Methodology

This study attempts to assess the relevance of raster population datasets in the events of natural hazards and compare patterns relating to urban agglomerations and population densities. For doing so, LandScan, 2008 raster population datasets from Oak Ridge Laboratory is used along with census datasets. It is useful to investigate and to assess the strengths and weaknesses of existing census data, methods (e.g., gaps in spatial and thematic coverage, counting individuals, proxy measures such as those derivable from Earth observations), and tools for estimating population living under risk conditions. It is imperative for decision-makers for identifying populations at risk categories, that are susceptible to the impact of natural or human-induced disasters, thus there are three critical elements of the data, each of which is a scale issue: spatial scale (how far below the national level can estimates be derived?); temporal scale (for how recent a time period can estimates be made?); and risk scale (how detailed are the available population characteristics of living, place of accommodation and shelter?) (Comenetz 2007). Since, Landscan data only reflects population number and density, thus for satisfying third element, more ground survey is required to understand household characteristics. Thus, only aspect of population captured by these approaches is total population size. No other demographic information that could identify risk (beyond being on the footprint of a Hazard), such as age, gender, or race or ethnicity, is currently available in these data collections. Nonetheless, one further do dasymet- ric modelling to integrate both datasets to have considerable utility and to be used for purposes including emergency planning and hazard management. To have confidence in Landscan data and its applicability, evaluation was carried out across various scales, namely state, districts and municipal wards. Later, application and estimation of vulnerable population across various municipal wards along Yamuna flood plain in Delhi was carried out (Fig. 1).

First, predicted population totals per state (district level for costal districts of India) were compared to the Census population adjusted estimates for the year 2008. The LandScan dataset was unsurprisingly nearly perfect, as the population data were matched approximately to Census population estimates in the modelling

Q

Fig. 1 Population comparison between census of India population (2008) and landscan 2008 (Authors work)

Fig. 2 Population comparison between census of India population (2008) and landscan 2008 (Authors work)

procedure. However, the aim is to observe how far away the LandScan datasets is from these most contemporary estimates. R² (0.999) were extracted and differences in population estimates per state/district were mapped.

Second, grid-based differences between datasets were measured. Unit-level absolute differences were mapped and plotted to explore tendencies in these differences. Third, the numbers of people predicted in towns and settlements with known population size have been compared. In order to allow the calculation of population predicted in small settlements (smaller than 1 km), the LandScan datasets were used for comparison (Fig. 2). (R²) between predicted and observed population in towns and settlements were extracted. The impact of the choice of population dataset on estimates of the population at risk was tested for flood event in Delhi before common wealth games in 2010.