| Summary: | The Minerals and Petroleum Resources Act (MRPDA) No 28 of 2002 of South Africa states that the holder of a mining permit remains liable for environmental consequences until a closure certificate has been issued, but does not stipulate the environmental standards required to obtain such a certificate. Monitoring of surface mining environments requires a consistent, repeatable and efficient method of monitoring that can be applicable to heterogeneous landscapes on large properties. To this end, this study forms a component towards the development and local testing of an internationally accepted, monitoring toolkit to monitor mine rehabilitation. Landscape Function Analysis (LFA) is a technique to rapidly determine broad biogeochemical processes occurring at the soil surface in heterogeneous landscapes. However, LFA is time consuming. Hyperspectral remote sensing (HSRS) is an alternative technique for monitoring large landscapes and is sensitive to both plant response to stress and soil minerals. The aim of this study is to derive LFA indices from HSRS (i.e. surface reflectance) data acquired with a hand-held spectrometer in order to predict landscape condition on deep-level gold mining surface environments in the Highveld region. The first objective was to test the potential of Partial Least Squares Regression (PLSR) modelling to predict LFA indices from the spectral data. The second objective was to test the potential for using Vegetation Indices (VI), calculated from hyperspectral data, to predict LFA indices. Twenty-three VIs, covering plant pigments (i.e. chlorophyll, carotenoids and anthocyanins), plant structural components (cellulose and lignin) and plant water content, were tested. The study was carried out in winter (dry season) as this is the season when disturbance is most visible, and both seasonal (deciduous) vegetation growth and annual species are absent. The study was carried out at two gold and uranium mining operations in the Highveld grassland biome: West Wits Operations near Carletonville (Gauteng Province) and Vaal River Operations near Klerksdorp (North West Province). At Vaal River, data was collected from high and low disturbance sites replicated three times, in each of four of the dominant vegetation types: wet grasslands, non-rocky grasslands, rocky grasslands and woody shrub sites representing increasing structural complexity. At West Wits Operations (n = 6 sampling plots), only non-rocky grasslands were sampled. Twenty five circular quadrats of 50 cm diameter were evenly distributed on five gradsects within each plot (Total quadrats = 750). Paired data acquired from each quadrat were reflectance data (44 cm field of view), LFA data (50 cm circular quadrat), and a photograph for later allocation of the remaining LFA data. Time constraints collecting LFA data reduced the total number of quadrats sampled in the field from 750 quadrats to 150 quadrats. Difficulties in accurately pairing the LFA and HSRS data further reduced the number of quadrats I used for statistical analyses to 105.The results of ranking the three LFA indices showed that stability was above the threshold value for sustainability, while infiltration was below threshold and nutrient cycling was close to threshold for all vegetation types and disturbance levels combined. These results suggest that soils were crusted and promoting run-off, and that disturbance was mainly impacting the vegetation component, rather than the soil component of the landscape. A comparison of non-rocky grasslands between the two mining regions showed that West Wits had higher LFA indices for infiltration and nutrient cycling (t-test, P ≤ 0.01, DF = 36.8 and 26.4 respectively) than Vaal River. All three LFA indices: stability, infiltration and nutrient cycling, differed between vegetation types (One-way ANOVA, P < 0.05, DF = 3, 101) with wet grasslands having consistently higher LFA indices than the other three vegetation types. Disturbance levels, combining vegetation types and mining region, also differed (t-tests, P < 0.01, DF = 81.8, 102.3 and 100.08 for stability, infiltration and nutrient cycling respectively), with high disturbance quadrats having lower LFA indices than low disturbance quadrats. When comparing LFA indices between disturbance levels within each vegetation type, low disturbance sites generally still had higher LFA indices than high disturbance sites (P < 0.05). These findings support the initial selection of distinct vegetation types and disturbance levels, with exceptions to this pattern believed to be a result of low replication (n = 5) for these vegetation types. The twenty-three VIs were not useful for predicting LFA indices from HSRS data under my experimental conditions. All the VIs had generally low indices as expected (in the case of chlorophyll and plant water-based VIs) for winter senesced Highveld grasses. All linear regressions between LFA indices and VIs had very weak coefficients of determination (r2 < 26%). The lignin index (NDLI) had the strongest coefficient of determination for both the stability (r2 = 25%, P < 0.01) and the nutrient cycling indices (r2 = 25%, P < 0.01). The infiltration index had the strongest coefficient of determination with the standard normalised difference vegetation index (NDVI) (r2 = 16%, P < 0.01). VIs had generally very low indices due to the winter senesced state of the Highveld vegetation. PLSR modelling produced much stronger regression coefficients of determination than did the VIs. The best PLSR model was a 15-component model to predict nutrient cycling (r2 = 54%, P < 0.01). A 13-component model predicting stability had an r2 = 38 % (P < 0.01), while a 17-component model was derived for infiltration (r2 = 32%, P < 0.01). In all three cases, these models were able to account for more than 90% of the spectral variability within the first two components. However, more than 16 components were required to account for 90% of the variability in the LFA measurements. It may be possible to reduce the number of components required for the PLSR modelling of the latter with a more standardised approach to the LFA data collection, i.e. having one observer who acquires all the LFA data in the field, and increased replication.
|
|---|
