Artisanal and Small-scale Mining (ASM) is on the rise in the Global South, providing a primary source of income for many communities, especially among the rural poor. However, many ASMs are often operated illegally or informally, leading to major social and environmental problems such as water pollution, low worker safety, and deforestation. Despite these problems, the informal nature of the sector currently presents serious challenges related to accurately locating ASM activities in a timely manner to support their sustainable development. Remote sensing methods are therefore suitable for regular monitoring, although the use of optical images has faced challenges such as spatial resolution and persistent cloud cover, which limit their capabilities for regular ASM monitoring in cloudy regions like southern Ghana. Synthetic Aperture Radar (SAR) offers a solution for regular ASM monitoring in cloudy regions due to its ability to penetrate clouds. However, the use of freely available Sentinel-1 (S1) Cband imagery can be affected by several factors, such as speckle noise, atmospheric effects from active rainfall events, and topography, leading to a greater or lesser degree of high false alerts and measurement uncertainties. These constraints have resulted in a lack of research to date that has explored the timely detection of ASM to facilitate rapid response. This thesis first investigates an optimal SAR processing methodology in addressing these constraints to subsequently develop a new scalable SAR-based method, the first to leverage Change Point Detection (CPD) and Deep Learning (DL) techniques for timely and reliable ASM monitoring using C-band S1 imagery in persistently cloudy regions. In this thesis, the use of S1 radiometric terrain-normalised γ0 backscatter products processed using Copernicus 30m digital elevation model (DEM) is demonstrated to be superior to conventional σ0 backscatter products in mitigating terrain-induced effects to improve the timely detection of ASM. This thesis makes an important contribution by proposing time series smoothing as an essential additional SAR preprocessing step and as a robust, practical approach for effectively smoothing S1 data with outliers caused by high-intensity rainfall and seasonality, enabling timely and reliable ASM detection, particularly during rainy seasons. CPD is a powerful statistical technique for analysing smoothed S1 SAR imagery to generate ASM-induced forest disturbance alerts in mining landscapes, reducing the delays in locating ASM from several months using traditional methods to within a few days, with possible delays of up to 18 days. In this thesis, the potential of DL to overcome the speckled effect associated with traditional pixel-based Machine Learning (ML) methods is demonstrated to improve the accuracy of distinguishing mining from other forest disturbances from S1 SAR imagery promptly, with a mean IoU of 0.82 and an overall F1-score of 89%. The methods tested in this thesis are a step towards achieving operational ASM early warning systems, as this will be crucial for mitigating their social and environmental impacts and providing emergency aid to protect lives and the natural environment globally.