Contextualizing UK moorland burning studies with geographical variables and sponsor identity

1. It has been claimed that geographical variability could alter conclusions from some studies examining the impacts of prescribed moorland burning, including the Effects of Moorland Burning on the Ecohydrology of River basins (EMBER) pro -ject. We provide multiple lines of evidence, including additional analyses, to refute these claims. In addition, new findings from EMBER study catchments highlight previously unconsidered issues of burning adjacent to and over watercourses, contrary to guidelines. 2. A systematic review confirms the EMBER conclusions are in line with the majority of published UK studies on responses to prescribed burning of Sphagnum growth/ abundance, soil properties, hydrological change and both peat exposure and erosion. 3. From this review, we identify an association between sponsor identity and some recent research conclusions related to moorland burning. This additional variable, which has not previously been incorporated into moorland burning policy debates, should be given greater consideration when evidence is being evaluated. We also show that sponsorship and other perceived conflicts of interest were not declared on a recent publication that criticized the EMBER project. River multiple impacts associated with prescribed vegetation on peatland.

. There have also been debates about the effects of changes in vegetation and catchment processes due to vegetation burning (e.g. McCarroll, Chambers, Webb, & Thom, 2016;Yallop, Clutterbuck, & Thacker, 2010). Ashby and Heinemeyer (2019;'A&H') added to the debate with their critique of four of the 'Effects of Moorland Burning on the Ecohydrology of River basins' (EMBER) papers published to date.
In our view, the A&H paper in several places made unfounded statements apparently intended to undermine all EMBER outputs.
A&H suggested that the EMBER work was problematic, proposing that geographical variation had not been considered. The critique represents part of an intense debate about UK moorland burning Brown, Holden, & Palmer, 2016;Davies et al., 2016;Douglas, Buchanan, Thompson, & Wilson, 2016;Evans et al., 2019). Most recently, some studies on peat and carbon accumulation (Heinemeyer, Asena, Burn, & Jones, 2018;Marrs et al., 2019b) were suggested to have overstated conclusions due to use of incorrect methods , and these papers have required corrections to clarify perceived competing interests (Heinemeyer et al., 2018;Marrs et al., 2019a). At the same time, as researchers are increasingly required to evidence societal impact of their work, perceived decreases in public funding mean that researchers are seeking to diversify research funding, which may include sponsors with some form of agenda. There has been no detailed analysis of the funding source or competing interests amongst contributors to these debates and, therefore, the extent to which such factors may or may not be influencing the discussion remains unclear.
Here, we address three issues. First, we examine the A&H assertion that geographical variability contributes to false conclusions drawn from EMBER studies. Second, from a review of the current literature, we seek to establish whether EMBER conclusions are in line with published studies on responses to burning of Sphagnum growth/abundance, soil properties, hydrological change or peat exposure and erosion. Third, we examine whether sponsor identity might be associated with published research outcomes.
We show that: A&H′s critique contains multiple incorrect portrayals of where geography (linked to site-and plot-specific analyses) was incorporated in EMBER analyses, and new analyses illustrate this further; we highlight a selective focus of the A&H critique, which ignored papers published since 2017 using EMBER data; we show their concerns about soil temperature responses are unfounded; we demonstrate that EMBER results are in line with the majority of other published studies; we provide new evidence that guidelines on burning near watercourses appear not to have been followed in EMBER study catchments, and we identify the possibility that, in some cases, published evidence could be associated with the particular agenda of sponsors-a concept known as sponsorship bias. We therefore contend that sponsor and other perceived conflicts of interest, in relation to authors of research outputs plus those conducting peer reviews, may need to be considered by journals and policymakers when interpreting research conclusions.

| Examination of A&H claims
A&H selectively focused on four publications (Brown, Johnston, Palmer, Aspray, & Holden, 2013;Holden et al., 2014Holden et al., , 2015, even though EMBER supported three more primary research papers to date (Aspray, Holden, Ledger, Mainstone, & Brown, 2017;Brown et al., 2019;Noble et al., 2018). We perceive the selective focus as an attempt to undermine the entire project. A&H claimed that altitude was unaccounted for in EMBER publications, and that because it would be linked with precipitation and temperature across the study sites, it should have been considered further. Altitude and precipitation data from Table 2 in A&H were assessed using linear regression, and the assessment was repeated with catchment outlet altitude (e.g. Brown et al., 2013). We tested for association between water temperature and catchment outlet altitude (Brown & Holden, 2020;Brown et al., 2013). A&H claimed altitude, catchment size and precipitation effects would likely affect river invertebrates but that this had not been considered even though the original analysis incorporated water temperature (associated with altitude). We fitted catchment size and run-off parameters (associated with precipitation) from Holden et al. (2015) to the non-metric multidimensional scaling (NMDS) solution using the envfit procedure, and assessed community composition data collected in five sampling periods using ANOSIM, as described in Brown et al. (2013). Papers ignored by A&H (Aspray et al., 2017;Brown et al., 2019) suggested that fine particulate organic matter (FPOM) from peat erosion (as expected following vegetation removal with fire) can have significant effects on ecosystem structure and functioning when deposited in rivers. Brown et al. (2013) were tested for association with catchment size and altitude, and with rainfall totals for the month of sampling from the modelled gridded precipitation records used by A&H. FPOM densities and macroinvertebrate community metrics discussed by Brown et al. (2013) were analysed further using mixed-effects models, to assess whether site-specific variables (water temperature, catchment size, geology, flow variables) were associated with responses alongside burn effects (see Supporting Information).

FPOM densities reported by
While there are no recorded cases of EMBER 'surface' thermistors being exposed periodically to sunlight and being warmed artificially as claimed by A&H, we tested the effect of this possibility to determine whether it alters conclusions. Statistical models were developed by Brown et al. (2015) to predict daily maximum soil temperature in plots burned 15+ years prior to the study. These models were applied to predict temperatures of plots burned 2, 4 and 7 years previously, with outliers from predicted temperatures (hereafter 'disturbances') enabling estimation of burning effect magnitude. Using the maximum temperature datasets, the top 10% of disturbances, encompassing the peak temperatures commented on by A&H, were discarded, and the analysis re-run following Brown et al. (2015). We also examined reference lists from recent publications, including other systematic reviews (e.g. Glaves et al., 2013), and from our own knowledge of relevant research outputs. We initially rejected studies not based in the UK uplands, those focusing solely on wildfire effects, review/opinion/comment papers or literature not available publicly for peer review (e.g. reports to water companies, summaries of unpublished data), and those with no obvious relevance to EMBER studies. The initial searches and shortlisting produced 135 potentially relevant peer-reviewed publications.

| Review to contextualize EMBER findings
We reviewed each shortlisted paper focusing particularly on abstract, results, discussion and conclusions to categorize papers according to seven ecosystem properties studied in EMBER (Table 1).
Finer scale properties for specific variables (e.g. pH, DOC, EC as part of stream water chemistry) were explored initially but returned low numbers of studies, hence our use of broader groupings. Overall, 68 papers were considered to be directly relevant. Our approach was to categorize papers based on statements and suggestions within each paper, accepting the expert judgement of the scientists involved based on their detailed evaluations of the datasets available to them (see Supporting Information). For four of the properties, we considered it possible to classify suggested responses to vegetation burning as positive, negative or having no/mixed effects (Table 1). We classified such responses when authors of those papers made clear suggestions that there was a burning effect, no burning effect or results were varied/inconclusive respectively. All papers that were found to be relevant to the first four ecosystem properties were classified in terms of a combined effect: + (only positive outcomes suggested across the four properties), − (only negative outcomes suggested) or 0 (no clear outcomes, or a mixture suggested). For the other three ecosystem properties (soil physical/chemical properties, stream water chemistry, hydrology), we classified responses in terms of whether there was a change/difference (yes) or no change/difference (no) suggested. The approach for these three properties was necessary because most of the studies lacked clear statements as to whether effects could be deemed positive or negative for peatland function.

| Sponsor identity
For each paper, acknowledgements, funding declarations (where present) and/or affiliations were used to determine sponsors and relevant competing interests, then combined into groups for analysis: (a) Grouse shooting industry compared to non-grouse shooting groups, and (b) Government agencies compared to non-government groups (see Supporting Information). We focused on these two comparisons because there is the possibility that scientists in

Ecosystem property response to burning Classification
Sphagnum growth/abundance + = positive response (e.g. increased growth and/or higher abundance/cover suggested) − = negative response (e.g. decreased growth and/or lower abundance/cover suggested) 0 = no or mixed response suggested Mean and/or maximum soil temperature + = decreased temperatures suggested − = increased temperatures suggested 0 = no temperature change Peat exposure and/or erosion + = reduced bare peat and/or erosion − = enhanced bare peat and/or erosion 0 = no or mixed response Aquatic invertebrate communities + = positive response suggested (e.g. higher diversity and/or densities of sensitive taxa) − = negative response suggested (e.g. lower diversity and/or densities of sensitive taxa) 0 = no or mixed responses Peat physical and/or chemical properties (including pore water chemistry)

| Examination of A&H claims
A&H suggested EMBER results are unreliable because they were based on a space-for-time approach with treatments located in geographically separate and environmentally distinct sites. A&H further A&H suggested slope varied between EMBER plots, but they made a fundamental mistake in their assessment of how EMBER incorporated slope. Three soil papers Holden et al., 2014Holden et al., , 2015 stated that plot locations were determined based on topographic index (TI) categories. Consequently, across the catchments, there were three groups of plot locations defined by the TI which incorporates both slope angle and upslope drainage length, which is a much more logical approach for comparing treatment effects than just using slope angle ( A&H used a 50 × 50 m digital elevation model covering each of the EMBER plots and suggested a significant difference in slope between unburnt (steeper) and burnt plots. However, their analysis suggests that the difference was <1°, which is so small in the context of UK moorlands that their criticism has no physical implications and so can be disregarded. First, given that EMBER plots were approximately 20 × 20 m, 1° lies well within the margin for error when calculating slope using a 50 × 50 m UK upland grid. Second, A&H did not show how this effect size could possibly be meaningful, particularly as blanket peatland is often found covering slopes up to 20° (Lindsay et al., 1988) and in extreme cases up to 30° (Ingram, 1967). Hence, this additional evidence indicates that the effects of burning were sufficiently large to override slope effects encountered within EMBER plots.
A&H questioned whether EMBER plots (by burn age or burned/ unburned) were distributed equitably by aspect, although in their own analysis, they did not find any significant effect. A&H noted that 'elevation exerts a strong influence on precipitation which, in turn, affects peatland water tables and overland flow'. By inference, they suggested that hydrological data from the EMBER sites are therefore problematic as catchments were (unavoidably) in different locations. However, their own analysis (Figure 2c in A&H) showed no significant overall difference in elevation between burnt and unburnt catchments. Furthermore, analysis of data in A&H's Table 2 reveals two significant weaknesses in their argument: (a) there was no significant relationship between mean elevation and mean   (Figure 1a). Using water-    and the non-metric multidimensional scaling solution in Brown et al. (2013). Two variables that correlated significantly (shown in italics) were not associated with altitude or catchment size ( Figure S1) A&H criticized Brown et al. (2013) for not considering any between-site differences when analysing macroinvertebrate community and river habitat responses to burning. The suggestion by A&H is inaccurate because the analysis did include water temperature data, which were associated with altitude (R 2 = 0.56, p = 0.013). Analyses already detailed in that paper showed water temperature was not associated with the NMDS solution, nor was catchment size (R 2 = 0.04; p = 0.365). Rainfall was not incorporated into this analysis because rainfall-run-off relationships are modified by catchment processes, and river invertebrates would thus respond to flow rather than rainfall.
Incorporation of mean flow metrics subsequently provided for each catchment by Holden et al. (2015) suggests that invertebrate communities may be associated with flow variability in our study (Table 2) including changes linked with vegetation burning. Additional analysis suggests these site variables were not associated with communities in  with conclusions of Brown et al. (2013). EMBER experimental evidence (Aspray et al., 2017;Brown et al., 2019) directly implicated peat sediment FPOM deposition as a driver of river macroinvertebrate community changes. A key question arising from these findings is where could the extra sediment in rivers draining burned catchments come from?
Vegetation removal with fire often exposes the peat surface, enhancing erosion potential, possibly through micro-rill development around exposed tussocks and other microforms (e.g. Lindsay, 2010), with sediment transfer to rivers then more likely. This erosion risk is recognized explicitly in the Heather & Grass Burning Code (Defra, 2007)  Eroded peat from patches burned close to watercourses could therefore be transported easily into headwater rivers. Aquatic ecosystem effects can occur quickly over hours-days after deposition of acute sediment pulses (Aspray et al., 2017), while in burned peatland rivers, sediment deposition is also likely to be a chronic stressor.
Burning adjacent to watercourses was not considered in a previous assessment, which suggested that prescribed burning on one selected moor followed the Defra code best practice (Allen, Denelle, Sánchez Ruiz, Santana, & Marrs, 2016). We contend that the wider extent of this problem needs to be quantified from further aerial imagery and ground-based assessments to measure compliance with voluntary burning codes.
A&H proposed that some soil surface temperatures measured in EMBER plots could be due to measurement error caused by sensors at the peat surface having parts exposed to sunlight. While Brown et al. (2015) used the term 'surface' for sensors placed shallowest in the soil profile, they also explained that sensors were placed horizontally in the top 1 cm of the peat-litter layer (i.e. not directly on the surface) and checked every 3 weeks. Some maximum temperatures recorded at B2 plots were similar to those reported by Kettridge, Thompson, and Waddington (2012) for Canadian peatlands after fire, even though they used a different sensor (see discussion in Brown et al., 2015). We therefore have confidence in the temperature data from our study. Several lines of evidence, including some available to A&H, provide further evidence-based assurance that the EMBER soil temperature data are robust: (a) higher temperatures were recorded with sensors buried at 5 cm depth in B2 plots  and surface temperatures at these locations would have been similar to/higher than 5 cm depth; (b) sensor exposure to sunlight cannot explain why the lowest temperatures were also recorded in recent burn plots, which in turn would enhance soil ice formation and erosion processes (Li, Holden, & Grayson, 2018); (c) further analysis of data with the top 10% disturbance values removed confirm findings in Brown et al. (2015;

| Review to contextualize EMBER findings
The ecosystem properties with the largest number of papers relevant to EMBER findings were those concerned with alterations to soil physical and chemical properties, and to Sphagnum growth/ abundance (

| Sponsor identity
Of the 68 papers reviewed, 11 had declared funding links to grouse shooting groups and 30 cited government agency funding. While there were no apparent statistical links between government agency funding and suggested burn impacts, for studies funded by the grouse shooting industry, there was a significantly higher probability of suggesting positive effects of burning for Sphagnum growth/ abundance and the combined findings ( Hydrological function n/a n/a n/a Stream chemistry n/a n/a n/a Peat physical/chemical properties 0.4368 n/a n/a n/a

Aquatic invertebrates
Hydrological function n/a n/a n/a Stream chemistry n/a n/a n/a Peat physical/chemical properties 0.302 n/a n/a n/a was associated with sponsor identity (Figure 4). Most of the + publications are based on data from a single experimental area on sloping blanket bog at Moor House, northern England.

| D ISCUSS I ON
Sponsorship effects are a known phenomenon in science (Lesser, Ebbeling, Goozner, Wypij, & Ludwig, 2007) Thus, it is unclear how these NERC projects could have funded the A&H critical analysis. We accept that the above discrepancies may have been unintended or there could be valid explanations, but due care is required by scientists when they submit papers to academic journals for peer review to ensure potential perceived conflicts of interest can be managed by the editorial process.
We have shown that: (a) geographical variability does not con- situations. Formal meta-analysis would provide an alternative way to evaluate any potential bias in comparison to our study-count approach, but researchers will routinely need to provide clearly defined and comparable effect size estimates to enable these kinds of analyses. We agree with A&H that policymakers need sound evidence to support the policy process on moorland burning. Fully transparent statements about funding and potential conflicts of interest, supplied at the outset of the peer review and publication process, represent a key part of the assurance that published research is reported and reviewed as objectively as possible. Any apparent weakening of this principle should be a source of concern to all who publish.

ACK N OWLED G EM ENTS
EMBER was funded by NERC (NE/G00224X/1) and Yorkshire Water.

AUTH O R S ' CO NTR I B UTI O N S
Both authors conceived the ideas and designed methodology; both authors collected the data; both authors analysed the data; both authors led the writing of the manuscript and gave final approval for publication.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available via University of Leeds Repository https://doi. org/10.5518/833 (Brown & Holden, 2020).