Abstract: Abstract: The non-point pollution landscape source and assembly pattern of a river basin and its spatio-temporal process analysis are the scientific theoretical bases for studying the pollution, production, transport, prevention, and policy of river water. In this paper, to mine the spatio-temporal character and advantages of remote sensing data, the definition schedule of the landscape of the river basin non-point pollution source and assembly was introduced. The first types of landscape in the river basin included the impervious surface area (ISA) and the pervious surface area. The second types of landscapes in the river basin consisted of 14 different types, including woodland, orchard, road, rural area, town, and paddy field, etc. The river basin non-point pollution landscape source and assembly pattern remote sensing parsing method was subsequently presented at the pixel, sub-pixel, and time-renewing levels. (1) A "globe-local" coupling information extraction model for ISA at the pixel level was established. Through the mining and integration of the spatial information in a local image area, the spectral instability of the whole scale was optimized. The model was divided into two main computing steps: the "global" prior classifier, and the "local" post classifier. The prior classifier only extracted pixels that would satisfy a certain accuracy threshold, which was based on the probability of classification. The unclassified pixel was handled by the post classifier, which first mined the partly classified spatial information result and then computed the new spatial features, such as distance, texture, pattern, and other features for assisting with the subsequent pixel classification. (2) A sample library-based ISP estimation model at the sub-pixel level was established. After comparing the two classic methods of ISP computation, spectral mixture analysis and the machine study model, a sample library-based ISP model was established. Research on the main technical method, including the establishment of a training sample set, model application strategy, and repair at the sub-pixel level was introduced. (3) An impervious updated model at a temporal scale was established. The model uses pre-existing impervious information to extract updated information on images from other periods. The information in the unchanged area was used to infer and update the attribute information in the changed area, which analyzed the information at a multi-temporal scale. Using the Jiulong river basin as an example, the extraction of ISA information at the non-point pollution source pattern in 2010 was achieved. Renewing and spatial analyses were also conducted in 2000 and 2005. The obtained results demonstrated that ISA increased by 33.38% from 2000 to 2010.