Introduction

Urban environments are continuously reshaping natural landscapes, subsequently influencing the behaviors of many wildlife species (Turner et al. 2005). Among these species, avian populations, particularly rooks (Corvus frugilegus), have showcased a profound capacity to adapt and thrive in urban environments. These birds are attracted to urban locales for reasons such as plentiful food supplies, fewer predators, and congenial microclimates (Marzluff et al. 2001).

The habitat associations of rooks in urban environments have been the subject of much research. Earlier studies emphasized the rook's propensity towards human-made amenities, spotlighting their dependency on easily accessible food sources (Byrkjedal et al. 2012). In a similar vein, research has also highlighted the rooks' affinity for the controlled climates of cities, valuing the protection from harsh winds and the milder temperatures they offer. Moreover, green spaces (Griffin et al. 2000; Clewley et al. 2016) play a critical role in these urban settings. These areas, blending the natural with the urban, serve as sanctuaries, enhancing not just the avian biodiversity but also suggesting a nuanced relationship between the rooks and various urban land use patterns.

Previously, researchers relied on manual collection of localized or historical data (Whitfield 2001). However, extrapolating these findings to larger areas is often restricted (Sanz 2002; Crick 2004). Ecological research often hinges on our grasp of diverse geographical concepts and spatiotemporal patterns, and large-scale studies can be logistically daunting (Dickinson et al. 2010). Citizen science offers a solution, aiding not only in broad-scale species conservation but also in identifying and monitoring invasive species over vast areas (Gallo and Waitt 2011). For extensive studies like avian research, citizen science helps bypass spatial constraints, thereby reducing costs (Devictor et al. 2010; McCaffrey et al. 2005; Braschler 2009).

Citizen science projects actively engage the public in collecting data for scientific research or policymaking. Empowered by technology, modern citizen science transcends many traditional barriers (Silvertown 2009). It has proven invaluable in locating rare organisms, monitoring species invasions, and tracking bird populations (Hochachka et al. 1999), presenting a robust approach to address environmental challenges in conservation biology, natural resource management, and environmental protection.

In this context, we examined the rook, an invasive species in South Korea now deemed harmful due to its nocturnal urban invasions and agricultural damage (Article 4 of the Enforcement Rules of the Wildlife Protection and Management Act 2018). Altered by environmental changes, the migratory patterns of rooks have shifted (Gordienko and Sokolov 2006). Since 2016, rooks have been observed seasonally in cities like Suwon from November to March, now spanning a vast geographical scope in and around the city.

To chart their movements, we initiated a three-month citizen science campaign from December 2021 to March 2022, gathering over 6,314 location data points on rooks. With the aid of participant-collected data, our study was structured around distinct yet interrelated objectives. Our primary aim was to (1) delineate the spatial locations of dense rook congregations within varied land use types across the city, analyzed in 3-h segments. By modeling their activity in 3-h segments, we unraveled the route by which rooks transitioned from rural areas—predominantly farmlands serving as their foraging grounds—to urban zones that provided their roosting areas. Such an approach is pivotal as it offers insights into the intricate dynamics governing rook movement amidst varied urban landscapes. Our secondary objective was to (2) discern the paramount environmental factors that influence rook distribution during both daytime and nighttime periods. By shedding light on the rook's habitat associations, we aspire to deepen our understanding of their urban ecology and the specific urban attributes that attract them.

Through these objectives, our research aims to navigate beyond the constraints of standard avian methodologies. We endeavor to provide a comprehensive lens into rook activities and behaviors within the city, offering insights for future urban planning and effective bird management.

Materials and methods

Study area

We conducted our study in the city of Suwon, Gyeonggi Province, South Korea (37° 17′ 28″ N; 127° 0′ 32″ E). Suwon comprises four administrative divisions: Paldal-gu, Jangan-gu, Yeongtong-gu, and Gwonseon-gu, which include urban areas and surrounding agricultural and natural lands (Fig. 1). Rooks arrive in Suwon—the third most densely populated city in Gyeonggi Province with a population of 1.18 million people—in November, when the average temperature is 7.9°C and annual precipitation is 104.2 mm (Suwon 2022).

Fig. 1
figure 1

Occurrence data of Corvus frugilegus acquired by citizen science throughout the 4 districts of Suwon. a sample picture of photo acquired by citizen science data of rook’s departure from their foraging area located in Gwonseon district. b sample picture of photo acquired by citizen science data of rooks roosting on the utility pole in the urban area located in Paldal district

Full size image

The areas of each district are Yeongtong: 2.75 ha; Paldal: 1.29 ha; Jangan: 3.33 ha; and Gwonseon: 4.71 ha. In the study area, Gwonseon District exhibits the largest expanse of agricultural land, encompassing 0.97 hectares. Residential zones are most prevalent in Yeongtong and Paldal Districts, with areas of 0.64 hectares and 0.31 hectares, respectively. Jangan District, on the other hand, features the most extensive forested land, covering 1.51 hectares (Table 1).

Table 1 The classification of land use and landcover types of Suwon City and the total area of the land use of the biotope for Gwonseon-gu, Yeongtong-gu, Jangan-gu, and Paldal-gu (Suwon 2019)
Full size table

Study species

The breeding season for rooks primarily spans May through June, with key breeding areas located in the Amur region, northeastern and southern China, and Mongolia. During the winter months, from November to March, rooks migrate to Korea, Japan, Taiwan, and eastern China. They frequently traverse Korea during the spring and autumn, acting as migratory birds in the country's southern regions.

Rooks favor habitats such as flatlands, open terrain, and forests in proximity to their roosting locations. As previously highlighted, rooks are classified as a harmful species under Article 4 of the Enforcement Rules of the Wildlife Protection and Management Act due to their tendency to cause prolonged crop damage in large groups. These birds typically gather in significant numbers, often in the hundreds or thousands (Madge and Burn 1994), to roost post-sunset, commencing their foraging activities after sunrise (Hubálek 2017). Consequently, we defined the preferred habitats of rooks based on the distinction between sunset and sunrise, utilizing data from December 2020 to February 2021 provided by the Meteorological Administration in Suwon.

Acquisition of citizen science data

Data on rook presence was collected using the smartphone app CADA v.1.0.15 (Kim 2020). We adopted a crowdsourcing approach—the most elementary among the four levels of citizen science as described by Shum et al. (2012): level 1: crowdsourcing, level 2: distributed intelligence, level 3: participatory science, and level 4: extreme citizen science. To motivate participation, citizens were financially rewarded 0.35 USD for recording sightings of Corvus frugilegus using their smartphones.

No training sessions were conducted prior to the campaign, with the intent to maximize participation. Instead of requiring participants to submit a predetermined number of photographs at specific times or locations, observations were made at the citizens' convenience. This was to ensure wide coverage without imposing rigid constraints on participants. To guide participants, the app incorporated a detailed description explaining the campaign's objective, given the escalating invasions of rooks in urban areas and the consequent disturbances they caused to the residents of Suwon. Visual guides were provided to help differentiate the rook from the raven based on physical appearance.

The data acquired included the participant's address, age, geolocation (longitude and latitude), photograph of the rook, time of day, and the date of the observation. Once a participant's smartphone was registered, our database logged the location data along with the photos. We filtered out any instances where a single individual reported multiple observations simultaneously throughout the day. Moreover, the application was configured to prevent participants from capturing multiple photographs at the same location; it disallows a user from submitting more than one entry within a 200-m radius of a previously submitted sighting for a duration of 20 min. This precautionary measure was essential to avoid over-representation of data from any single point and to minimize the risk of duplicating records within a short timeframe.

Crucially, the app was enhanced with a machine-learning algorithm designed to filter out fraudulent photos not featuring rooks. To ensure the integrity of the data, two researchers were also involved in the final scrutiny of all collected images. Thus, by bolstering validation steps, we could confidently rely on the broader public without compromising the quality of data collected.

In total, we amassed 6,314 rook location records across Suwon's four districts through crowdsourcing. After refining the dataset, we used 4,522 valid location points. Of these, 2,528 points (55%) were from Gwonseon, 1,023 (22%) from Yeongtong, 963 (22%) from Paldal, and 9 (0%) from Jangan.

Acquisition of environmental variables

We referenced previous studies to select environmental variables favored by rooks appearing in urban areas; these were primarily artificial influences and structures such as food supply, mild temperatures, wind shields, and green spaces (Byrkjedal et al. 2012; Ciah and Fröhlich 2017; Clewley et al. 2016; Griffin et al. 2000). Among them, we combined 15 types of land use from the biotope map and five other variables were extracted—elevation, Euclidean distance from buildings with three different classes of floor numbers (1st ~ 5th, 6th ~ 20th, 30th and above), Euclidean distance from agricultural land, Euclidean distance from a utility pole, and Euclidean distance from streetlamp—which we selected as rook-favored variables (Table 2). As interest in landcover and land use maps increases as a practical measure that can consider specific natural environments and ecosystems in various urban development plans, the Ministry of the Environment of South Korea has prepared guidelines for creating “biotope” maps and distributed them to each local government.

Table 2 Predictor variables used in the study
Full size table

The biotope map of Suwon city (Suwon 2019) was produced using the classification of the following: land use, land cover, green cover ratio, vegetation rate, and landscape data. Within the classifications, in this study, we utilized the land use and the height (number of floors) of the buildings provided by Suwon’s biotope map, which has a native resolution of 5 m2. The distance from the utility pole was extracted by eliminating the areas of poles constructed underground provided by the Korea Electric Power Corporation. After extraction, the area poles were placed on top of transportation facilities at intervals of 30 m. Altitude was extracted with a resolution of 10 × 10 m using the digital topographic map provided by the National Territory Information Platform. The Euclidean distance from streetlamps was applied after acquiring the location data of streetlights in Suwon provided in 2021 from the public data portal (www.data.go.kr). We weighed these variables over 12-h periods at 3-h intervals according to our MaxENT results.

Processing maximum entropy model

We employed MaxENT version 3.4.4 to predict the preferred habitats of rooks based on their occurrence data. Recognizing the potential limitations posed by small sample sizes, we employed 10 replicates within MaxENT. This approach involves running the model multiple times (in this case, ten times) using different random subsets of the data for training and testing. By doing so, it generates an ensemble of models rather than relying on a single iteration. As a result, it provides a more robust and reliable output, as any variance in individual runs due to potential sample biases gets averaged out over the multiple iterations. This is especially crucial when dealing with limited data points, as it reduces the chance of overfitting to any specific subset of the data (Phillips and Dudík 2008). In line with best practices, we utilized k-fold cross-validation (k = 10), a method shown to provide a good balance between variance and bias in many empirical studies (Kohavi 1995). We also designated a random test percentage of 25%, a frequently cited and practiced split ratio that typically offers a good balance for model training and validation (Hastie et al. 2009).

Pearson’s correlation analysis was performed to screen for environmental variables with high multicollinearity; however, none were identified with r = 0.8 threshold (Fig. S1). To adjust for uneven sampling bias, we transformed a bias file—capturing the density of Suwon’s population aged 18 and above within 5 × 5 m pixels—into a normal distribution. This ensured enhanced relative gain contributions and weighted areas with lower population densities appropriately (Kramer‐Schadt et al. 2013).

The movements of rooks to foraging and roosting zones are influenced by light intensity (Swingland 1976). Data from the Meteorological Administration in Suwon, spanning December 2020 to February 2021, indicated sunrise and sunset times as 7:35 and 17:41, respectively (Table S1). Considering civil twilight, which persists for 20–30 min post-sunset and pre-sunrise, noticeable changes in light intensity commence around 08:00 and 18:00 (Andre and Owens 2001). After determining the daylight and nighttime weightings of the rook predictor variables, we applied these to the presence records. These records were segmented into eight 3-h intervals, factoring in sunrise and sunset durations, allowing us to monitor the rook population's spatial distribution throughout the daily cycle (Table 3).

Table 3 Sum of photographs submitted by citizen scientists every 3 h documenting rook locations
Full size table

Results

Model performance

MaxENT AUC values of ≥ 0.70 are considered indicative of reliable models (Phillips and Dudík 2008). The AUC values for rook distribution during the periods after sunrise (08:00–18:00) and after sunset (18:00–08:00) were both equal to or greater than 0.87. These values suggest that the models possess robust explanatory power (Fig. 2)

Fig. 2
figure 2

MaxENT results of Corvus frugilegus throughout Suwon after sunrise (08:00 ~ 18:00) and sunset (18:00–08:00): a sample crowd-sourced photograph collected on December 25, 2020 at 14:33 in the Gwonseon district (longitude: 127.0281729, latitude: 37.2678822), and b sample crowd-sourced photograph collected on January 3, 2021 at 07:23 in the Paldal district(longitude: 126.9543047, latitude: 37.2791152)

Full size image

Species distribution model results

The relative contributions of the top three environmental variables to the MaxENT model for rooks in the time zone after sunrise were Euclidean distance from agricultural land (31.5%), elevation (29.1%), and Euclidean distance from streetlamps (12.8%), Euclidean distance from buildings 21 floors and above tall (10.9%) (Table 4). According to the Euclidean distance from agricultural land, the probability of rook appearance was above 50% between the range of 0–100 m. According to elevation, the probability of rook appearance was 50%–60% or more within a low range between 25 and 30 m. The probability Euclidean distance form streetlamp and buildings 21 floors and above dropped sharply beyond 500m (Fig. 3a).

Table 4 MaxENT percentage contribution and permutation importance for Corvus frugilegus after sunrise and sunset
Full size table
Fig. 3
figure 3

The 3 main response curves of environmental variables after sunrise and sunset for Corvus frugilegus habitat distribution model after sunrise (refer to Table 2 for land use type codes)

Full size image

The relative contribution rates of the top four environmental variables to the MaxENT model for rooks in the time zone after sunset were Euclidean distance from buildings 6–20 floors tall (20.7%); land use (19.4%); elevation (19%); and Euclidean distance from a utility pole (14.5%) (Table 4). The environmental variable with the highest gain when used in isolation was land use. The environmental variable that, when omitted, most decreased the gain was the Euclidean distance from a utility pole. Among the 15 types of land use, mixed complex areas and the areas of lakes and wetlands contributed 50% or more. The probability of appearance within 0–10 m of a pole was found to be between 55%–60% and dropped sharply to less than 50% beyond 20 m (Fig. 3b).

Differentiation in the assemblage locations by ‘real-time’ tracking using citizen science

Daily citizen science data from November to February indicated that rooks started to move to their foraging areas in the Gwonseon district after 06:00. In this region in western Suwon, agricultural land comprises 20.7% (Fig. 4c). After 09:00 presence probabilities above 0.8 were concentrated mainly in these agricultural lands where the rooks settled to forage until 18:00 (Fig. 4d–f). Movement to the urban areas could be perceived after 18:00 in the Paldal and Yeongtong districts, located in eastern Suwon, where mixed residential and business land use comprises over 20% (Fig. 4g).

Fig. 4
figure 4

MaxENT-derived three-h interval distribution of Corvus frugilegus a 24:00–3:00, b 03:00–06:0, c 06:00–09:00, d 09:00–12:00, e 12:00–15:00, f 15:00–18:00, g 18:00–21:00, h 21:00–24:00 (AUC < 0.85)

Full size image

Discussion

In this study we explore the daily habits and congregation patterns of rooks within urban environments, offering a nuanced understanding of their behavioral patterns. Notably, the rooks’ movement congregation relied highly upon land use. During the post-sunrise hours, our findings consistently revealed that rooks prefer to gather within 50–100 m of agricultural areas. This clear-cut preference echoes established research highlighting their foraging tendencies, which are significantly influenced by resource abundance and size in such territories (Griffin et al. 2000).

The role of artificial lighting, specifically its presence around streetlamps, cannot be understated. Illuminated areas seem to present an irresistible lure for rooks. The reasons are multifaceted: the increased visibility that such lighting provides allows for more extended foraging sessions, thereby maximizing their chances of finding food, especially during hours when other predators or competitors might be less active (Clewley et al. 2016). Paldal, located in southeast Suwon, is a case in point. Its prominent residential character coupled with interspersed commercial areas made it a consistent hotspot for rook sightings (Fig. 2) (Dwyer and Doloughan 2014). Another interesting finding from our research is the role of utility poles for these birds, notably in locales dominated by open spaces with sparse tree cover. In such settings, utility poles supersede even streetlamps as preferred congregation points (Dwyer and Doloughan 2014). The structural design of urban spaces also plays a pivotal role. Buildings of medium height, specifically those between 6 to 20 stories, offer rooks shelter from gusty winds, making these areas attractive for roosting and shelter (Gorenzel and Salmon 1995).

An intricate dance of movement is observable when analyzing rook behavior over a 24-h period. The waning daylight serves as a signal for the birds, directing them to the agricultural fringes to the west of Suwon (Fig. 4). Seasonal variations, particularly the shortened daylight during winter months, amplify these patterns, with light intensity emerging as a predominant factor dictating their movement (Andre and Owens 2001).

Our study delineates the daily congregation patterns of rooks derived from non-invasively procured citizen science data. This knowledge can underpin strategies to either curtail or harmoniously coexist with urban-invading corvids. Presently, Suwon is trialing the laser pointer technique to deter rooks from city roosting—an approach similarly employed in Sunnyvale, California to manage rook invasions (Kim 2019; Picon 2022). This method relies on laser beams mimicking the eye color (green) of rook predators to scatter them. In Rochester, Minnesota, USA, a composite technique featuring laser pointers, recorded crow calls, and starter guns has been deployed to discourage American crows (Corvus brachyrhynchos) from downtown roosting (Richert 2019). Other locales, including Sacramento and Hanford in California, have introduced trained Harris’s hawks (Parabuteo unicinctus) guided by lasers to disband crow congregations. As we expand our scope beyond Suwon City, leveraging citizen science tools akin to eBird will prove invaluable in discerning rook habitat predilections. By judiciously analyzing time-segmented citizen science datasets, we aspire to evaluate both immediate and enduring outcomes of laser interventions and to monitor rook adaptive responses to varied laser intervention frequencies.