Opportunities for prioritizing and expanding conservation enterprise in India using carnivores as flagships

Conservation interventions in developing countries are frequently thwarted by socio-economic agendas, severely limiting the scope and rigor of biodiversity and habitat conservation. Very few ecological assessments incorporate human interests in conservation prioritization, creating asynchrony between planning and implementation. For conservation actions to be logistically feasible, multiple criteria including ecological, social, economic and administrative aspects must be considered. Understanding how these different dimensions interact spatially is also important for gauging the potential for conservation success. Here, we use a guild of select mammalian carnivores (wild canids and hyenas) in India to (i) generate distribution maps at the spatial scale of administrative sub-districts, that is relevant to management, (ii) examine ecological, social and biogeographic factors associated with their distribution, quantify key threats, and identify areas important for their conservation, (iii) use prioritization tools for balancing habitat conservation, human needs and economic growth, and (iv) evaluate the spatial congruence between areas with high conservation potential, and areas currently in focus for protection efforts, conservation investments, and infrastructure development. We find that the current Protected Area system does not adequately cover or represent diverse habitats, that there is immense potential for States to increase financial investments towards alternative conservation strategies, and, most infrastructure projects may be jeopardizing important carnivore habitats. Our framework allowed for identifying locations where conservation investments would lead to the highest dividends for flagship carnivores and associated species across habitats. We make a case for re-evaluating how large-scale prioritization assessments are made, and for broadening the purview of conservation policies in India and other developing countries.


154
Carnivore presence records: We collected carnivore distribution data in three phases. In phase 1, 155 we used citizen-science data from countrywide web-based surveys for three months in 2018 (see 156 SI Appendix, section 1 for detailed survey protocol). In addition to our own survey, we also 157 included carefully vetted records from other citizen-science portals (India Biodiversity Portal: 158 http://www.indiabiodiversity.org; iNaturalist: https://www.inaturalist.org). In phase 2, we 159 extracted data from wildlife, nature and photography pages on social media, online wildlife presence record was then assigned to an administrative sub-district within the species' plausible 174 range. We used an occupancy modelling framework that accounted for partial detectability to

183
For analyses, we collapsed data from 48 months into four 12-month blocks, resulting in one temporal replicate per year. At the sub-district scale (mean = 1400 sq. km; range = 3 -51,000 sq. 185 km), we assumed that the distribution status of the focal species remained stable during the four- 186 year period. For each species, we built a set of candidate models with singular and additive 187 effects of a set of explanatory variables, based on specific a priori predictions (SI Appendix, 188 section 2). We fit occupancy models to detection/non-detection data using package 'unmarked' 189 in program R v3.4.1 (R Core Team 2018).  Table S1. Schedule 3). Scores for area of occupancy were based on the predicted proportion of habitat that 211 a species currently occupies within its plausible range in India (scores: 1 to 10; 1 = species 212 occupies 0-10% of plausible range; 10 = species occupies 90-100% of plausible range). Through 213 surveys of field experts, we elicited scores for perceived population trend (1 = decreasing; 3 = 214 stable; 5 = increasing). We also obtained scores for 'level of threat' attributed to habitat loss, prey decline, direct persecution, road-related mortality, illegal trade and negative interactions 216 with free-ranging dogs (1 = fatal; 2 = high; 3 = medium; 4 = low; 5 = not a threat). The total 217 scores for each category were averages from scores of individual experts. The final conservation 218 score was obtained by summing across all categories, weighting area of occupancy at 0.5, 219 protection status at 0.3 and expert responses on threats at 0.2. The unequal weighting is because 220 area of occupancy is a quantitatively estimated metric, protection status is derived from global 221 datasets but without the same analytical rigor, and threat information is based on expert opinions 222 (which could be anecdotal, or limited to insights from local/regional experience). We re-scaled 223 the sum out of 100; a lower conservation score implied a higher threatened status (within India).  (Table S2). We  Fig. S1. In addition to 243 these, we also included human poverty index and projected human population for 2020, both of 244 which qualified as "costs" in our assessment. Our rationale was that sub-districts with larger 245 human populations and higher poverty require focus on economic growth and infrastructure development and should therefore be of lower priority for conservation (see Table S2 for 247 details). Following exploratory runs with different combinations of settings for Boundary Length 248 Penalty (BLP) and Warp Factor (WP), we set BLP at 0 and WP at 100 as a trade-off between 249 computation time and reliability of spatial maps.

261
We collated a total of 4437 presence records of the target species across three phases (Table 1).

262
Model-averaged occupancy estimates ranged from 0.21 (SE 0.02) for desert fox to 0.75 (SE 263 0.002) for golden jackal (Table 1; Fig. 2). We could not formally analyze or generate estimates 264 for Tibetan fox because the data were too sparse. Tibetan wolf occupied the least overall area 265 (152,180 sq. km) and golden jackal was the most widely distributed species (2,259,361 sq. km).

266
Combining protection status, distribution extent, population status and anthropogenic threats (the 267 latter two elicited from surveys of field experts in India; n = 45), golden jackal and red fox 268 scored the highest, while Indian fox and striped hyena scored the least (Table 1).

270
Carnivore-habitat associations generally agreed with our predictions. Forest cover and large wild 271 prey were important for dhole and Indian wolf ( Table 2). Extent of grasslands, scrublands,  (Table 2). Interestingly, high probabilities of golden jackal occurrence were clustered around 280 large settlements with high densities of humans and free-ranging dogs. This may be attributed to 281 high availability of provisioned food resources, and their ability to adapt to human-modified 282 areas. While these results were based on singular covariate effects, the final occupancy estimates 283 for all species were derived from averaging across models with comparable statistical support 284 (based on AICc scores and weights; SI Appendix section 2, Table S3). We generated both, 285 distribution maps at the sub-district level (Fig. 2), and maps depicting occurrence probabilities 286 within plausible habitats (Fig. S2).

288
Based on results from the prioritization analysis, we chose sub-districts that accounted for the top 289 30% priority scores (n = 703) as the most important areas for conserving wild canids, hyenas, 290 and their habitats (Fig. 3). The high priority areas-which we term 'Canid Conservation Units' 291 (CCUs)-cover 1,475,558 sq. km, of which 545,070 sq. km (37%) constitute focal habitats (i. e., 292 habitats deemed important for the focal species; Fig. 3). We further classified these areas into tier 293 1, tier 2 and tier 3, based on level of priority (top 10, 20, or 30% scores). Protected Areas 294 currently cover around 26% of key habitats within CCUs (Fig. 3). Summed conservation ranks of 295 CCUs showed that 12 States fared relatively better than India's 17 other States (Fig. 4). Of these, Pradesh have high conservation scores, but will need to invest higher monetary resources to be 301 able to work towards conserving CCUs in their States (Fig. 4).    Our focal species align themselves mid-way between obscurity and ubiquity, i.e., they are neither 387 so rare that citizens are unaware of their existence, nor are they so abundant that calling for their 388 conservation seems futile. And since these carnivores also occupy mid-high trophic levels, we 389 believe they can serve as ideal flagships across important habitats (forests, alpine and arid/semi- Protected Areas. Certainly, executing these approaches would entail choosing habitat-specific 416 flagships relevant to ecological, regional and social contexts, and developing monitoring 417 techniques that can be co-opted by conservation managers and planners. In sum, our study 418 represents an 'ensemble approach', combining a range of data sources, methods and analytical 419 frameworks, and incorporates human population, poverty, infrastructure projects, and 420 administrative potential and likelihood in conservation prioritization. We believe these are 421 crucial for critically evaluating large-scale prioritization assessments, and for rethinking country-  Table 1. Summary of data records collated from three survey phases, estimated occupancy 708 (standard errors in parentheses), extent of occurrence in India and conservation score for the 709 focal carnivore species   Table 1.      Table 2. Estimated slope coefficients from singular covariate models for each of the carnivore species (standard errors in parentheses). Abbreviations: fcov-combined forest cover; dfor-tropical dry forests; tfor-temperate forests; gsor-grasslands, scrublands, open (barren) habitats and ravines; ohab-open (barren) habitats; agri-agricultural areas; prod-production agroforests; hamt-high altitude mountains; habtcombined area of plausible habitats (species-specific); rock-rocky outcrops and escarpments; pptn-annual precipitation; rugg-terrain ruggedness; prey-wild prey index; elev-elevation; catl-cattle population; shot-sheep and goat population; lstk-livestock (cattle, sheep and goat) population; hpop-human population; dogs-free-ranging dog population; sett-human settlements; ifra-density of roads and railways; resv-extent of Protected Areas