Ground Truthing of Dumpsites using Remote Sensing and Machine Learning Approaches in Peri-Urban Settings

The open dumping of waste poses severe environmental and public health hazards when exposed to the atmosphere. Therefore, to reduce these consequences, it is crucial to identify waste disposal sites across large areas. However, the local government agencies and environmental groups often pose significant challenges in obtaining the information on dumpsite location data promptly. Hence, the present study focused on the detection of the existing dumpsites in Madhya Pradesh using the Random Forest (RF) machine learning technique with the use of Sentinel-2 images for the year 2022. The logistic regression function was then implemented to analyse the influence of Land surface temperature (LST), Normalized Differential Vegetation Index (NDVI), and Normalized Differential Built-up Index (NDBI) on classified dumpsite features. The RF technique achieved an overall accuracy of 86.49%. The LST, NDVI, and NDBI values were extracted for the 37 sample datasets. The extracted temperatures for dump sites vary from 35.47 to 39.58 ℃, whereas the NDVI and NDBI range between 0.04–0.25 and − 0.06 to 0.12, respectively. Subsequently, the overall accuracy of logistic regression obtained was 88%, showing the collective findings of LST, NDVI, and NDBI demonstrate a significant contribution to the dumpsite detection. Using this novel approach, 60 undocumented dumpsites were successfully detected. The developed methodology effectively detects dumpsite locations; however, it lags in analyzing the morphological and compositional attributes of such areas. Therefore, the further studies will focus on integrating the field investigations with high-resolution remote sensing data to assess characteristics of the identified dumpsite locations.