Poor‐quality monitoring data underestimate the impact of Australia's megafires on a critically endangered songbird

Catastrophic events such as south‐eastern Australia's 2019/20 megafires are predicted to increase in frequency and severity under climate change. Rapid, well‐informed conservation prioritization will become increasingly crucial for minimizing biodiversity losses resulting from megafires. However, such assessments are susceptible to bias, because the quality of monitoring data underpinning knowledge of species' distributions is highly variable and they fail to account for differences in life history traits such as aggregative breeding. We aimed to assess how impact estimates of the 2019/20 megafires on the critically endangered regent honeyeater Anthochaera phrygia varied according to the quality of available input data and assessment methodology.

burned and more than a billion animals are estimated to have perished in eastern Australia (Wintle et al., 2020). Initial assessments estimate 327 threatened species have been impacted by the fires, defined as taxa having >10% of their known or predicted range burnt. These figures include 31 species listed as "critically endangered" under federal biodiversity legislation (Commonwealth of Australia, 2020aAustralia, , 2020b. The development of rapid and strategic conservation responses to such unprecedented events will play an increasingly crucial role in limiting future global biodiversity loss (Wintle et al., 2020).
Rapidly and accurately assessing the relative impact of extreme events on threatened species is crucial not only for prioritizing emergency conservation investment (Wintle et al., 2020) but also for informing how, where and when these funds should be utilized to maximize conservation returns (Bottrill et al., 2008). In this instance, it is desirable to devise a quick, simple and repeatable metric of relative impact assessment that decision-makers can use to help identify those taxa in most need of conservation effort (Ward et al., 2020).
In other words, how is conservation prioritization in response to Australia's bushfires best implemented, given the unprecedented scale of the fires and the short time frame required to commence on-ground action?
Determining how to rapidly and accurately prioritize conservation funds across entire ecosystems is challenging. In such circumstances, an attractive option is to determine a species' predicted distribution using area of occupancy (AOO), extent of occurrence (EOO) or species distribution models (SDMs), overlay remotely sensed fire severity mapping and calculate the percentage of a species' range that has been fire-affected (Ward et al., 2020). The principal advantage of utilizing these methods as a first-phase assessment is that they can be implemented using data that already exist for the vast majority of affected species. Consequently, the Australian Government have adopted these methods, alongside consideration of pre-fire imperilment and other threats, to help guide the conservation response to the bushfires (Commonwealth of Australia, 2020a, 2020b; Legge et al., 2020).
Despite practical advantages, using species distribution data to assess bushfire impacts may be subject to systematic biases . The quantity and quality of baseline monitoring data varies substantially across fire-affected species, but is poor for the majority of taxa Scheele et al., 2019).
Interspecific differences in life history traits such as ecological niche, dispersal capacity, sociality and competitive ability (i.e. body size) will also affect species' capacity to recover from the impacts of bushfire, both with and without conservation assistance Woinarski & Recher, 1997). Highly mobile species, whose functional habitat for survival and reproduction in any given year may be only a tiny proportion of their entire range (Runge et al., 2014;Webb et al. 2017), will be particularly at risk if bushfire affects those critical areas (at least until habitats in such areas have recovered). If high-quality monitoring data on breeding activity for such species are not available, bushfire impacts may be underestimated using species distribution methods. Underestimating bushfire impact due to variability in the quality of available monitoring data could cause prioritization decisions to misallocate conservation resources, potentially overlooking actions that could help prevent extinction (Woinarski et al., 2017).
Here, we assess how estimates of fire impact vary with the quality of available monitoring data for a critically endangered, nomadic songbird, the regent honeyeater. Specifically, we assessed fire impact using six monitoring datasets that differ in quality and temporal span: EOO, AOO, a MaxEnt SDM, public sightings, targeted monitoring and nest monitoring. We predicted that fire impact would be underestimated using methods that draw on datasets with longer temporal spans and opportunistic monitoring data such as public sightings. These datasets may fail to account for habitat functionality (e.g. annual rainfall and dynamic resource distribution) and species-specific life history traits (e.g. flocking and group nesting).
Consequently, they contain biases in survey effort that could result in overestimation in habitat availability or the size of the species' contemporary range.

| Study species
The regent honeyeater is endemic to Australia's eastern seaboard and was abundant and widespread as recently as 60 years ago (Franklin et al., 1989). Extensive land clearing has led to a rapid population decline, with fewer than 350 individuals estimated to persist in the wild today in a range exceeding 600,000 km 2 from Victoria to southern Queensland . Regent honeyeaters nest primarily in association with flowering events in a small number of Eucalyptus tree species, which show very high spatio-temporal variation in flowering phenology (Birtchnell & Gibson, 2006;Franklin et al., 1989 et al., 2017). The entire population represents a single genetic management unit, but the core remaining population persists within the greater Blue Mountains area of central/eastern New South Wales (Crates, Olah, et al., 2019;.

| Fire severity mapping
We used the Australian Google Earth Engine Burnt Area Map (GEEBAM, Commonwealth of Australia, 2020a, 2020b), derived remotely from Sentinel 2 satellite imagery. GEEBAM calculates the difference between pre-fire (April 2018 to April 2019) and post-fire (November 2019 to May 2020) normalized burn ratio using nearinfrared and shortwave infrared spectral data. Fire severity classes reported in GEEBAM include low (little change), medium (crown unburnt), high (crown partially burnt), very high (crown fully burnt) and unclassified (i.e. non-native vegetation or areas outside of the fire footprint). Further details of the fire severity mapping are available at https://www.envir onment.gov.au/syste m/files/ pages/ a8d10 ce5-6a49-4fc2-b94d-575d6 d11c5 47/files/ ageeb am.pdf.

| Regent honeyeater monitoring datasets
We used six monitoring datasets based on varying degrees of data quality (Table 1). Since 2015, we have used these datasets to establish a standardized, targeted and range-wide monitoring programme for the regent honeyeater-the National Regent Honeyeater Monitoring Program (NRHMP). The aim of the NRHMP is to increase the quality and quantity of monitoring data available for the regent honeyeater, with the ultimate goal of informing targeted conservation action to help prevent the species' extinction in the wild. Developed over the past 6 years through extensive field surveys, habitat modelling and expert elicitation, the NRHMP now surveys over 1,300 sites stratified in known or potential breeding areas throughout the species' contemporary breeding range ( Figure S1) during the Austral spring and early summer . The NRHMP sampling regime aims to account for both the nomadic life history and the breeding biology of the regent honeyeater (Crates, Terauds, et al., 2017). The six monitoring datasets based on wild birds are as follows: 1) Extent of occurrence (EOO): minimum convex polygon of verified regent honeyeater sightings since 1990, sourced from BirdLife Australia and used for IUCN classification.
2) Area of occupancy (AOO): 1 km × 1 km grids containing a verified regent honeyeater sighting since 1990, sourced from BirdLife Australia. or by members of the public (n = 6) since August 2015 . We also included three nests involving a captive-released regent honeyeater if the partner was of wild origin.

| Spatial and statistical analysis
We used ArcGIS Desktop 10.8 (ESRI, 2020) for all geoprocessing.
Prior to spatial analysis, we projected all spatial data to EPSG: 3,577 (GDA94-Australian Albers), ensuring equal area between raster cells. We resampled the GEEBAM data to 40-m resolution during projection, then converted from raster to polygon for later use in analysis. We clipped the EOO minimum convex polygon to the coastline and converted each pixel of the MaxEnt raster (i.e. 0 | 1; unsuitable | suitable) into a distinct polygon using a three-step procedure involving: i) raster to point conversion; ii) creation of a fishnet based on the raster dimensions; and iii) conversion of point feature to polygon with raster values, using the raster to point output for labelling the final polygons. We projected all point data for known nests, public sightings and NRHMP sightings from WGS84 to Australian Albers, buffered by 500-m radius, and converted from circular polygons to square polygons. We used a 500-m buffer to account for foraging and dispersal movements of regent honeyeaters (Geering & French, 1998) and created square buffers around point data to ensure consistency and comparability between reporting of impacts based on raster data (1 km 2 cells) versus vector distribution data.
Because the sightings and nesting databases confer data at the indi-   (Figures 1-3). Twenty-three per cent of habitat near fire-affected nests burned with high or very high severity. Logistic regression revealed the proportion of fire-affected cells in each database differed significantly (Table 3, Figure 2).

Minimizing biodiversity losses from catastrophic events such as
Australia's megafires depends on the implementation of urgent conservation actions for severely affected species (Wintle et al., 2020).
Implementing effective prioritization to identify priority species from an estimated 327 fire-affected threatened taxa  requires accurate estimates of bushfire impact, alongside accurate information on pre-fire imperilment, other threats and population trajectory. This in turn ideally requires high-quality, species-specific monitoring data (Boer et al., 2020;Bottrill et al., 2008). However, available occurrence records and monitoring data for many fireaffected taxa are poor (Scheele et al., 2019), which reduces confidence in the results of conservation prioritization and potentially misallocates conservation resources. We demonstrate with our case study of the critically endangered regent honeyeater that estimates of fire impact can vary dramatically depending on the input dataset.
Specifically, we found that methods to estimate species distributions that are available for many vertebrate species, such as EOO and AOO, underestimated fire impact by at least half in comparison with records derived from a targeted, contemporary monitoring dataset. We examine some of the underlying discrepancies between the estimates obtained from the different datasets and outline recommendations for improvement.
The datasets that led to the most severe underestimation of impact were AOO and EOO, which utilized verified sightings data from the 1990's onwards (including the targeted contemporary monitoring data), broadly in line with the temporal period considered in other assessments . However, rapid population decline has seen a concurrent range contraction in regent  (Runge et al., 2016;Welbergen et al., 2020). Many of these nomadic species rely on the co-occurrence of multiple dynamic habitat features, such as eucalypt blossom, water and tree hollows, for potential habitat to become functional habitat . In any given year, only a tiny proportion of a nomadic species' breeding range may be suitable (Webb et al., 2014), and if these areas are fire-affected, the capacity for such species to breed successfully in the years during and following fire may be seriously limited (Runge et al., 2014). If these critical areas are unknown due to a lack of systematic, spatially extensive monitoring, then bushfire impacts on nomadic species could be substantially underestimated.
Our results raise a question central to efforts to assess the biodiversity impact of major catastrophic events such as Australia's recent megafires: What is the appropriate input dataset upon which to base assessments? Outdated, incomplete or inaccurate input datasets risk biasing assessments, which could either over-or underestimate impacts. As we observe for the critically endangered regent honeyeater, underestimated fire impacts may be common in species that have undergone rapid declines and for which occurrence records that are not yet a decade or two old are already redundant because the population has continued to decline. This is worrying, as these species may be the ones most vulnerable to extinction due to stochastic events (Melbourne & Hastings, 2008). For species experiencing rapid declines, we recommend assessments use only reliable contemporary distribution data (ideally less than a decade old and preferably less). For species without targeted monitoring programmes, available data within this time period are likely to be insufficient (i.e. since 2015, only 5% of known regent honeyeater nests were found by the public). In the case of rapidly declining species, the assumption that earlier occurrence records mean previously occupied areas still contain functional habitat that can be exploited TA B L E 2 Summary of the spatial databases used to evaluate the impact of Australia's 2019/20 megafires on regent honeyeater habitat. See Table S1 for area of occupancy and extent of occurrence impact estimates based on data from 1996 onwards a Differences between % of fire-affected cells and % fire-affected area is because it was not always the case that 100% of the area within fireaffected cells was burnt.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13385. Jason MacKenzie is a freelance spatial ecologist and principal of geoADAPT.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found online in the Supporting Information section.
Poor-quality monitoring data underestimate the impact of Australia's megafires on a critically endangered songbird.