Machine Learning of NDVI Time Series Identifies Ghost Villages in Uttarakhand, India
Sarvjeet Kumar, Prakhar Misra · 2025
The phenomenon of rural depopulation, leading to abandoned villages is known as ‘Ghost Village’ phenomena. It has been emerging as a major challenge in various Himalayan states in South Asia, including the hilly state of Uttarakhand in India. As per the last population census conducted in 2011, there are already more than 1000 ghost villages in Uttarakhand. As a result, more and more villages in Uttarakhand are becoming abandoned, and some districts, like Pauri, Garhwal, and Almora, are experiencing negative population growth. This research utilizes long-term normalized difference vegetation index (NDVI) time-series (from 2000 to 2010), derived from the Moderate Resolution Imaging Spectrometer (MODIS) satellite sensor, to investigate temporal farmland patterns and identify ghost villages within districts of Uttarakhand. Performance of two common machine learning algorithms, logistic regression, and random forest is assessed for identifying the absence of agricultural and human activities, a pattern associated with ghost villages. The random forest classifier performed better than logistic regression, with the former having an overall accuracy of 81%. Training the model on longer time-series also increased the detection accuracy. Using the trained random forest model, 83 low-population villages were predicted to have converted into ghost villages in the Pauri Garhwal and Almora district from 2010 to 2020. Our research can be extended for inventorying ghost villages across Uttarakhand, having direct implications for studying urbanization and migration, anthropogenic land use and land cover change, and policy development in the Himalayan region.