Application of evolutionary algorithms to develop a rule set for assessing the rehabilitation status of asbestos mines in South Africa J. J. Bezuidenhout, D. Liebenberg, S. Claassens & L. Van Rensburg
Environmental Earth Sciences ISSN 1866-6280 Volume 70 Number 7 Environ Earth Sci (2013) 70:3267-3275 DOI 10.1007/s12665-013-2391-2
1 23
Your article is protected by copyright and all rights are held exclusively by SpringerVerlag Berlin Heidelberg. This e-offprint is for personal use only and shall not be selfarchived in electronic repositories. If you wish to self-archive your article, please use the accepted manuscript version for posting on your own website. You may further deposit the accepted manuscript version in any repository, provided it is only made publicly available 12 months after official publication or later and provided acknowledgement is given to the original source of publication and a link is inserted to the published article on Springer's website. The link must be accompanied by the following text: "The final publication is available at link.springer.com”.
1 23
Author's personal copy Environ Earth Sci (2013) 70:3267–3275 DOI 10.1007/s12665-013-2391-2
ORIGINAL ARTICLE
Application of evolutionary algorithms to develop a rule set for assessing the rehabilitation status of asbestos mines in South Africa J. J. Bezuidenhout • D. Liebenberg S. Claassens • L. Van Rensburg
•
Received: 28 March 2012 / Accepted: 5 March 2013 / Published online: 21 March 2013 Springer-Verlag Berlin Heidelberg 2013
Abstract Asbestos mining has left a legacy of pollution in former mining areas that continues to negatively affect both the environment and local communities. In 2007, the Rehabilitation Prioritisation Index was developed as a scientific tool to indicate the preferred sequence for mine site rehabilitation and served as a departure point for the present investigation in which a database for the rehabilitation success of asbestos sites was developed. Broadbased quantitative and qualitative data, typically used for monitoring rehabilitation success, including amongst others, soil cover depth, physical and chemical soil properties, microbial activity, vegetation properties and small mammal abundance were analysed using multivariate statistics, specifically a redundancy analysis. The most representative model was subsequently selected for the classification of the rehabilitated sites. The multivariate analysis revealed those factors typically associated with rehabilitation success or failure, as well as essentials to be addressed. The feasibility of development of a rule set for rehabilitated site classification was firstly investigated using neural networks which also assisted in the selection of significant parameters. Results from the neural network approach were then used to guide parameter selection for the evolutionary algorithm software. The coordinate scores for the first two axes of the redundancy analysis served as targets for the evolutionary algorithms. Overall, a targeting match of 71 % for the first axis coordinates and 38 % for the second axis coordinates were obtained. Contributing parameters
J. J. Bezuidenhout (&) D. Liebenberg S. Claassens L. Van Rensburg Unit for Environmental Sciences and Management, North-West University, Potchefstroom Campus, Private Bag X6001, Potchefstroom 2520, South Africa e-mail:
[email protected]
for the rule set included: Cl, K, pH, percentage organic carbon, Zn, NH4 and SO4 content of the sites. Keywords Asbestos rehabilitation Evolutionary algorithms Neural network approach
Introduction Derived from the Greek for inextinguishable flame, asbestos is one of the oldest and most widely used minerals known to mankind and was referred to as the miracle mineral by the Greeks. The widespread use of asbestos, however, turned out to be one of the most controversial issues surrounding the industrial mineral industry (Hart 1988; Virta 2003). Factors contributing to the controversy surrounding asbestos include its carcinogenic nature, an overall lack of knowledge of minimum safe exposure levels, its widespread use for more than 100 years and the long latency for the development of lung cancer and mesothelioma (Virta 2003). Mining of asbestos generates vast amounts of residue material, which is chemically not that different from the original rock, however, the fineness of asbestos is problematic as it renders it more susceptible to weathering. The residue dumps are also unsightly and subject to wind erosion. Revegetation of the dumps is not only aesthetically desirable but is also a means of stabilising material, which, if airborne, is a potentially serious health hazard (Meyer 1980). Factors contributing to problems associated with vegetation establishment on asbestos tailings include extremely alkaline conditions, low nutrient concentrations such as P, K and Ca, and surface crusting (Hossner and Hons 1992). Asbestos have been implicated in three major diseases namely asbestosis, lung cancer and mesothelioma. All
123
Author's personal copy 3268
types of asbestos are known to cause asbestosis, other pleural disorders and cancer. Asbestosis is an occupational disease confined to the workplace caused by the inhalation of asbestos fibres and is a non-malignant lung disease associated with exposure to amphiboles. As the disease develops, it may result in a crippling fibrosis of lung tissue, leading to loss of lung elasticity and a reduction in lung function (McCulloch 2003). This disease is characterised by a mixture of symptoms which are associated with the scarring of lungs and general fibrosis, which will cause the victim to suffer from progressive shortness of breath. The disease can be fatal and will not be diminished by removing the individual from the hazardous environment in which the disorder was contracted. In contrast, mesothelioma can result from trivial exposure. This implies that the risk of injury crosses the boundary that usually distinguishes occupational from environmental hazards (McCulloch 2006) is mainly associated with crocidolite exposure, with amosite regarded as being less potent. Mesothelioma is usually fatal and is a primary cancer of the lining of the lung or the abdominal cavity. This inoperable malignancy of the lung lining is characterised by progressive pain and shortness of breath. Pleural effusion is another disease caused by asbestos that can be described as the accumulation of fluid between the layers of the membrane lining the lung and the chest cavity (Nel 2006). Asbestos fields occur in several provinces throughout South Africa. Crocidolite occurs mainly in the North-West Province and the Northern Cape Province. The crocidolite fields of the Northern Cape stretch over 450 km from just south of Prieska on the Orange River to the Botswana border (Hart 1988). Crocidolite occurs in cross-fibre seams in the banded ironstones of the Asbestos Hills Formation of the Griquatown Group that range in thickness from less than 1 mm to about 50 mm. The maximum fibre length is about 150 mm (Howlling 1937; Hart 1988). Amosite is found almost exclusively in the Mpumalanga and Limpopo Provinces. The amosite field occupies portions of the Polokwane (Pietersburg) and Letaba District, and extends from Chuniespoort in the west to the Steelpoort River in the east, a distance of some 90 km. In this region, the asbestos is confined to the banded ironstone of the Penge Formation of the Chuniespoort Group (Hart 1988). There are many deposits of chrysotile in the Limpopo and KwaZulu Natal provinces. The most important chrysotile deposits are those located in the Barberton area, where the chrysotile bodies are hosted in ultramafic intrusions within the Swartkoppies Formation, which forms part of the Onverwacht Group of rocks (Hart 1988; McCulloch 2003). Mining of the asbestos resulted in vast amounts of residue material dumps, which despite objectionable aesthetics, is also subject to wind erosion (Van der Walt and de Klerk 2009) and pose unique rehabilitation challenges (Van
123
Environ Earth Sci (2013) 70:3267–3275
Rensburg and Pistorius 1998). From a rehabilitation perspective, revegetation of these discard dumps addresses the aesthetic concerns, but also serves as a means to stabilise the material and thus avert the associated serious health and environmental hazards (Meyer 1980). Besides aesthetic aspects of asbestos mining and rehabilitation, it represents an ecological risk as the fibres may contribute to atmospheric and water pollution at such sites. Ultimately this may have health implications for nearby human settlements (Emmanouil et al. 2009). Problems associated with asbestos is perhaps well publicised for South Africa, but also hold true for various sites across the globe, e.g. Greece (Emmanouil et al. 2009), Italy (Giacomini et al. 2010), and the USA (Langer et al. 2011), to name but a few. A common theme across the mentioned and other related studies is the importance of the rehabilitation of these sites to prevent impacts to the surrounding environment. Alongside the importance of rehabilitation of mining sites (asbestos and other mining activities), assessment of the rehabilitation status or success thereof plays a critical role in management decisions regarding these sites. In general, rehabilitation assessment evaluates specific risks associated with the particular mining activity and as such criteria used in these assessments differ from one mining activity to another (Hancock et al. 2006; Neri and Sa´nchez 2010). In this regard, the generation of an index value to score the rehabilitation status of a particular site was found to be particularly useful (Thompson et al. 2008; Neri and Sa´nchez 2010). Development of such indices may alleviate the problem of unclear and unformulated criteria for the evaluation of mine closures, possibly preventing the issue of the wrongful closure of mines as well as serving as a tool for rehabilitation efforts by mine groups to benchmark their progress against (Fourie and Brent 2006). Asbestos fields occur in several provinces throughout South Africa and continue to affect former mining areas and surrounding land as a source of pollution. As part of an ongoing effort to rehabilitate asbestos discard dumps and to negate the associated pollution problems, a Rehabilitation Prioritisation Index (RPI) was developed in 2007 (Van Rensburg et al. 2008). The RPI provides a scientifically based method to indicate the order of rehabilitation of specific asbestos sites by evaluating and quantifying the risk associated with pollution sites. This index was implemented by the South African Department of Minerals and Energy as part of an integrated approach towards asbestos rehabilitation (Van Rensburg et al. 2008). To ensure relevant and accurate risk assessments and subsequently the long-term success of rehabilitation, the sustainability of the rehabilitation measures applied should be ascertained. This requires the regular revisiting of information used to develop an index such as the RPI. It also necessitates the development of a comprehensive database
Author's personal copy Environ Earth Sci (2013) 70:3267–3275
established by means of continuous monitoring of relevant qualitative and quantitative properties of rehabilitated or partially rehabilitated asbestos sites. In this regard, the Rehabilitation Monitoring Index (RMI) was developed as the next phase of the asbestos rehabilitation effort (Liebenberg et al. 2012). A multidisciplinary approach was applied to facilitate the development of the RMI, whereby qualitative and quantitative parameters of asbestos sites in various stages of rehabilitation were assessed. The parameters assessed were chosen for their relevance to rehabilitation success and/or for being representative of specific ecosystem functions critical to sustainable rehabilitation. For example, dehydrogenase activity was assayed as an indicator of microbial activity, which is indicative of microbial community function in soil. In turn, microbial community function is considered critical to soil health and the maintenance of soil ecosystem processes (Tate and Rogers 2002). Multivariate analysis of the resulting data allowed for the identification and grouping of parameters into the following descriptors: (i) success parameters, (ii) essentials to be addressed, (iii) non-distinguishable parameters and (iv) reasons for failure. A description of the identified parameters and their relevance is summarised in Table 1. The ordination obtained was representative for the rehabilitation states of the various asbestos sites assessed (Liebenberg et al. 2012). However, this does not provide a model or equation whereby future classification of new sites could be performed. Furthermore, the addition of new data to the established dataset could modify the ordination and consequently the classifications obtained for sites previously assessed. The present investigation overcomes these obstacles by the application of artificial neural networks (ANNs) and evolutionary algorithms. To date, ANNs and hybrid evolutionary algorithms have found various applications due to their ability to map the non-linear relationships between the variables and the fact that no assumptions about the model in terms of mathematical relationships or distribution of the data are required (Wilson and Recknagel 2001). Furthermore, hybrid evolutionary algorithms can assist in the development of generic predictive rule sets from ecological data (Welk et al. 2008). Due to the fact that hybrid models can allow the simulation of the various features of complex systems, it is finding increased application in important decision making tools (Parrot 2011). The aim of this study was to apply ANNs and evolutionary algorithms to generate a predictive rule set with which rehabilitated or partially rehabilitated asbestos sites can be ranked into the following specified descriptors: (i) success parameters, (ii) essentials to be addressed, (iii) non-distinguishable parameters and (iv) reasons for failure.
3269
Materials and methods The dataset from the RMI study (Liebenberg et al. 2012) is an extensive one and was therefore suitable to serve as the basis for the application of the ANN and evolutionary algorithms of the present investigation. The study was conducted on selected asbestos mines from the Limpopo, North-West and Northern Cape Provinces of South Africa. The status of the asbestos mines can be divided into three classes: (i) rehabilitated, (ii) partially rehabilitated and (iii) no rehabilitation carried out. Sites monitored in the RMI study included asbestos mines from all three of these classes. In the Limpopo Province, 12 mines which included 52 discard dumps were monitored. Seven mines were monitored in the North-West Province, which included 30 discard dumps. In the Northern Cape Province, 7 mines were monitored which included 34 discard dumps. Quantitative and qualitative parameters (Table 2) were assessed at each of the asbestos discard dumps Multivariate statistical analysis, specifically redundancy analysis (RDA), was performed on all quantitative data obtained during the RMI investigation (Liebenberg et al. 2012) to determine the most prominent parameters influencing the progress of rehabilitation at the various sites (Canoco for Windows 4.5) (Ter Braak and Smilauer 1998). During the generation of the RDA, the software package generates a solution file containing the coordinate scores for the plotting of the sites within the ordination space as well as a correlation matrix of the input parameters and ordination axes. In the current investigation, the coordinate scores served as targets for the neural network software and evolutionary algorithm software. For the neural network analysis, Forecaster XL (Alyuda Research LLC) was used to evaluate the feasibility of applying scoring targeting to generate a predictive rule set for the classification of the rehabilitation sites. Though the results from the neural networks seemed quite promising, neural networks have the limitation that it requires specialist software and no rule set or equation is generated to provide insight into how the results were obtained (Bezuidenhout et al. 2008). This is an obstacle that can be overcome by the application of evolutionary algorithms. Even though modelling of the dataset by an ANN prior to using EA is strictly not required, we have found that using the ANN and EA approaches in series does have advantages. During the ANN step a good insight can be gained as to the possible success of continuing to the EA step and it allows for the rapid evaluation of dataset combinations. Upon evaluation of the results from the ANN step, the process may either be aborted, or if successful, refinement of the model and generation of a predictive rule set can be pursued in the EA step.
123
Author's personal copy 3270
Environ Earth Sci (2013) 70:3267–3275
Table 1 Description of different parameters influencing rehabilitation Quantitative data
Reasons why parameter is of significance for rehabilitation
References
Essentials to be addressed Mg
Mg is a metallic constituent of chrysotile and amosite. Nutrient imbalances may have been present because Mg occurred in extremely high concentrations compared to Ca and K. Low Ca:Mg and K:Mg ratios may restrict plant growth. Ca added to the tailings in solution will be rapidly replaced by Mg; the replacement of K is less effective. There are three ways in which Ca and/or Mg levels can affect vegetation: (a) deficiency of Ca; (b) toxicity of Mg; or (c) unfavourable Ca:Mg ratio. An inverse relationship exists among Mg, Al, B, Co, Mn, P and Na. It was concluded that Mg uptake might have an antagonistic effect and reduce the uptake of other important minerals. Ni toxicity is reduced in the presence of plant nutrients such as N, K and Mg
Hossner and Hons (1992), Proctor and Woodell (1975), Moore and Zimmerman (1977), Van Rensburg and Pistorius (1998)
NO3
Tailings are almost universally deficient in N. N is absorbed by plants in both NO3 and NH4 forms. The application of NO3 containing fertilisers can stimulate the accumulation of macro-elements particularly Ca and therefore could assist in alleviating Ca deficiency in plants where its uptake is depressed by Mg excess
Hossner and Hons (1992), Van Rensburg et al. (2004), DeGrood et al. (2005) and Ellerly and Walker (1986)
pH
Addition of gypsum/lime to lower pH. Increasing Ca reduces toxicity of heavy metals
Proctor and Woodell (1975)
Cl
Due to both the high Mg and Cl contents in the soil, there is a probability that it could form MgCl2, which if absorbed by plants, leads to chlorosis as well as the rapid development of brown necrotic lesions, because the metabolism of the plant is disturbed. High Cl content could be responsible for the low P concentrations in plant tissue because of Cl accumulation
Van Rensburg and Pistorius (1998)
K
High Mg concentrations are often a cause of poor K status and inhibit plant growth when accompanied by low K levels. K has been reported to reduce the toxicity of Ni
Proctor and Woodell (1975), DeGrood et al. (2005) and Van Rensburg and Pistorius (1998)
Success parameters NH4
Tailings are almost universally deficient in N. N is absorbed by plants in both NO3 and NH4 forms
Hossner and Hons (1992), Van Rensburg et al. (2004), DeGrood et al. (2005) and Ellerly and Walker (1986)
Cu
Heavy metals reduce enzymatic activity and the microbial and microfauna populations in soils. Higher stability of Cu complexes favours the leaching of Ni. Particularly elevated concentrations of Ni, Cr and Co are known to occur. The interaction with other heavy metals and micro-elements such as Fe, the influence of soil pH on heavy metal solubility and organic matter content makes Ni availability a complex process. One of the primary factors affecting Ni uptake in plants is the pH. By increasing the soil pH with liming, a significant reduction of Ni occurs. pH also influences the precipitation of Ni with other compounds, such as phosphates. High Mg concentrations could ameliorate the toxic effect of high Ni availability
Hossner and Hons (1992), Proctor and Woodell (1975) and Van Rensburg and Pistorius (1998)
Zn
High pH leads to Zn and Fe deficiency
Thorne (1957)
%C
Organic matter is critical to bind excessive micro-elements. Increases water-holding capacity, aggregation of soil and nutrient status. Organic matter improves the physical nature of the rooting medium by:
Van Wyk (1994), Pascual et al. (2000), Tordoff et al. (2000) and Van Rensburg and Pistorius (1998)
• increasing water and nutrient holding capacity, • provision of plant nutrients in a slow-release form, facilitating vegetation establishment, • complexing of heavy metals, thereby reducing phytotoxicity. The presence of organic matter in soils can also reduce Ni toxicity by removing this metal as a chelate complex
123
Author's personal copy Environ Earth Sci (2013) 70:3267–3275
3271
Table 1 continued Quantitative data Microbial activity
Reasons why parameter is of significance for rehabilitation
References
Dehydrogenase activity depends on the metabolic state of soil microorganisms and it is used as an indicator of microbial activity in soils. In areas where soil conditions are marginal, the function of soil microbes is critical for supporting plant growth and revegetation success. Soil microbes also play a fundamental role in the establishment of biogeochemical cycles and are involved in the formation of the soil structure
Ros et al. (2003), Pascual et al. (2000) and DeGrood et al. (2005)
Reasons for failure Fe
Main component of amosite and crocidolite. Ni toxicity involves an induced Fe deficiency; however, due to Fe being a main component of amosite and crocidolite, tailings would be enriched in Fe and this together with the addition of organic matter makes Ni availability a complex process
Mn
Intensification of Ni toxicity symptoms in plants caused by high levels of Mn. Ni in soils with high levels of soluble iron may be less toxic and in soils with high levels of soluble Mn more toxic. Mn functions as an enzyme activator in biological systems and is not very soluble under alkaline conditions. Mn precipitates as a hydroxide at high pH values. High Mg concentrations in the soils can reduce Mn uptake owing to the competition between the ions of these metals B Low Ca may exacerbate the phytotoxicity of boron and may cause the build-up of iron in the plant biomass to phytotoxic concentrations. The uptake of K ions is many times higher in plants with a good B status. Conversely, high K availability can intensify even if B is deficient, unlike Ca it cannot reduce B toxicity, seeming rather to intensify it. There is a positive correlation between B uptake and both Mg and K content in grasses. B has been found to be deficient in most asbestos-rich soils Non-distinguishable entities
Van Rensburg et al. (2004), Proctor and Woodell (1975) and DeGrood et al. (2005)
Proctor and Woodell (1975) and Van Rensburg and Pistorius (1998)
Van Rensburg et al. (2004) and Van Rensburg and Pistorius (1998)
Total P
Acidification may improve the solubility and availability of P and other plant nutrients that are in short supply, but the benefit may be offset by increased heavy metal toxicity. Phosphorus deficiencies are common. Iron which is abundant combines with P to form unavailable and insoluble phosphates. Similarly P interactions with trace metals are also a plausible explanation for deficiencies in available P
Ellerly and Walker (1986) and Van Rensburg and Pistorius (1998)
PO4
Soluble or available phosphorus. Mg improves the uptake of phosphorus in plants
Berger (1968)
Vegetation density
Where the re-establishment of natural plant communities is the objective, success is recorded on the basis of both the cover vegetation achieved and its composition
Coaltech (2007)
Basal cover
Basal cover is the measure of the proportion of ground, at root level that is covered by vegetation and more specifically, by the rooting portion of the cover plants
Coaltech (2007)
Woody crown height
Important for the rain drop erosion effect. The foliage of plants breaks raindrops into smaller particles which decreases the effect of erosion
Redco (2008)
The evolutionary algorithm software Multimodelling, a hybrid evolutionary algorithm for ecological informatics, was used to generate a rule set for the classification of the datasets into the identified descriptors. Targeting for the X-axis coordinates and Y-axis coordinates was done separately. Combinations of the full dataset and a reduced
dataset from the RMI study (Liebenberg et al. 2012) were also evaluated as input for the software. From the results, a rule set was obtained that is easily adaptable to spreadsheet software. For each dataset combination, the software was run at least 10 times and the generated models evaluated until the best model was selected.
123
Author's personal copy 3272
Environ Earth Sci (2013) 70:3267–3275
Table 2 Quantitative and qualitative parameters assessed at each of the asbestos discard dumps Quantitative parameters
Qualitative parameters
Parameter
Description
Parameter
Description
Soil cover depth
Depth of soil layer covering asbestos discard material
Footprint area
The area occupied by each discard dump
Physical and chemical soil properties
Particle size distribution, Ca, Mg, K, Na, P, Cl, NO3, SO4, NH4, CEC, Bsat, EC, pH
Land use
Type of land use, e.g. grazing or no land use
Microbial activity
Assay of dehydrogenase activity as an indication of microbial community function
Erosion
Erosion or flood damage—extent, severity and potential risk
Vegetation properties
Dominant species, plant density, basal cover, dominant growth form, impact by grazers, crown cover and crown height of herbaceous and woody species, respectively
Secondary pollution
Areas where rehabilitation was ineffective and asbestos fibres were again exposed
Small mammal abundance
Species, numbers, sex, mass, lactating, pregnant
Water control structures
Evaluation of gabions, contour walls, retaining walls and speed reducers to determine reasons for success or failure in the reduction of erosion
Air quality
Presence/absence of asbestos fibres through atmospheric sampling
Ca calcium, Mg magnesium, K potassium, Na sodium, P phosphorus, Cl chloride, NO3 nitrate, SO4 sulphate, NH4 ammonium, CEC cation exchange capacity, Bsat base saturation, EC electrical conductivity
Results and discussion The ordination obtained from the RDA of the quantitative data (Fig. 1), was applied to explain the grouping of the different measured parameters in terms of four descriptors: 1.
2.
3.
4.
Success parameters (Class I) parameters that exert the most significant influence on the success of asbestos rehabilitation. If these parameters are not primarily addressed, rehabilitation will fail. Essentials to be addressed (Class IV) parameters that have a high influence on the rehabilitation success of asbestos sites and should be attended to in order to ensure a higher success rate and achieve potentially sustainable rehabilitation. These factors have to be addressed in combination with ‘‘success parameters’’ mentioned above. Reasons for failure (Class II) parameters that have been shown to contribute to the failure of the rehabilitation attempt should they not be addressed. These parameters will not necessarily always present a problem for the rehabilitation attempt. Non-distinguishable entities (Class III) parameters that had no prominent effect on the success of asbestos rehabilitation.
Therefore, the order from the most to least important category is: success parameters [ essentials to be addressed [ reasons for failure [ non-distinguishable entities.
123
Fig. 1 RDA ordination diagram illustrating the quantitative results for all three provinces. Red vectors represent the environmental parameters and blue vectors the vegetation properties and microbial activity. Eigenvalues for the first two axes were 0.052 and 0.021, respectively. DHA dehydrogenase, Dens density, BasalC basal cover, HerCC herbaceous crown cover, HCH herbaceous crown height; WCC woody crown cover; WCH woody crown height
When comparing the forecasted coordinate scores against the actual coordinate scores, a high correlation ([90 %) was observed for the coordinate scores for the X-axis (Fig. 2a).
Author's personal copy Environ Earth Sci (2013) 70:3267–3275
3273
Fig. 2 a A regression analysis between actual and forecasted data to obtain coordinate scores for the first axis (X) used as targets for the ANN. b Contribution of various parameters to the coordinate score. Correlation and fit for the ANN was [90 %
Fig. 3 a A regression analysis between actual and forecasted data to obtain coordinate scores for the second axis (Y) used as targets for the ANN. b Contribution of various parameters to the coordinate score. Correlation and fit for the ANN was [90 %
Furthermore, for the X-axis coordinate scores, parameters such as dehydrogenase activity (DHA), woody crown cover (WCC), soil cover depth (Depth) and pH contributed more than 10 % to the generated model (Fig. 2b). When comparing the coordinate scoring results for the Y-axis (Fig. 3a, b), a high correlation ([90 %) was obtained. For the Y-axis coordinate scores, the following parameters contributed more than 10 % to the model: soil cover depth (Depth) and organic carbon (C). However, despite the good coordinate scoring results, the inherent limitation of the ANN approach that includes the inability to generate a formula or rule set as well as the fact that this approach require specialised software, limits its potential applicability. The most applicable rule set for targeting the X-axis and Y-axis coordinates, obtained from the results of the evolutionary algorithm, will subsequently be presented and discussed. The following rule set was generated for the X-axis coordinate scores: IF (Cl C 0.226) THEN f(x) = (((K ? (K ? pH)) - C)/(((Zn/5.312)/ NH4) - 6.085)) ELSE f(x) = (((K ? (K ? pH)) - (C ? (C ? 5.779)))/(((Zn/ 4.494) ? SO4) - 2.086))
The following rule set was generated for the Y-axis coordinate scores: IF ((NO3 C 1.743)OR(Fe C 11.783)) THEN f(x) = (ln(|HCH|) ? (Fe/((K ? 10.388) ? (1.290/ C)))) ELSE f(x) = (ln(|SO4|)*K) A summary of the result obtained for the rule sets is presented in Fig. 4a and b. For the X-axis coordinates, an R2 value of 0.710 was obtained, corresponding to a 71.0 % accuracy, while for the Y-axis an R2 value of 0.376 was obtained, corresponding to a 37.6 % accuracy. It is important to remember that during redundancy analysis, each successive axis captures less of the variation in the data than the previous axis. Also, the ordination graphs represent a two-dimensional representation of the multidimensional model, which will make exact targeting of the values not possible. When examining the rule sets themselves, it is clear that for the X-axis, parameters such as Cl, K, organic C, pH, Zn, NH4 and SO4 play an important role in the rule set. In this regard, it is interesting to note that despite the fact these factors could exert direct positive (K, NH4, SO4 and Zn) or negative effects (Cl) on vegetation from a plant nutritional perspective (Bergmann 1992), the potential growth medium acidification impact of the NH4, SO4 and organic C
123
Author's personal copy 3274
Environ Earth Sci (2013) 70:3267–3275
Fig. 4 a (left), b (right) Scatterplot comparisons for the X-axis and Y-axis coordinates. AX1 and AX2 (Y) represents the coordinate scores that served as targeting values for the evolutionary algorithm for the X-axis and Y-axis coordinates, respectively, CalcAX1 and CalcAX2 (Y) represents values calculated from the rule set for the respective axes
cannot be ignored. For the Y-axis, parameters such as NO3, Fe, herbaceous crown cover, K, C and SO4 were observed to be determinant for the rule set. Similarly, apart from their obvious potential effect on plant nutrition, the presence of NO3 and Fe (and absence of organic C and SO4) is typically associated with more alkaline growth medium conditions. This pH associated effect in the growth medium correlates well with the data provided in Fig. 1, i.e. the ‘‘success parameters’’ being associated with the presence of organic C and potentially acidifying NH4, whilst the Cl, SO4 and NO3, amongst others, grouped under the ‘‘essentials to be addressed’’. When classifying the sites into a Class system and comparing classification between the ordination data and the rule set data, it was apparent that from the 237 cases in the dataset, 137 (57.8 %) matched the original classification, whilst a classification mismatch occurred in 100 (42.2 %) of the cases. When examining the classification mismatches, it became clear that, in most cases, the reclassification seemed to involve borderline cases in the dataset. In about 24 cases, a worse classification than the initial classification was obtained, for example success to failure. There were about 29 cases where a better classification was obtained, i.e. a reclassification for failure to success, non-distinguishable entities to success, and failure to success to name a few examples. There were 11 cautions, i.e. reclassification from success to essentials to be addressed, and about 36 neutral reclassifications, i.e. from another class to essentials to be addressed. Conclusions The study was indeed successful in the development of a rule set to allow for the classification of rehabilitation sites within an existing representative model. The rule sets developed were able to classify approximately 60 % of the sites into their proper categories. Also, sites at the rule boundaries of the various categories can be subjected to closer inspection,
123
particularly in cases where a site appears to be transitioning from a successful category to a category pointing to possible problems requiring intervention. Furthermore, the approach of using ANNs as a rapid evaluation tool for the dataset of subsets of the dataset also holds great potential. Also the major parameters in the generated rule sets can be linked to existing literature concerning plant nutrition and soils. The application of the rehabilitation status index for asbestos mines will go a long way to negate unclear and unformulated criteria for the evaluation of asbestos rehabilitation sites in South Africa in particular. By enabling a measureable benchmark for the rehabilitation status of asbestos mines, premature closure before successful rehabilitation can be prevented. The rehabilitation index can serve as a tool for the companies involved to measure their progress against, and if required, perform suitable interventions to ensure success of their rehabilitation efforts.
References Berger KC (1968) Introductory soils. Macmillan, New York Bergmann W (1992) Nutritional disorders of plants. development, visual and analytical diagnosis. Gustav Fischer Verlag, Jena Bezuidenhout JJ, Du Plessis S, Van Ginkel C, Van Rensburg L (2008) Applicability of Ann models as a possible early warning tool for algal bloom prediction. In: Proceedings of Conference on Environmental Science and Technology. Houston Coaltech (2007) Guidelines for the Rehabilitation of Mined Land. Chamber of Mines of South Africa DeGrood SH, Claassen VP, Scow KM (2005) Microbial community composition on native and drastically disturbed serpentine soils. Soil Biol Biochem 37:1427–1435 Ellerly KS, Walker BH (1986) Growth characteristics of selected plant species on asbestos tailings from Msauli Mine, eastern Transvaal. S Afr J Bot 52(3):201–206 Emmanouil K, Kalliopi A, Dimitrios K, Evangelos A (2009) Asbestos pollution in an inactive mine: determination of asbestos fibers in the deposit tailings and water. J Hazard Mater 167:1080–1088 Fourie A, Brent AC (2006) A project-based mine closure model (MCM) for sustainable life cycle management. J Clean Prod 14:1085–1095
Author's personal copy Environ Earth Sci (2013) 70:3267–3275 Giacomini F, Boerio V, Simona Polattini S, Tiepolo M, Tribuzio R, Zanetti A (2010) Evaluating asbestos fibre concentration in metaophiolites: a case study from the Voltri Massif and Sestri, Voltaggio Zone (Liguria, NW Italy). Environ Earth Sci 61:1621–1639 Hancock GR, Grabham MK, Martin P, Evans KG, Bollho¨ffer A (2006) A methodology for the assessment of rehabilitation success of post mining landscapes—sediment and radionuclide transport at the former Nabarlek uranium mine, Northern Territory, Australia. Sci Total Environ 354:103–116 Hart HP (1988) Asbestos in South Africa. J S Afr Inst Min Metall 88(6):185–198 Hossner LR, Hons FM (1992) Reclamation of Mine Tailings. In: Lal R, Stewart BA (eds) Soil Restoration. Advances in Soil Science, vol 17. Springer-Verlag, New York Howlling GE (1937) Asbestos. Imperial Institute, London Langer WH, Van Gosen BS, Meeker GP, Adams DT, Hoefen TM (2011) The dispersion of fibrous amphiboles by glacial processes in the area surrounding Libby, Montana, USA. Environ Earth Sci 64:157–168 Liebenberg D, Claassens S, Van Rensburg L (2012) A multidisciplinary approach for the assessment of rehabilitation at asbestos mines in South Africa. Environ Earth Sci 67(4):1237–1244 McCulloch J (2003) Asbestos mining In Southern Africa, 1893–2002. Int J Occup Environ Health 9:230–235 McCulloch J (2006) Saving the asbestos industry, 1960 to 2006. Public Health reports (974) 121(5):609–614 Meyer DR (1980) Nutritional problems associated with the establishments of vegetation on tailings from asbestos mines. Environmental pollution. Series A. Ecol biol 23(4):287–298 Moore TR, Zimmerman RC (1977) Establishment of vegetation on serpentine asbestos mine wastes, southeastern Quebec. Canada J Appl Ecol 14:589–599 Nel J (2006) The current risk of asbestos exposure to the citizens of the Prieska area, Northern Cape Province in South Africa. (Minidissertation). North-West University, Potchefstroom Neri AC, Sa´nchez LE (2010) A procedure to evaluate environmental rehabilitation in limestone quarries. J Environ Manage 91:2225– 2237 Parrot L (2011) Hybrid modelling of complex ecological systems for decision support: recent successes and future perspectives. Ecol Info 6:44–49 Pascual JA, Garcia C, Hernandez T, Moreno JL, Ros M (2000) Soil microbial activity as a biomarker of degradation and remediation processes. Soil Biol Biochem 32:1877–1883 Proctor J, Woodell ARJ (1975) The ecology of serpentine soils. In: Macfadyen A (ed) Advances in Ecological Research, vol 9. Academic Press, London, pp 256–385 Redco/Nemai Consortium (2008) Monitoring of previously rehabilitated asbestos sites in South Africa. Report prepared for the Department of Minerals and Energy. South Africa
3275 Ros M, Hernandez MT, Gercia C (2003) Soil microbial activity after restoration of a semiarid soil by organic amendments. Soil Biol Biochem 35:463–469 Tate RL, Rogers BF (2002) Soil ecosystem properties, microbial diversity, and ecosystem assessments. In: Violante A, Huang PM, Bollag J-M, Gianfreda L (eds) Developments in soil science, vol 28B. Elsevier, New York, pp 79–93 Ter Braak CF and Smilauer P (1998) Canoco reference manual for user’s guide to Canoco community ordination, ver. 4. Microcompuer Power, New York Thompson SA, Thompson GG, Withers PC (2008) Rehabilitation index for evaluating restoration of ecosystems using the reptile assemblage as the bio-indicator. Ecol Ind 8:530–549 Thorne W (1957) Zinc deficiency and its control. In: Norman AG (ed) Advances in agronomy. Academic Press Inc., New York, pp 31–66 Tordoff GM, Baker AJM, Willis AJ (2000) Current approaches to the revegetation and reclamation of metalliferous mine wastes. Chemosphere 41:219–228 Van der Walt IJ, De Klerk TC (2009) A model to determine nonoccupational human exposure to crocidolite asbestos fibers in the Northern Cape. South Africa. Phys Geogr 30(1):79–87 Van Rensburg L, Pistorius L (1998) An investigation into the problems associated with revegetating chrysotile tailings. S Afr J Plant Soil 15(4):130–140 Van Rensburg L, Maboeta MS, Morgental TL (2004) Rehabilitation of co-disposed diamond tailings: growth medium rectification procedures and indigenous grass establishment. Water Air Soil Pollut 154:101–113 Van Rensburg L, Claassens S, Bezuidenhout JJ, Jansen van Rensburg PJ (2008) Rehabilitation of asbestos mining waste: a Rehabilitation Prioritisation Index (RPI) for South Africa. Environ Geol 57(2):267–273 Van Wyk S (1994) ‘n Strategie vir die rehabilitasie van versteurde mynbougebiede in suidelike Afrika. (Proefskrif—PhD Omgewings wetenskappe). Potchefstroom Universiteit vir Christelike Hoer Onderwys. Suid-Afrika Virta RL (2003) Worldwide Asbestos Supply and Consumption trends from 1900 through 2003. US Geological Survey Circular 1298. http://pubs.ugs.gov/circ/2006/1298/. Accessed Feb 2009 Welk A, Recknagel F, Cao H, Chan W, Talib A (2008) Rule-based agents for forecasting algal population dynamics in freshwater lakes discovered by hybrid evolutionary algorithms. Ecol Info 3:46–54 Wilson H, Recknagel F (2001) Towards a generic artificial neural network model for dynamic predictions of algal abundance in freshwater lakes. Ecol Model 146:69–84
123
