National survey

A total of 2019 people took part in the surveys. Most of the participants were from 102 bird watching societies, universities and institutes, or relevant organizations with basic background of field work and bird watching. All volunteer observers were trained to identify M. squamatus, especially from Common Merganser (M. merganser) or Red-breasted Merganser (M. serrator). The connection of ecological process is used to define “site”, e.g., a river section between two dams, or a reservoir with closed boundary. In the first year, the sites were based on historical distribution, and stretched to adjacent river sections or reservoirs, with consideration of site condition (hydrology, land cover, anthropogenic disturbance, etc.) during the survey. In the following years, surveyed sites were improved according to the last year’s situation.

The National Wintering Survey of M. squamatus was conducted during the weekends in late December to early January when the wintering population of the duck is mostly stable (Zeng et al. 2015a). Effective monitoring requires that data collection ensures a relatively high probability of detecting the target species (e.g. Conway 2011). Considering the weather conditions and available time of volunteer observers, a strict synchronous survey at national scale is quite difficult due to large numbers of participants and geographic range. Each group could confirm their preferred survey date based on a suggested time periods. All field surveys were conducted on clear days, avoiding snowy, rainy, or strong windy days. Each team was comprised of at least three observers, who walked along the river or travelled by car/boat, and used binoculars to identify and locate M. squamatus. A GPS receiver was used to record the location and as the GPS logged coordinates corresponding to the location of the observers, not the actual position of the ducks, we used GoogleEarth to plot the precise coordinates. The ducks resting on land or foraging in water were also recorded. To avoid repeat counting, only birds flying from one direction (upper or lower reach) were counted. A photo of the species as well as the habitat ensured quality control and review of the data. Although the surveys were designed to target the Scaly-sided Merganser, observers were encouraged to record detections of other species. Supplementary information including disturbance (e.g., number of boats that use the river section during the count period), and other relevant data from volunteers and local people were also collected during the survey.

A datasheet was used with a unified format, including basic information that identifies the observer and describes how the count was conducted (start time, duration, and distance traveled), and information of birds (number, sex ratio, location, behavior), habitats (wetland types, bottom types, water depth, speed) and anthropogenic influence (disturbance, river regulation projects). Each team was required to complete the paper form in the field and input data later into an Excel sheet. The paper datasheet, the e-form, and field photo were all submitted to CBA. Observers were encouraged to enter their data into the pooled web-based biodiversity citizen science database (http://www.birdreport.cn). These data are collected and organized with the same protocol, and reviewed and confirmed by experts aided by field photos provided by volunteers.

Environmental variables

Although many routes meander over 20 km, centroid of each route path was used to characterize site for the geographic analyses (Flather and Sauer 1996; Sauer et al. 2013).

Data analysis

National wintering survey

The three surveys covered 144, 223, 317 sites in 2014, 2015 and 2016 respectively (Fig. 1). These areas included mainstreams and branches of ten major basins (Yangtze River, Yellow River, Huaihe River, Zhujiang River, Lancang River, Liaohe River, Haihe River, Luanhe River, Changhua River and rivers at southeast coast) and adjacent reservoirs. Coastal areas and major artificial rivers such as Beijing-Hangzhou Grand Canal and Rivers of South-to-North Water Diversion Project were also surveyed. The distance from the south to the north is c. 2600 km and east to west is c. 2400 km. In total, 441, 634 and 1138 individuals of M. squamatus were recorded in winters of 2014, 2015 and 2016 respectively, the largest count almost 1/10–1/2 of its population size of 2400–10,000 individuals (Wetlands International 2017).

Species distribution model

Using independent testing data (i.e. the testing data were randomly selected and were not included in model training), we verified the fitted models. The performance measures for model testing did not vary substantially from these for model training (Table 3), suggesting the stable performance of the fitted models.

Distribution range overlap and shift

The predicted distributions of wintering distribution of M. squamatus in mainland China for 2014, 2015, 2016 and a maximum distribution range summed up three SDMs were presented in Fig. 2. Although there were thousands of habitat patches modelled for each year (Table 4), the patches were rather concentrated with the main distributions included in the river system of Dongting Watershed in Hunan, East Qinling mountains in Henan, Dabie Mountain in Anhui, and the river system of Poyang Watershed in Jiangxi (Fig. 2).

Landscape metric	2014 distribution	2015 distribution	2016 distribution	Total distribution
No. of habitat	6226	6017	8660	12,500
Total area (km2)	218,839	165,983	259,863	517,386
Proportion (%)	5.01	3.80	5.95	11.84
Mean patch area (km2)	35.15	27.59	30.01	41.39
Largest patch area (km2)	20,287	14,795	41,894	80,880
Total core area (km2)	106,281	72,600	118,269	271,833
Core area proportion (%)	2.43	1.66	2.71	6.00
Mean core area (km2)	17.07	12.07	13.66	21.75
Max patch core area (km2)	13,903	11,392	31,010	61,170
Habitat cohesion index	9.83	9.82	9.89	9.92

The key landscape metrics of the predicted distribution range was reported in Table 4. Interestingly, although the number of survey sites increased year by year (138, 207 and 294 for 2014, 2015 and 2016, respectively) and more birds were counted, the total number of habitat patches, the size of distribution range, and largest patch area did not increase accordingly (Table 4). The 2016 wintering distribution range was the largest, and the 2015 distribution was the smallest. In the 2014 wintering season, 5.01% of the total land (218,839 km², 6266 habitat patches) might be occupied by M. squamatus (Table 4). In 2015 winter, 165,983 km² (3.80% of the total land) and 6017 patches were predicted to be used by the duck. In 2016 winter, these values were 259,863 km² (total distribution area), 5.95% (percentage of land) and 8660 (number of habitat patches).

Schoener’s D index was 0.96, 0.97, 0.99 between 2014 and 2015, 2014 and 2016, and 2015 and 2016, respectively, which indicated that the wintering range of M. squamatus was highly identical for the 3 years, especially between 2015 and 2016 (D = 0.99). The highly overlapped distribution ranges between years suggested that the wintering range of M. squamatus in mainland China were relatively stable.

Wintering range

Over large spatial extents, range predictions are typically derived from expert knowledge, which are useful at coarse resolution, and suitable for delineating unoccupied regions (Merow et al. 2017). Point records with species distribution models could provide finer-scale occurrence information, which often rely on the availability of a sufficient amount of occurrence and/or abundance data representative of the species’ distribution (Elith and Leathwick 2009). A sufficiently large volume of data with relatively lower per-datum information content can contain more information for broad-scale species distribution estimates than a smaller amount of higher quality data (Hochachka et al. 2012). Normally, established large-scale monitoring programs are limited to the breeding season, and relatively few data are available about bird populations during migration and winter (Munson et al. 2010). The 3-year wintering surveys operated consistently and at broad geographic scale, contributed a great deal of information about the wintering range of M. squamatus.

Prior to the citizen science surveys by CBA, there were several national and regional surveys, notably the National Terrestrial Wildlife Resources Surveys organized by Chinese State Forestry Administration, and regional waterbirds surveys covering the middle and lower Yangtze and coastal regions. These surveys found that the majority of the wintering M. squamatus flocks were located along the lower Yangtze River, especially in Jiangxi Province (He et al. 2002, 2006). Targeted single species surveys were later designed specifically for M. squamatus in Jiangxi Province and southern China by several institutes and research teams, and confirmed that the majority of M. squamatus wintered in Jiangxi Province (Barter et al. 2014). Many studies have been compromised by the logistical or financial challenge of collecting data, and large-scale surveys are most likely not feasible for one individual expert or team (Kelling et al. 2009). Citizen science with scientific research conducted, in whole or in part, by nonprofessional scientists, is an alternative and provides advantages over conventional science (Hand 2010). In addition to the field surveys, geolocators logging (Solovyeva et al. 2012) and satellite tracking devices (Liu et al. 2014) were used to identify the wintering grounds of this species. However, these techniques were limited by either inaccuracy or small sample size, and did not delineate the entire distribution range. The 3-year citizen science surveys found that the wintering range of M. squamatus was much more widely dispersed than previous reported, and Henan, Hunan and Hubei provinces were found to support large populations. The two main sites in Dabie Mountains and East Qinling Mountains provided updated knowledge of wintering distribution of the species in these areas. And the surveys are consistent with a previous study, which used Maxent model based on historical data to predict wintering distributions (Zeng et al. 2015b). Maxent with presence-only data produced a map of occurrence probability, and a threshold was used to transfer the probability to presence/absence. Therefore, the threshold had a great effect on the distribution map. However, the random forest models used observed presence/absence data in the study area and it should be more accurate, while largely depends on the data quality. Both studies with Maxent model and random forest models suggested that rivers in Jiangxi, Hunan and Hubei provinces were highly suitable habitats for the endangered duck (Fig. 3), while the 3-year surveys define a larger wintering range. Some sections of rivers recommended to be systematically surveyed by Zeng et al. (2015b) have been visited during this study; M. squamatus were recorded on Xiangjiang River in Hunan Province, but not on Futun and Nanpu rivers in Fujian Province. More rivers ought to be included for the surveys in the future.

Limit in wintering population estimate

Data-collection methodologies affect the robustness of results (Kelling et al. 2009). When properly designed, carried out and evaluated with appropriate protocols, training and oversight, volunteers can efficiently generate high-quality data equal to those collected by experts. Therefore, citizen science can provide sound science and help solve problems (Danielsen et al. 2014; McKinley et al. 2017). In a case study in Australia, strong correlations in reporting rates of bird species were found between informal area searches in locations selected by volunteers, and similar searches conducted using a formal stratified-random design (Szabo et al. 2012). Although the national-scale survey shows its supremacy on prediction of the wintering distribution of M. squamatus, we have to accept that it presents challenges of observer capability and variability that are not easily addressed (Weir et al. 2005). The small count of M. squamatus in 2014 might be due to first-time effect among observers, who tend to have lower counts in the first year they surveyed (Sauer et al. 1994; Kendall et al. 1996). And the counted number of 1138 in 2016 also indicated a gap between wintering and breeding population. It failed to reveal the real population for the deficiency of spatial coverage and temporal extent. Although the surveys covered the main watersheds, investigation coverage was relatively poor due to low manpower and large geographic area, as well as inaccessibility of some habitats, especially reservoirs. There is an increasing trend of abundance with the number of observers, while the limited data (number of surveys) prevents drawing reliable conclusions. The gap of numbers between field observations and population estimate may possibly be filled in by increased survey effort and coverage of more areas, which evidenced the great contribution of citizen science to the census of cryptic species.

The 3-year surveys are not long enough for accurate estimates of population size and trends. Reliable estimates of population trends will probably require > 5 years (and perhaps as much as 15–20 years) of survey data (Conway and Timmermans 2005). A study in southern Ontario of US used 45 years eBird data to estimate population change of twenty-two species (Walker and Taylor 2017). Using long term data to estimate population trends would be a promising avenue for future work. Long term and more focused surveys linked with the modelled distributions, are needed to accurately estimate the wintering population and its trend.