Fueled by migration from rural areas, valleys and plains; the growth of religious, ecological and adventure tourism; and recent social unrest, towns and urban centers are expanding. Although the region is still predominantly agrarian, migration to urban centers is an increasingly important livelihood strategy for rural households, and non-farm income is an increasing component of household incomes. As recently as 1981, less than 10% of the Himalayan population lived in a town or city. By 2000, the urban population in the region had doubled to 20%. The growing urban population, an urbanizing economy, and associated land use and land cover changes are transforming the Himalaya. Construction of buildings, deep roadcuts in steep hillsides, and unplanned urban development, all of which require cutting into bedrock or crossing geologically weak areas, have resulted in increased and more severe occurrences of hillside collapse, landslides, debris flows, and rock slides are putting millions of people at risk.

In the Himalaya, the scale of urban growth—which is expected to be small in spatial extent, but geographically distributed throughout the region—poses a challenges for remote sensing analysis. Given that the materials used for urban development have spectral signatures similar to fallow land and barren soil, care must be taken to avoid confounding urban change with agricultural land. Since this particular challenge is amplified if a pixel at a single point in time is examined, using a time series helps distinguish different land uses because the temporal signatures of urban settlements and urban growth differ significantly from the phenology of vegetation. It is for these reasons that we are adapting two time series algorithms which have been designed for flat terrains, the Continuous Change Detection and Classification (Zhu and Woodcock, 2012) and an econometric time series approach similar to those developed by co-PI Seto. A novel contribution of our remote sensing analysis to characterize urbanization is developing algorithms using the entire Landsat TM archive from 1982 to present.

The study region is covered by 41 Landsat tiles, each of which has 300-600 images for the entire study period. In total, we expect to process around 20,500 images totaling 16TB and 1.3 terapixels. In January 2018, we conducted a scoping exercise in Nepal and India and will conduct an accuracy assessment field survey this summer.