one of the studies performed a simplification in the model by the results obtained with the GSA. Gómez-Delgado and Tarantola (2006) [
52] executed the GIS-MCE model twice, first with the 11 factors and their respective weights and second with the three main criteria, observing that the best suitability distribution was not substantially modified. Those results were consistent with the conclusions obtained in the present study. Therefore, a simplification in the model could be achieved without affecting the results of the suitability map, at least in the most optimal location for METland. Finally, GSA could be addressed for identifying the most influential variables in other natural-based solutions.