Seasonal population estimates based on night-time lights

Highlights

• Night-light images are known to strongly correlate with ambient population.

• SUOMI/VIIRS satellite monthly composit night-lights are used as a proxy to derive seasonalpopulation.

• The method is based on measuring for each region how much each month is different compared to March which is assumed to be closest to the census/resident population.

• The method is tested in several islands where seasonal population can be estimated by official statistics since arrivals by air and by sea are recorded.

• Validation shows a strong correlation of observed by remote sensing and reference (official statistics) seasonal population.

Abstracts

The objective of this paper is to present a method for estimating seasonally specific ambient population counts. The central assumption is that the variation in observed night-lights is a valid proxy for ambient population. Island populations are used for validation, where it is possible to derive estimates of ambient population from national statistics. The method is then applied to the whole of Greece. The validation shows a strong correlation amongst night-lights derived estimates and the reference dataset. Based on the proposed method, national maps are produced showing the month when seasonality is in its peak, the peak value during that month and the overall length of the season, in terms of how many months exceed a certain threshold. Different seasonality patterns are revealed. An advantage of the proposed method, compared to other contemporary approaches, is that it is based on public domain, global data.

Introduction

Population counts and estimates are normally available from censuses and registers. Census residential population is counts of people at their permanent residencies (a.k.a. “night-time” population). Ambient population, as opposed to residential population, is the total population present at each particular location at a given time (Amaral et al., 2005, Sutton et al., 2003). The main benefit of ambient population is that it can be used to track the movement of people within the day (a.k.a “day-time” population). Ambient population shows for example the concentration of people in shopping centers during the day and in stadiums during the weekends, both void of resident population. Censuses are typically performed once per decade. But people spend time in places other than their residence, generating a demographic concept known as the seasonal population. People move in space for multiple reasons, including commuting to work and taking holidays. Depending of the length of stay, different seasonal population groups are formed, with quite distinct characteristics. A first group refers to tourists traveling to places with some kind of touristic interest. Tourists that only make short stays are also known as visitors. A second group includes seasonal workers attached to seasonal jobs (in tourism, agriculture, construction etc.). Third is the group that includes second-house owners. They are typically expected to spend more days than tourists at their second-house location. Fourth, migrants, registered or unregistered, that move in space for several reasons including refugees escaping conflicts as well as people moving due to socioeconomic factors.

Seasonal population, typically exhibits a peak time at each place. Seasonality peak-time depends on what is actually attracting the additional, non-resident, people. For tourists it can be summer holidays, in coastal areas, or winter holidays, in mountainous regions. It may be religious events triggering the movement of people throughout the year (Roman & Stokes, 2015). In any case, time-specific population distributions within a year are hard to estimate by conventional means. Custom surveys are inevitably limited in scope and costly, rendering repetition infrequent. Proxy variables such as water and electricity consumption can also be used to estimate seasonal population. The reliability of this approach is however reduced by the fact that (i) consumption data are available at two or three month intervals (ii) billing is sometimes based on estimated rather than actual consumption, for long periods of time before an actual measurement is made (iii) ephemeral events are present in the data (measurement errors, leaks etc.) (iv) per capita consumption of water and electricity is not constant throughout the year, consumption patterns change during the summer due to swimming pools, increased irrigation needs, excessive use of air conditioning etc.

Occupancy of tourist accommodation establishments is also frequently used as a metric for estimating the additional present population. The recorded data include percentage of beds occupied in the establishments. However, this data is affected by the shadow economy (illegally operating establishments) and by tax evasion (legally operating establishments not declaring all stays). Recently, occupancy data are becoming even more unreliable due to the rapid takeover of disruptive technologies such as the AirBnB platform (Airbnb, 2017) which currently escapes national statistics. It has been suggested that Airbnb already provides a viable alternative for certain traditional types of overnight accommodation (Zervas, Proserpio, & Byers, 2016).

The phenology of population and the ability to capture and anticipate its dynamics has crucial implications in several domains. For example, seasonal population is useful in risk management in order to better estimate population at risk at each time (Smith et al., 2015). It is also important for improving energy demand forecasting (Roman & Stokes, 2015) and in epidemiology to monitor disease outbreaks (Bharti et al., 2011). In planning, both the amount and the type of seasonal population is essential because it substantially alters the demand for services. The demographic and socioeconomic profile of residents vs. the several non-residential groups is typically very different. Imposing a different strain on services and infrastructure.

Seasonal population is also one of the primary factors in determining the carrying capacity of a particular place. Carrying capacity is defined as the point where a place becomes insufficient to meet, without degradation, the needs of both resident and seasonal population due to natural or anthropogenic (infrastructure) constrains (Coccossis & Parpairis, 2000). Respecting the capacity of the local system is important both in the context of sustainable development and for maintaining the attraction and competitiveness of touristic destinations (Coccossis & Mexa, 2004).

Recent efforts to estimate seasonal population have mainly focused on mobile phone-call records (Deville et al., 2014, Erbach-Schoenberg et al., 2016, Hanaoka, 2016, Ratti et al., 2006, Reades et al., 2009, Wesolowski et al., 2015, Wilson et al., 2016, Yang et al., 2016). Essentially, phone-call records are treated as big-data. While this dataset provides powerful insights in population movements, it is limited by the fact that phone call data are not public domain. Therefore usage is restricted to those having access to the data. Roman and Stokes (2015) recently exploited a different data source, daily satellite night-light images, revealing seasonal movement at fine intervals of space and time. However, the night-light averages used for periods of less than a month are currently also not public domain.

In general, the strong correlation of night-lights with population has been proven in numerous studies. Elvidge, Hsu, Baugh, & Gosh, 2014 observed a strong correlation between night-lights and population in most countries of the globe. Stathakis (2016) noted a very strong correlation between night-lights and resident population for Greece in specific. Amaral et al. (2005) noted a strong correlation between night-lights and urban population in Brazil. Other than demography, previous studies used night-lights in numerous domains such as to assess the economic performance of areas (Ma et al., 2012, Triantakonstantis and Stathakis, 2014, Stathakis et al., 2015), urbanization (Stathakis, 2015, Zhang and Seto, 2011, Zhang and Seto, 2013) health-related topics (Kloog, Haim, Stevens, & Portnov, 2009) and conflicts (Li & Li, 2014).

Our objective is to propose a method to estimate seasonal population based on global, publicly available data. The proposed approach is based on monthly composites of night-light satellite images that only recently became available. The dataset is described in detail in the next section. The method to use it as a proxy for monthly population estimates is then introduced. The main novelty of this approach is that the strong correlation of night-lights with population is exploited to derive monthly rather than annual estimates, based on a new method developed to efficiently process the satellite data. We believe that the proposed approach can fill the current data-gap with respect to monthly population estimates and trigger practical applications in this newly available human-geography time-scale.