Integrating Remote Sensing for Spatial Interpolation of Water Quality Parameters using Deep Learning

Abstract Submission: Natural and anthropogenic features such as river meanders and weirs create significant spatial heterogeneity in rivers, complicating water quality evaluation across time and space. This variability is influenced by river depth and slope, which are shaped by watershed characteristics, land use, and tributary inputs. In this regard, traditional in-situ sampling methods often fail to capture water quality dynamics fully, particularly in areas with sparse monitoring stations. In this study, we integrate Sentinel-2A remote sensing data with deep learning to improve the spatial interpolation of four key water quality parameters: dissolved oxygen, nutrients (total phosphorus and total nitrogen), and turbidity along the Kissimmee River in South Florida. In our preliminary analysis, we developed a deep neural network (DNN) to classify upstream and downstream stations of the river using spectral data from Sentinel-2A imagery. Preliminary results suggest that our model accurately captures spatial variations in water quality along the river, achieving a classification accuracy of 0.91 in distinguishing between upstream and downstream areas. Given its adaptability and ability to leverage high-resolution remote sensing data, this method aims to improve the accuracy of water quality interpolation. This approach will offer a more detailed and accurate understanding of water quality variability. Unlike conventional methods such as kriging and inverse distance weighting (IDW), which struggle with non-linear and complex spatial patterns, integrating remote sensing data with DNN provides a more flexible and promising solution. Our proposed approach aims to improve water quality assessments by addressing these limitations to monitor rivers across a broader spatial extent.

Learning Objectives/Expected Outcome (Optional) :