Geochemical Sampling Scheme Optimization on Mine Wastes Based on Hyperspectral Data

Spatial sampling optimization is an important issue for both geo-chemists and geo-statisticians. Many spatial sampling optimization methods have been previously developed. In this paper, we present a spatial simulated annealing method is presented using hyperspectral data.This sampling method was applied in a project concerning environment assessment of the Dexing Copper Mine. Mine waste contains high concentrations of metals, mostly of a non-economic value. Most of them are discharged without any decontamination, for example, acid-generating minerals. Acid rock drainage can adversely have an impact on the quality of drinking water and the health of riparian ecosystems. To assess or monitor environmental impact of mining, sampling of mine waste is required. Optimal geochemical sampling schemes, which focus on ground verification of mine wastes extracted from hyperspectral data, was derived automatic from a JAVA program.Hyperspectral data help to identify ground objects by a larger spectral range. Spectral angle mapper classification technique is carried out to obtain rule images. A rule image provides weights that are utilized in defining the objective function for the sampling scheme. These are optimized by means of simulated annealing. The simulated annealing uses the Weighted Means Shortest Distance (WMSD) criterion between sampling points. The scaled weight function intensively samples areas where an abundance of weathering mine waste occurs. A threshold is defined to constrain the sampling points to certain areas of interest. 1. OVERVIEW OF OPTIMIZATION OF SAMPLING SCHEMES To obtain a better cost-effect spatial sampling method, geo-statistical research has been carried out. In comparing traditional sampling schemes Burgess found that a regular grid results in only slightly less precise estimates than a triangular grid, for the same sampling density. Different grid designs can reach the same precision with different sampling costs(Burgess, 1981) . Some specially designed grids can be sampled at a lower price than a regular grid, while maintaining the same precision (Debba, 2006). We distinguish two kinds of sampling schemes. First, a retrospective sampling scheme that contains sample locations that are already planned. During assessment, some points will be removed from or added to the existing sampling scheme. Second, a prospective scheme contains sample locations that are pre-determined before actual sampling in the field. Both of these schemes require optimization. Optimization of a sampling scheme means a reduction in the numbers of sampling points with the same or even a higher accuracy of certain unknown parameters. In this paper we employed the second sampling scheme to explore the abundance data from an area where no prior field work is done. The mine waste deposits normally cover large area. Some piles already exist for ages. Intensive sampling over the whole area is costly. With the aid of remote sensing data the design of an optimal sampling scheme is vital to provide the researcher with the relevant information about mine wastes. 2. METHODOLOGY Figure 1 shows graphically the overview of the sampling process. The sampling process consists of three parts, of which each has several sub-processes. First, we need to decide on the area to perform the sampling scheme. We need to know the size of the area and how many samples would be deployed in this area. Prior knowledge or background information can play a key role in arriving at a suitable answer to the above. For example, the geological setting in the area, the deposit type, mineral type, land cover, land user and weather are important traits in making such decisions. Such information helps our work through the whole period. For example, based on weather information we can predict the approximate time for satellite to acquire data as the quality of hyperspectral satellite image strongly relies on weather condition. If too much cloud gathers in the air, the satellite can not get an image of a sufficient high quality. After the acquisition of an image, a quality assessment should be performed before further analysis. The second stage is the analysis of the image data. During this stage the ground weathering waste rock should be identified and mapped. Atmospheric correction is a prerequisite to most hyperspectral imagery data analysis approach. During the correction, the atmospheric disturbances should be removed. Normally, cloud areas should be masked and then further analysis will exclude these areas. After image preparation, the most important stage is ground weathering waste rock spectral discrimination and acquirement. We must ensure what we wish to identify from the image data. The ground weathering waste rock spectral data can be acquired from three different ways. First, the best option is by means of field observed spectral