Invisible biodiversity: extinction debts and colonization credits amongst US


 Habitat change is a major source of biodiversity loss. However, the response of animal communities to land-cover changes is not instantaneous. In fact, extinctions and colonizations may take up a long time before they fully manifest, leading to extinction debts and colonization credits. Current quantifications of debts and credits do not consider multiple land cover types and their directionality of change. Here we quantify the relative contribution of past and present landscapes to the current species richness of 5500 USA bird communities, explicitly measuring the response of biodiversity to increases and decreases of five land covers. We identified extinctions debts across 12% of the USA whilst colonization credits were instead present in 16% of the land area. Future species loss is predicted in the East following recent urbanization, while new species colonizations are expected mainly in the North-West as a result of forests conversion into grasslands. Furthermore, lags in biodiversity response were dependent not only on the type and amount of land cover change, but also on its directionality. Effective conservation actions rely on reliable predictions of future biodiversity. Our results highlight the essential need of considering past landscape changes when setting policy targets aiming at halting biodiversity loss.

15-year period in the contiguous USA and validated our predictions utilizing independent 70 data from a more recent survey. 71 We used species richness data from the North American Breeding Bird Survey (BBS,72 Fig. S1, S2), comprising information on the abundance of 541 bird species across the 73 contiguous USA 15 . Birds are an ideal taxon for analyses of spatial and temporal biodiversity 74 changes because they have long been monitored over broad spatial scales and they are highly 75 sensitive to anthropogenic disturbance 16 . We also sourced high spatial resolution (30m 2 ) land 76 cover data from the National Land Cover Database CONUS products 17 , as well as 77 temperature data (mean across May and July) from the PRISM climate dataset 18 (Fig. S3,78 Table S1). Using these datasets, we developed a generalized mixed effects model (GLMM)  Our analysis revealed for the first time an East-to-West spatial gradient in debts and 89 credits across the contiguous USA (Fig.1). Most of the species losses are predicted in the 90 South East, where up to 11 species are expected to go locally extinct. Strikingly, many areas 91 of debts are localized in metropolitan areas, such as in Atlanta,Orlando,Chicago,92 Indianapolis, St. Louis and Houston. Conversely, predicted colonization credits of up to 10 93 species are largely concentrated in the North West, but West Texas is also expected to gain 94 species. Neglecting such debts and credits could lead to overestimates of the species richness 95 a landscape can support by up to 34%, whereas in other locations to underestimates of up to 96 61%. Overall, 12% of the total contiguous USA is expected to lose species, 16% to gain 97 species, and the remaining 72% is currently in balance (Fig. 1). 98 Our model was able to accurately estimate species richness in 2016 (Pearson 99 correlation test, R = 0.65, df = 5688, p < 0.01, Fig. S4). Importantly, we further validated the 100 model's predictive power with novel bird data from 2018, to confirm that the predicted debts 101 and credits matched recently observed changes in species richness from 2016 to 2018.

102
Without using any land cover change information from the same period, and despite the 103 relatively short time interval (we expect most of these debts and credits will require more 104 than two years before they become fully realized), changes in species richness since 2016 105 have largely been in the direction predicted by our model (Pearson correlation test, R = 0.39, 106 df=5688, p <0.001, Fig. S6).

Fig. 1. Extinction debts and colonization credits across US bird communities. The
109 estimated distribution and magnitude of extinction debts (red) and colonization credits (blue) 110 across the contiguous USA. Accounting for lags in community responses to land cover 111 change, these values were computed as the difference between the long-term predicted 112 species richness for a given location (given enough time to adapt to the land cover change in 113 the landscape) and the predicted number of species in 2016. We estimated that 16% of the 114 contiguous USA land area is, as of 2016, experiencing colonization credits while 12% of it is 115 paying for extinction debts (pie chart).

117 118
The debts and credits identified by our model highlighted substantial lags in the 119 response of biodiversity to land cover change (Fig. 2). The magnitude of these lags varies 120 depending on the type, amount and directionality of land cover change. For instance, a 25% 121 increase or decrease in each land cover type resulted in the current species richness being 122 explained more by the past than by the present land cover (past timepoint contribution always 123 > 0.5, Fig. 2). Amongst the land cover change types, urbanization leads to the strongest lags, 124 with just a 10% increase in urban land leading to a present bird community still being almost 125 completely explained by the past land cover composition (past timepoint contribution = 0.92, Fig. 2A). In comparison, for a 10 % increase in cropland, the species richness in 2016 is 127 largely explained by the present landscape, implying that community response is faster (past 128 timepoint contribution = 0.25, Fig. 2E). The magnitude of the lags also varied according to 129 the directionality of change. For example, losses of forest caused greater lags compared to 130 gains (Fig. 2D), while the opposite is found for cropland (Fig. 2E).  Land cover changes have not been homogenous across the contiguous USA (Fig. 3).

149
For instance, the North West has experienced large-scale forest loss in the 15-year period of 150 our study (Fig. 3B). Here forests have mostly been converted to grasslands ( Fig. 3B and D), 151 including pasture lands. Conversely, urban development has mostly occurred in the East (Fig.   152 3A). We therefore hypothesized that the longitudinal gradient in debts and credits reflects 153 spatial segregation of different types of land cover changes. To test this hypothesis, we modelled the magnitude of the predicted species losses 163 and gains as a function of changes in land covers ( Fig. 4 and Table S3). We found extinction 164 debts to be significantly associated with urban, cropland and forest gains, as well as grassland 165 losses. Reductions in cropland, forest and wetland and gains in wetland and grassland were 166 instead associated to colonization credits. Increases in cropland and urban cover are well 167 known to be associated with species loss 19 . Although it might at first appear counterintuitive 168 that increases in forest are related to future species loss, it is important to recognize that 169 recent forest gains in the USA have been largely due to plantations 20 , which are often 170 species-poor 21 . It is important to note that this present model does not consider community 171 composition and rarity. While loss of wetland seems to lead to colonization credits, this might 172 underlie a disproportional gain of more common regional species compared to a limited loss 173 of keystone waterbirds.  Our analysis emphasizes that lags in the community response are dependent not only 183 on the type and amount of land cover change, but also on its directionality. By accounting for 184 all these aspects in our model, we show a strong spatial heterogeneity in the distribution of 185 future species extinctions and colonizations across a large geographical area. This highlights 186 areas of conservation concern in the South-East of the USA, which have already experienced 187 catastrophic losses of avian diversity over the last 50 years 16 . Our results show that this 188 decline is far from being over and that more avian diversity will be lost if urgent conservation

194
All of the statistical analyses were conducted using the R programming language version high wind and rain (as indicated by the Run Protocol ID field being equal to 1), which could 224 affect bird occurrence and detectability. Following this filtering process, the total number of 225 BBS routes analysed was 1138. For higher precision when inferring the relationship between 226 avian diversity and environmental variables, we subdivided each route into five segments of 227 equal length, consisting of 10 count locations each. This approach was motivated by the need 228 to more closely associate bird communities with the land cover composition in the area in 229 which they are found. 230 We assumed the effect of observer experience on the biodiversity metrics to be 231 negligible, based on previous studies using breeding bird survey data that found no consistent difference between the ability of experienced and novice recorders to detect species 26,27 . 233 Further, we did not model detectability issues associated with traffic noise and disturbance, 234 because previous studies have found no evidence for such effects 28,29 .

235
Following these procedures, our final dataset included species richness and evenness 236 data for 1138 routes (Fig. S1), each divided into five segments, giving a total of 5690 237 observational units (that we refer to as "segments"). Alpha diversity measures of species 238 richness and Pielou's evenness index 30 were calculated for each observational unit for each 239 of the two timepoints.   277 We developed a bespoke statistical model that conceptualised the problem of extinction debt 278 and colonisation credit by combining the two concepts: (1) the equilibrium of avian 279 communities in a given landscape composition and (2) the lagged response in the species 280 diversity of a given landscape due to past land cover changes (i.e. a system which is 281 approaching equilibrium). We reasoned that, given enough time, and with no further changes in land cover, species richness at a given location would equilibrate. The equilibrium 283 distribution of species richness emerges from the effect of different land cover types in 284 encouraging or impeding the recruitment and survival of particular species and functional 285 groups. We did not model these ecological mechanisms directly, but instead expressed the 286 equilibrium of species richness, and the rate of approach to this equilibrium, as empirical 287 functions of environmental covariates. It is important to keep in mind that during a finite time 288 interval following environmental change, it is likely that our observations of species richness 289 represent a system in a transitory state towards its new equilibrium. It has to be noted that a We assume that, under landscape change, the system is in a state of flux and that survey Here, the function represents the equilibrium distribution for different configurations of the Equilibrium model 325 We defined the equilibrium distribution of species richness at a given timepoint as a function

400
We ran 4 chains, sampling for 10 4 iterations with a burn-in period equal to half the 401 number of iterations. The number of iterations was sufficient to achieve chain convergence.

402
The JAGS sampling was run on four parallel threads on a multi-core Intel i7 -8750H 403 processor with a maximum clock speed of 4.1 GHz. bounded between zero and one to act as a weighted proportion when quantifying the lag 409 generated by each environmental variable change.