Shifts in bird ranges and conservation priorities in China under climate change

process, collected the area of each DRM and the contribution value of each variable for each species. We excluded 69 species with few localities’ density by the criteria that the number of points less than 1 per 3˚ � 3˚ (about 1 � 105 km2) in the DRMs, and excluded variables that had no contribution in the model for each species [64]. Finally, we used 1,042 species with 161,004 localities for second round simulation. In the second round, we used different environment variables for each species (details see S1 Table). Background selection method is also used for each species for sampling bias correc- tion, which had been proven useful in data-sparse situations [46, 65]. We collected province- level distribution records of each species [66] as the background selection sampling in the MaxEnt, aiming to create statistically reliable results [67]. Other settings are the same as the first round. The code for background selection and variables selection shown in S2 File. The mean training AUC value for 1,042 species was 0.908±0.055, and the mean test AUC value was 0.861±0.085 (see S1 Table). We overlaid all DRMs from the second-round simulation of 1,042 species to produce spe- cies richness maps for the current scenario and the 2070 scenario. We projected these richness maps on the DEM as a 3D model by ArcMap to show the change related to the terrain. To measure the effect of climate change on the shift direction of the distribution range, we identi- fied the median center point (the location that minimizes the overall Euclidean distance to the

maps) of each range map under different RCPs. Then, we calculated the displacement vectors of the distribution range under different scenarios. 2.3 Conservation priority modeling We used Zonation v4.0 as the planning tool to generate conservation priority maps for differ- ent RCPs [28]. Zonation is capable of large-scale, high-resolution spatial conservation prioriti- zation with big data, and it can achieve a higher conservation value aggregated across biodiversity features and space than other planning software [68]. All original Cloglog output files of species for each RCP scenario were applied as species layers in the tool to retain infor- mation for the distribution maps. We used the percentage change value of the distribution range of each species as the weight value in Zonation to indicate the impact of climate change (species with high habitat loss were given high weight values, and species with high habitat increases were given low weight values). We added an extra 0.5 to the weight value if the spe- cies is endemic to China (for details, see S1 Table). Considering the conservation value of each species, we used the basic core-area zonation (CAZ) result for cell removal rule and set the wrap factor to 1,000; thus, we were able to identify areas with high occurrence levels for a single rare/highly weighted species [69]. We produced two model scenarios for each RCP to meet different conservation planning target: 1) Random ranked scenario that only considering species layers and their weights; 2)

NNRs masked scenario that aim to map priorities out of NNR regions by apply the additional NNR layer as the mask layer in the Zonation. 3. Results 3.1 Identifying the species with the range reduction and expansion Among the 1,042 studied species, approximately 25% of bird species may lose part of their range, and approximately 75% of may expand their distribution range due to the climate change by 2070 (Table 1) under RCP 2.6 or RCP 8.5. The histogram of range change for all bird species (Fig 1) shows that in general, species will have a broader range under RCP 8.5 than under RCP 2.6 (Z = -4.49, P-value = 3.5E-6), although more range may be lost under RCP 8.5 for the 30% species that reduce their ranges (Table 1). For RCP 2.6, we found that 30 PLOS ONE Bird range changes call for innovative conservation in China PLOS ONE | https://doi.org/10.1371/journal.pone.0240225 October 8, 2020 5 / 20