Dr. Yichun Xie
Professor, Institute for Geospatial Research and Education, Eastern Michigan University, USA
Speech Title: Extracting Key Indicators of Black-Odorous Water from Spectroradiometer Data and Machine Learning

Abstract: China, through 40 years of the economic reforms, has witnessed fast economic growth, which has been accompanied with severe environmental challenges. One of those concerns is the surface water pollution, called black-odorous water, which have occurred in almost all big rivers as well as small streams. The Sate Council of China on April 16, 2015 unveiled its first Action Plan for Water Pollution Prevention and Control, aiming at cleaning up 70% black-odorous water in major rivers and cities by 2020. However, there are several technical challenges hindering this Action Plan: 1) What are the key indicators of black-odorous water? 2) Is the traditional direct measurement approach too time-consuming and costly; 3) Can remote-sensing techniques be used to replace the traditional measurement? 4) Are there effective methods to identify key indicators of black-odorous water from the remote sensed data? We surveyed current water quality indicators through literature review and standard operational manuals. We took 52 black-odorous water samples in the Chebei and Yonghe Rivers in the City of Guangzhou from September to October in 2017. We measured the indicators of dissolved oxygen (DO), ammonia nitrogen (NH3-N), total phosphorus (TP), total nitrogen (TN), chemical oxygen demand (COD), five-day biochemical oxygen demand (BOD5), and suspended solids (TSS). We also collected the full spectral data from 350nm to 2500nm at 1nm spectral resolution synchronously by using a spectroradiometer. Together the water quality indicators and the spectral reflectances composed a big data-set. We are investigating four research questions: 1) how and why each water quality indicator contributes to the severity of black-odorous water; 2) how and why each wavelength reflectance relates to the severity of black-odorous water; 3) which wavelength reflectances complement with what water quality indicators to reveal the most sensitive responses to the severity of black-odorous water; and 4) how we shall scale up the spectroradiometer measurement to other remote sensed data collected through drone-based or airborne or satellite sensors. We are adapting the double-Lasso data mining as the primary variable selection technique to answer the research questions and will report the findings at WRE 2019.