Monitoring of mine tailings using satellite and lidar data

Transcription

1 Surveying Monitoring of mine tailings using satellite and lidar data by Prevlan Chetty, Southern Mapping Geospatial This study looks into the use of high resolution satellite imagery from RapidEye and WorldView 2 sensor systems in conjunction with high resolution lidar data for mine tailings pond rehabilitation monitoring in South Africa. A tailings pond is an area of refused mining tailings where the water borne material is pumped into a pond to allow for the sedimentation of the solid particles. The dangers associated with tailing ponds include dam failure and acid mine drainage from seepages. This leads to the need for continuous monitoring of the water passing through the pond and its immediate surroundings. Remote sensing provides cost effective alternatives for the accurate monitoring of mine tailings and allows for the detection of seepages. Additionally, the rates of sedimentation within the pond can be monitored. Cost effective satellite imagery was used for the monitoring of the mine tailings and surrounding vegetation. The mine tailings were spectrally classified and mapped, enabling the multi-temporal monitoring of sediment and water movement. The satellite imagery acquired additionally allows for the monitoring of vegetation health around the site by using vegetation indices. dates of the imagery utilised are seen in Table 1. WorldView-2 provides panchromatic imagery of 0,5 m resolution, and eight-band multispectral imagery with 1,8 m resolution. RapidEye provide panchromatic imagery of 5 m ground resolution, and five-band multispectral imagery that includes the Red-Edge band which is sensitive to chlorophyll content. Topographical data Southern Mapping Company conducted a lidar survey over the tailings pond at the area of interest prior to the satellite based project. The survey was conducted on 8 November 2012 with a capture density of 2 points per m² and a vertical accuracy of 8 cm. The lidar data was used to generate a 1 m digital elevation model (DEM). Study area The study area experiences a subtropical climate and is located on the Highveld Plateau of South Africa. Most of the mining activity in this region is centred around platinum on the Merensky Reef which stretches from the west of the Pilanesburg game reserve towards Marikana and Brits in the East. The average days with rainfall per month is shown in Fig. 1, which was used to select the archived satellite imagery. As indicated in Fig. 1 there are significantly fewer rainfall events between April and September, while September to January experiences more rainfall events. It is apparent that the vegetation response to rainfall at the site would be stronger Data sets Satellite imagery 50 cm satellite derived multispectral imagery from the WorldView 2 satellite and 5 m resolution multispectral imagery derived from RapidEye was used to assess the mine tailing. The Sensor Resolution (cm) Acquisition date WorldView /08/15 WorldView /07/21 RapidEye /04/24 RapidEye /12/16 Table 1: Satellite imagery used. Fig. 1: Average rainfall days Rustenburg airfield. 22 PositionIT September 2013

2 in the summer months (October to January) and progressively weaker during the winter months (February to September). Satellite based monitoring indices Vegetation indices Vegetation indices are combinations of surface reflectance at two or more wavelengths in the electromagnetic spectrum that have been designed to highlight a particular property of vegetation. They are derived using the reflectance properties of vegetation. The electromagnetic spectrum is a continuum of electromagnetic waves in which energy is transferred, arranged according to frequency and wavelength. It is measured in microns (µm), which provides a measurement for the length of the waves. Light is a particular type of radiation that can be seen by the human eye. The visible range of the spectrum ranges from approximately 0,4 µm (blue) to 0,7 µm (red) [1]. The shortest waves are gamma waves, while the longest waves are radio waves. Shorter wavelengths generally have a higher energy content while longer wavelengths have a lower energy content [2]. The Normalised Difference Vegetation Index (NDVI) The Normalised Difference Vegetation Index is numerical indicator that operates within the visible and near infrared bands of the electromagnetic spectrum [1]. An NDVI image is an image index that is sensitive to photosynthetic activity on a pixel-by-pixel basis. It is used to assess the rigour of vegetation cover within a target area. Fig. 2 shows how electromagnetic waves have been broken into different categories based on their wavelengths. Remote sensing is able to exploit these different characteristics of the spectrum, by using channels, which relate to a specific waveband or part of the electromagnetic spectrum. As part of the process of photosynthesis, plants absorb incoming radiation from the sun. At the same time, some of the radiation is scattered in the near infrared. A healthy plant differs from a withering plant in the way that it absorbs radiation, and this difference provides a measure of the greenness of the vegetation. The NDVI is therefore an index that is used to measure that Fig. 2: The electromagnetic spectrum [1]. difference. Healthier plants would generally absorb most of the visible light that is radiated towards it, while reflecting a large portion of near infrared light [3]. Unhealthy plants reflect more visible light, and less infrared light. The Enhanced Vegetation Index The Enhanced Vegetation Index (EVI) is an optimised index designed to enhance the vegetation signal with improved sensitivity in high biomass regions and improved vegetation monitoring through a decoupling of the canopy background signal and a reduction in atmosphere influences. An EVI analysis was employed in this study owing to its improved sensitivity and reduction in atmospheric influences. Wetness indices Water has a significantly lower reflectance value than that of soil. Thus, wet soil would have a lower reflectance value than the surrounding soil. The moist soil can be spectrally classified in each image, and a wetness map can be generated. This technique has been used effectively in this study, as this procedure allows for the monitoring of water and sediment fluctuation within the mine tailing. Methodology and results Image processing The imagery acquired was orthorectified using ground control points from the Chief Directorate of National Geo-spatial Information (CD:NGI) orthoimagery. The imagery was then subset and pan-sharpened to cover only the mine dump area. The imagery Display Numbers (DN) values were then converted to Top of Atmosphere (ToA) reflectance values which enabled comparisons to be made between the images. This procedure normalises the effect the atmosphere has on the imagery, because the atmosphere is in a constant state of change. Various indices were then calculated from the imagery which includes the Enhanced Vegetation Index (EVI) and the wetness indices. Stream flow analysis processing The DEM showed that the topography across the site decreases in a northwesterly direction. It is therefore expected that in the event of flooding or seepage, the bulk of the material would accumulate towards the north-western portion of the site. Fig. 3 provides a three-dimensional view of the site, which shows the lowest height values seen towards the north-west section. By overlaying the stream flow analysis and simulating a flooding event, the accumulation location can be verified as shown in Fig. 4. Using Global Mapper s Stream Flow tool, a stream flow was generated that is based in the elevations throughout the surface. These streams were then refined and used to produce a stream flow map. PositionIT September

3 Fig. 3: 3D DEM of mine tailings pond. Fig. 5: Stream flow analysis labelled. Fig. 4: Simulated flood event 3D. Since the most probable location of contamination in the event of the tailings pond failure had been identified, an EVI analysis was conducted along the stream flow lines towards the north-western region of the site. An analysis of the EVI values was conducted along the stream flow lines towards the north-western region of the site. The streams have been labelled for reference purposes and are shown on Fig. 5. EVI processing An EVI is computed following this equation: EEEEEE = GG xx (NNNNNN RRRRRR) (NNNNNN + CC1 xx RRRRRR CC2 xx BBBBBBBB + LL) where NIR/red/blue are atmospherically-corrected or partially corrected (Rayleigh and ozone absorption) surface reflectance (L) is the canopy background adjustment that addresses non-linear, differential NIR and red radiant transfer through a canopy, and C1, C2 are the coefficients of the aerosol resistance term, which uses the blue band to correct for aerosol influences in the red band. A pseudo-colour table was generated for the vegetation indices. EVI output demonstration The following section demonstrates the EVI capabilities in identifying vegetation degradation using examples from the mine tailings pond area. The healthy areas seen in the EVI map in the April 2009 image (Figs. 6 and 7) would remain healthy in the December 2009 image due to the rainfall in the region shown in Fig. 1. The areas that turn unhealthy between the April 2009 and December 2009 image would represent vegetation degradation. Figs. 6 and 7 show cultivated fields next to the mine tailing. The fields circled in black in April 2009 show healthy signs, and are visibly green in the natural colour satellite imagery. The black features seen represent very low EVI responses, which were likely to be water bodies which have a low reflectance. These values were masked out using an automated system. The same fields in December 2009 (Figs. 8 and 9) now show lower EVI values which relates to unhealthy conditions. This is because of the practice of burning cultivation in the post growing season which is common practice in the region. The fields in the natural colour image are visibly browner and show signs of clearing. Towards the central part of the tailings dam, EVI values indicate sediment build-up. Using the reclassification procedure described in the methodology section, and subtracting the two images, the areas that experienced decreases in vegetative health can be identified easily. Reclassification of EVI Values The methodology to detect areas that have been subjected to decreases in vegetation involves the following procedures: Reclassification of the raster based images into new values that are representative of areas with good vegetation response and poor vegetation response across all the EVI images. To do this, a histogram was plotted and the ranges of values that are healthy and unhealthy were assigned to a different class. For example, the range may be representative of healthy response amongst the plants, and can be reclassified as PositionIT September 2013

4 Subtract the pre-growing season image by the post-growing season image, which would reveal the areas that have experienced decreases in vegetative response. As the lidar-derived DEM has shown, it is anticipated that the areas to the north-western part of the mines tailing are the most likely to show signs of seepage in the event of a failure. Thus, the areas that show healthy responses in the image before the growing season here can be monitored to see if they show unhealthy vegetation response in the post-growing season image. This would be represented as a high value in the result of the subtraction between the two images. Figs. 10 and 11 show stream B with the EVI response map for the dates April 2009 and December 2009 respectively. The results show that there are no visible signs of vegetation health deterioration according to the EVI maps produced along stream B. The EVI values are now higher in December 2009 as opposed to April 2009 due to the rainfall the region experiences. The vegetation health has been maintained throughout the image, except for the stream areas which show increases in vegetation health. Where EVI decreases are present, the decreases can be attributed to other factors such as development and farming. The WorldView Imagery provides a similar response, but the differences are more clearly shown in the RapidEye imagery due to the imagery acquisition dates. Wetness index processing The spectral bands and indices were then used in a supervised classification Fig. 6: Orthoimage fields 2009/04/24. Fig. 7: EVI image fields 2009/04/24. Fig. 8: Orthoimage fields 2009/04/24. Fig. 9: EVI image fields 2009/04/ PositionIT September 2013

5 procedure to extract the water, water film, moist soil and dry soil classes. The classified imagery was then vectorised to form polygons of the different classes created during the classification process. Once dissolved (a process whereby the individual polygons with a common feature are joined to form a single bigger polygon), the pond tailings could be examined in greater detail, as area calculations could be performed to monitor movement within the pond itself. Wetness index analysis A wetness index was performed by spectrally separating the water, water film, moist soil and dry soil. The RapidEye imagery based wetness index images are shown in Figs. 12 and 13. In the RapidEye imagery, no apparent water film can be observed in the April image. This could be due to the resolution of the imagery (5 m), as the WorldView 2 imagery shows the water film. The water, moist soil and dry soil classes have been defined accurately. The tailings were vectorised as described in the methodology which then allows for accurate monitoring of the tailings ponds with regards to the water and water film within the pond. Fig. 14 shows the movement of the tailing s water from April 2009 to December The tailings pond s water surface in April 2009 measured approximately 3 km², while in December 2009 the pond measured approximately 15 km². The depth of the water within the tailings pond is unknown, but this can be derived by in field in-situ measurements. While this imagery shows no water movement between the dates August 2010 and July 2011, the water film could be spectrally classified due to the higher resolution of the sensor (50 cm). The spectrally classified imagery was also vectorised, but the imagery showed little water movement within the mine tailings pond. Conclusions The results showed that the extent of the pond's movement from April 2009 to December 2009 can be accurately monitored, with respect to the water, the water film, the moist soil and the dry soil. Acquisitions of data relating to the water depth will further strengthen the monitoring of the water within the mine tailings. It must be noted though that the extraction of all classes is particularly difficult without ground truth. Therefore, the in-field verification of the classes would further strengthen the classification process. The results from the Enhanced Vegetation Index have shown that this approach has the ability to accurately detect areas that are showing signs of decreasing vegetation health. Using the approach outlined in the methodology, the vegetation can be monitored readily to detect signs of seepage from the tailings pond. Based on the EVI response analysis on the site, there are no signs of seepage of tailings pond failure based on the imagery used in this study. The only instances of EVI decreases through time were seen on farms where the farmers would post growing season burn techniques, and on land that was being developed upon. The pond s walls show notable decreases in plant health which can be expected. The EVI difference map illustrates wetland flushes that are being fed by the mine dump. The water seepage appears non-toxic in nature to the specific plant species at the observed locations. Recommendations For improved co-registration of the imagery, an up to date digital elevation model is required. The use of an up to date DEM would produce improved satellite imagery which would affect the classification and EVI analysis results. To effectively utilise satellite based monitoring over a mine tailings pond site, a monthly monitoring programme is proposed which is based on the acquisition of 50 cm WorldView2 imagery with the blue, green, red and infrared bands. The programme includes the Fig. 10: Stream B EVI analysis 2009/04/24. Fig. 11: Stream B EVI analysis 2009/12/16. PositionIT September

6 Fig. 12: Classified Wetness map 2009/04/24. Fig. 13: Classified Wetness Map 2009/04/24. Fig. 14: Mine tailings' water movement between 2009/04/24 to 2009/12/16. EVI processing, spectral classification and a report based on the findings. The WorldView2 satellite can be programmed for a new acquisition every month (dependent on cloud cover). Acknowledgment This paper was presented at the South African Surveying and Geomatics Indaba 2013 and is republished here with permission. References [1] Odis. Near Infrared Research, [Online]. Available: ndvi.html [Accessed]. [2] JW Rouse, RH Haas, JA Schell, and DW Deering: Monitoring vegetation systems in the great plains with ERTS. Third ERTS Symposium, NASA, 351, [3] D Conway and S Donelly: Remote Sensing and Ground truthing. Geoscience, Contact Prevlan Chetty, Southern Mapping Geospatial, Tel , prevlan@southernmapping.com 28 PositionIT September 2013