Remote sensing of land use changes and impacts on river turbidity in a small-scale mining region

Abstract

The effective management of artisanal and small-scale mining (ASM) on regional and national scales must be based on good understanding of land and water footprints from various land use and land covers. The diffuse, dynamic and often remote nature of ASM means that traditional ground-based surveys are likely to be impractical except for local scale studies. Remote sensing offers a low-cost option for surveying land use changes and water turbidity, and quantifying the impact of ASM on water quality. However, there are questions about the reliability of remote sensing products for these tasks, and there is a need for recommendations about suitable products, data resolutions and analysis techniques. A case study of the Addalam river basin in the Cagayan region, situated in Luzon forming a part of the Philippine archipelago, was used to address these research questions. The value of alternative satellite products was tested using independent sources of land use, suspended sediments and turbidity data from project partners OceanaGold (Philippines), International RiverFoundation, and local government agencies. The unpredictable climate in wet tropical regions, and the spatial limitations of current satellite imageries are the challenges for remote detection of ASM. Pleiades and SPOT imageries were identified as potentially suitable and were tested. Historical spatial data on location and type of ASM mines were collected from the field, and were utilised as training data for classification through the OB-SVM classifier. The analysis resulted in overall accuracy between 87% and 89% for three different images; Pleiades-1A HiRI sensor for the 2013 and 2014 images, and SPOT-6 NAOMI sensor for the 2016 image. The main land use features, particularly the Didipio large-scale mine, were well identified by the OB-SVM classifier; however, the presence of small-scale mines was slightly under identified. The lack of consistency in their shape, and their small scale compared to the pixel sizes, meant they could not be reliably distinguished from other land clearance types. The biased-adjusted surface areas were acquired to determine the best possible estimates of the area variation in small-scale mines throughout the year. The image analysis indicated an increase in small-scale mining area from 91,000 m2 or 0.2% in March 2013 to 121,000 m2 or 0.3% in May 2014, and then a decrease to 39,000 m2 or 0.1% in January 2016. Various land use features in a mining region have different sediment yields, which have significant influence on the concentration of suspended solids in rivers. In-situ sampling can only describe the integrated impact of the upstream land uses. A model of total suspended solids (TSS) through the acquired surface reflectance from calibrated satellite images can be used to assess the fate and transport of sediments throughout the catchment of Didipio. The surface reflectance data from the satellites were used to develop a regression model of the TSS concentration. A linear model was derived between the surface reflectance data from Pleiades-1A and the corresponding ground-based measurements of TSS concentration. The regression using the red channel reflectance gave R2 values of 99% and 58%, respectively, for the two separate images of Pleiades-1A in 2013 and 2014. However, the regression model of the integrated dataset of surface reflectance and TSS from both images resulted to R2 value of 20%, and an RMS error of 937 mgL-1. The generated model represents sediments from different sources. It was noted that the increase in small-scale mines from 2013 to 2014 resulted to additional stripping of topsoil, which has lower surface reflectance when compared against fine particles with similar magnitude of TSS. The model was used to generate a map of continuous concentrations of TSS from the upstream to downstream of the catchment. However, the lack of images and simultaneous ground-based measurements of TSS concentrations that are higher than 3,580 mgL-1 means that the model cannot be confidently applied over the full relevant range of TSS. Overall, it is concluded that remote sensing shows promise in capturing the land and water quality footprints of ASM. However, the lack of cloud-free images in the case study and wet tropical regions in general is a significant problem in terms of capturing a sufficient number, and range of samples to develop the models. The integration of more images and simultaneous TSS samples, including higher concentrations, is recommended. On the other hand, further works on classification of land use features can be performed over the regions with less obstruction from cloud coverage (e.g. sites at lower ground elevation) and investigate the applicability of recently available moderate resolution satellite imageries (10 m or better) such as Sentinel-2 that can maintain the continuity of satellite images at no cost.