A high-resolution prediction system for birch pollen in Sapporo
Introduction
Hokkaido is the northernmost main island of Japan (Fig. 1a) with a population of about 5.3 million, and is an area with a high prevalence of birch pollinosis, which causes springtime allergy-related symptoms, such as stuffy nose, watery eyes and sneezing, and asthmatic reactions (Gotoda et al. 2001; Baba and Nakae 2008). About one-third of the population is concentrated in Sapporo, the capital of Hokkaido Prefecture. Residential and commercial areas extend north and east over the plain about 10 km from the centre, and there are mountainous areas to the south and west, with Mt. Teine in the west, Mt. Moiwa in the southeast, and Mt. Yakeyama in the south (Fig. 1b). Because birch is a shade-intolerant species, there are clusters of Betula platyphylla var. japonica, B. ermanii, and B. maximowicziana around the summit and in the ridgelines of the mountains. Moreover, there are birch trees, mostly B. platyphylla var. japonica, planted along major roads and in city parks administered by the Greenery Promotion Department of Sapporo City. Given that we expected that the bulk of pollen would be deposited near the source plant (Bricchi et al. 2000; Tampieri et al. 1977; Wright 1953), the instantaneous amount of airborne birch pollen was expected to be sensitive to the density of birch trees (Skjøth et al. 2008). Hence, the prediction of daily or hourly amount of pollen in Sapporo requires primarily a high-resolution map of birch tree distribution, which would help pollinosis patients take preventive action (Peden and Reed 2010). However, no map has been compiled because the trees are distributed irregularly in roads, parks, estates, and forests owned by various organisations and individuals.
The prediction of pollen concentration has long been a subject of study and requires a dynamical model system that accounts for the total amount, flowering, emission, dispersion, and deposition. The total amount of pollen during a single season has often been evaluated by seasonal pollen index, a proxy for allergenic exposure of the population (D'Amato et al. 2007; Gioulekas et al. 2004). The magnitude of flowering varies between years and is affected by the environmental conditions and the flowering intensity of the previous year (Dahl and Strandhede 1996; Ranta et al. 2008; Stach et al. 2008a). The total amount of pollen exhibits a pronounced biennial rhythm in Sapporo (Kobayashi et al. 2013). The total amount of pollen is also highly correlated with the number of catkins (Ranta et al. 2008), and can be predicted the year before (Yasaka et al. 2009). Shirasaki et al. (2014) found a significant positive correlation with the sunshine hours in June of the preceding year. Birch flowering timing is driven by heating degree days relative to a base temperature of 3.5°C from 1 March (Larsson 1993; Linkosalo et al. 2010). The onset date ranges from 70 to 120 degree days in central Europe (Sofiev et al. 2013; Verstaeten et al. 2019), and a pollen calendar model has been developed for predicting daily concentrations of airborne pollen in Serbia (Šikoparija et al. 2018). High temperatures trigger pollen emission into the air (Clot 2001; Ribeiro 2003) and the diurnal cycle of temperatures and sunshine also affects pollen emission (Fehér and Járai-Komlódi 1998). Based on these results, a dynamical model system for pollen prediction was developed using a conventional transport-deposition model for aerosol particles (Sofiev et al. 2006a, 2013). The dynamical model predicted the transboundary pollen transport effectively (Siljamo et al. 2012), and Sofiev et al. (2015) performed the first ensemble modelling experiment on birch pollen in Europe. However, because up to 1% of released pollen grains can travel far from the source despite the large dry deposition velocity of around 1.2 cm s−1 (Sofiev et al. 2006b), the accuracy of pollen prediction strongly depends on information about plant distribution.
Inatsu et al. (2014) reported a simulation of daily birch pollen in Sapporo. They constructed a multiple regression model with two independent predictors of lagged pollen amount and diurnal temperature range (Kawashima and Takahashi 1995), based on observed data from 2001 to 2010 in Sapporo (Kobayashi et al. 1998). A hindcast simulation for 2011 reproduced the daily variation in pollen amount with a correlation coefficient with the observed data of 0.84. However, using the pollen amount from the previous day (Inatsu et al. 2014; Stach et al. 2008b) may not be practical because the current day's pollen data are not available in time to produce tomorrow's pollen forecast. Ritenberga et al. (2016) excluded this nudging predictor by preprocessing, including normalisation by the seasonal pollen index, using heat sum as a time axis, projecting meteorological variables to pollen concentrations, and removing the mean pollen season curve. Nevertheless, no matter how complex the system, statistical prediction only provides the pollen amount as a relative value that optimally fits training data at a particular site. Because most of the released pollen grains settle near the source, the pollen amount slightly away from the site must differ substantially from the statistically predicted amount at the site, although these two amounts may be strongly correlated with each other. Moreover, other aerosol studies (Neubauer et al. 2014) suggest that the spatial distribution, although dependent on the tree density distribution, may have a much larger effect than the temporal variation. Hence, the statistical models for a single site may not be suitable for evaluating the higher exposure risk around birch trees. Even though the absolute amount of airborne pollen grains is helpful for pollinosis patients, the spatial distribution has not been examined, probably due to the lack of a high-resolution birch tree map.
The purpose of this study is to develop a high-resolution prediction system for birch pollen in Sapporo, including mapping the tree density and modelling the airborne birch pollen transport and deposition. A crucial stumbling block for such a prediction is the lack of a high-resolution birch tree distribution map. About one-third of birch pollen is deposited within 800 m of the source (Bricchi et al. 2000); thus, we first create a map of Sapporo city centre with 400 m spatial resolution based on a field survey along public roads and footpaths and Street View on Google Maps (Section 3). Second, we originally develop a transport-deposition model driven by uniform meteorological data for the city (Section 4). The model includes all the components necessary for birch pollen prediction, but it should be as simple as possible to facilitate a practical forecast. Third, we perform a hindcast simulation from 2001 to 2011 in Sapporo and compare the result with the pollen observations at a site (Section 5). We also compare the spatial variation with the temporal variation in the simulation result, which has not been reported in previous studies. Finally, we discuss the improvement of our prediction system (Section 6).
Section snippets
Data
We used the data for airborne pollen amount between 2001 and 2011 observed at the Hokkaido Institute of Public Health (HIPH) in Sapporo, which is located at 43°05′N, 141°20′E (Fig. 1b). The pollen was collected on petrolatum-coated glass slides in a Durham sampler installed on the rooftop 16 m above ground level (AGL). Durham samplers have a comparable performance to Burkard samplers (Kishikawa et al. 2009). After the slides were stained with gentian violet, the number of pollen grains was
Birch tree map
The horizontal distribution of birch trees was mapped onto a 400-m horizontal grid within 10 km of Sapporo city centre (Fig. 1b), based on a field survey, a search by Google Street View and Google Maps and an extrapolation. The procedure is illustrated in Fig. 3. First, we generated about 36,000 1-ha blocks in the circle with MANDARA (Geographic Information System, http://ktgis.net/mandara/). An author (RY) manually counted the number of birch trees along the city roads in a block in the
Prediction system
The prediction system we developed contains three components: flowering time, emission, and transport and deposition.
Hindcast simulation
We performed the hindcast simulation with the model described in Section 4. The target years were 2001 to 2011, and the daily deposition amount of the simulation at the grid cell containing HIPH was compared with the daily amount of pollen grains (unit of cm−2 day−1) collected in the Durham sampler there (Fig. 8)3
Conclusion and Discussion
We developed a high-resolution prediction system for birch pollen in Sapporo. We created a birch tree density map within 10 km of Sapporo city centre in 16 ha blocks by a field survey along public roads and footpaths and using Google Maps Street View (Section 3). We developed a system consisting of components such as flowering time, emission, and transport and deposition (Section 4). To facilitate practical forecasting, the model was designed to be as simple as possible, based on phenological
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
We thank the two anonymous reviewers for providing us with insightful comments that helped improve this paper. We are deeply indebted to Hokkaido Agricultural Research Center of National Agriculture and Food Research Organization, Hokkaido Research Center of Forestry and Forest Products Research Institute, Hokkaido Agricultural Technical College of Hakkou Gakuen, and Sapporo Development Corporation for assistance with the field survey of birch tree distribution in their domain. The Greenery
References (42)
- et al.
Environmental and occupational allergies
J Allergy Clinic Immunol
(2010) - et al.
Male flowering of birch: Spatial synchronization, year-to-year variation and relation of catkin numbers and airborne pollen counts
Forest Ecology and Management
(2008) - et al.
Statistical modelling of non-stationary processes of atmospheric pollution from natural sources: example of birch pollen
Agri Fore Meteorol
(2016) - et al.
A dispersion modelling system SILAM and its evaluation against ETEX data
Atmos Environ
(2006) - et al.
Long-term and short-term forecast models for Poaceae (grass) pollen in Poznan Poland, constructed using regression analysis
Environ Exp Bot
(2008) - et al.
Medium range transport of airborne pollen
Agric Meteorol
(1977) - et al.
National epidemiological survey of nasal allergy 2008 (compared with 1998) in otolaryngologists and their family members
Progress in Medicine
(2008) - et al.
Experimental results about Platanus pollen deposition
Aerobiologia
(2000) Airborne birch pollen in Neuchâtel (Switzerland): Onset, peak and daily patterns
Aerobiologia
(2001)- et al.
Predicting the intensity of the birch pollen season
Aerobiologia
(1996)