Effectively integrating experiments into conservation practice

Making effective decisions in conservation requires a broad and robust evidence base describing the likely outcomes of potential actions to draw on. Such is typically generated from experiments or trials that evaluate effectiveness actions, but for many missing incomplete. We discuss how can be by incorporating into practice. This most efficient if opportunities carrying out informative, well-designed are identified at an early stage during management planning. consider navigate way between stringent requirements statistical textbooks complexities ecological real world considering practical approaches key issues replication, controls randomization. suggest routinely sharing results could increase both value money argue with planning small additional input effort, important new learning gained implementation actions. 1. 生物多様性保全において効果的な意思決定を行うためには、特定の対策がもたらす可能性のある効果についての広範で頑健な科学的根拠が必要となる。このような科学的根拠は対策の有効性を評価する実験や試験から得られるものの、保全の現場では利用できる科学的根拠が欠如しているか、根拠が得られても不完全であることが多い。 2. 本稿では、保全活動に実験的アプローチをより積極的に組み込むことで、どのように必要な科学的根拠が得られるようになるかについて考察を行った。どうすれば有益で適切に設計された実験が行えるかを保全管理計画の初期段階で理解しておくことが最も効率的であると考えられる。 3. 統計学の教科書では実験計画法について厳しい要求が行われている一方で、現場で行われる生態学的な実験には様々な複雑さが関わり、その中で反復や対照群、無作為化といった重要な問題に実際どのように取り組むべきかを考察した。 4. 日常的に実験結果を共有することで、保全活動の費用対効果と有効性の両方を高めることができるだろう。 5. 早期に計画を立て、わずかな労力を追加するだけで、多くの保全活動において重要な新しい知見を得ることができると考えられる。 Despite increasing appreciation importance using policy, lack sufficient (Christie et al., 2021). Indeed, implemented interventions have no their (Sutherland 2020). Decisions taken on basis incomplete inaccurate information lead inefficient use limited resources available poorer outcomes. Routinely testing interventions, integrating practice, significantly decision makers, improving (Feinsinger, 2001; Cadotte However, designing yield useful feasible implement as part practice challenging. Although ecology undergraduate students taught experimental design, these courses frequently focus ideal scenarios, large numbers uniform units (such petri dishes field plots) where treatments replicated, randomized stratified ease. While important, such principles difficult apply complex natural systems. leads divergence terms design data analysis expectations academic researchers realities practitioners working ground 2017). The minimum standards often recommended necessary inference appear unachievable within projects. Our experience, including editing reviewing articles submitted Conservation Evidence Journal (AT, NO, TA WJS) Ecological Solutions (MC) conservationists faced challenge (MHH PT-M), there produce which realized changes intervention (e.g. Douglas 2019). In this article, we some associated generating knowledge neglected question when ‘non-ideal’ replicates without untreated controls) worth implementing. propose series steps identify include will randomization need controls. intention inclusive, therefore not assumed too much prior knowledge, hope reaching wide audience, particularly those ground. believe significant carry simple manipulative that, modest inform ongoing adaptive improve wider community. order generate first component informatively efficiently integrated (Figure 1). three requirements: (i) action being undertaken, better understanding would make difference (i.e. uncertain costs uncertainty significant); (ii) skills analyse experiment available; (iii) included existing workplans relatively easily repeated times allowing different compared; staff capacity monitor outcome). These components may vary, through time across so tests should regularly reviewed. Once opportunity answer has been identified, next address it, given available. questions presented Table 1, along worked example Box describe stages process considered by, example, site manager, officer reserve team. list, linear require iterative feedback adaptation selected treatment becomes unfeasible conditions change). Considering all before work gets underway project more informative than monitoring retrospectively reframed format. A its population ground-nesting birds (1). long-distance footpath cuts one dogs off belonging walkers path thought disturbing nesting (2). team like know change behaviour reduce disturbance (3). other options, closing prohibiting dogs, were decides option put up sign near (4). sure whether emphasising emotional motive keeping (5). Therefore, they decide trial: sign, asking keep birds, photo dog dead wader chick appeal. signs changed livestock manager visit each day. It take day create schedule few minutes switch (6). monitoring, tasks same observe people who passed document lead. about 10 Results recorded cards (7). assistant warden, looking challenges helped plan experiment, oversee (she suggests chi-squared test). Time allocated (8). set-up decided, considers it experiment. fact long distance provides opportunity, only pass once (the handful known local regular counted once). person independent, facilitating analysis; used day, might estimated person-days total, carried short periods interfere normal management. expected two days week write up, done wet winter days. seems worthwhile understand (9). Finally, shared. findings warden published paper after her MSc keen collaborate. Together submit open access journal Evidence, even study does detect behaviour. stated organisation's annual (10). There range test (stage 6 Many come down distinguishing effects caused variation exists, sites individuals. Suppose rotated weekly basis. possible weeks coincides period especially pleasant weather, visitor groups, number walked tendency let Any observed then due differences groups response rather treatments. To another if, following grazing intensity field, density orchids increases, well result altered regime Hutchings, 2010). also factor – perhaps rabbit decreased or, unbeknown researcher, farmer fertilizer regime. cornerstones control chance ‘background’ variables overwhelmingly influencing outcome certainty any ascribed taken. If based observations single population, serious concern peculiarities covariance variable interest unknown above example). Implementing independent site, individual usually confident action, respond consistently. basic principle ‘the better’, because applying larger increases accuracy hence detecting treatment, limit independently replicated (How wetlands restored? How islands invasive species eradicated from?). costs, time, replicates. although provide accurate information, must balanced against practicalities needed specified effect size calculated power (Lehr, 1992; Crawley, 2015). cases system understood, estimate present calculate required size. essence, smaller detect, greater variability exists replicate units, replication needed. accurately known. addition, specify target do wish specific amount, rate, simply it? cases, guided type what practically possible, magnitude (compared present), body already (Table 2). remember chosen affect widely applicable inferences be. forest grazers excluded plots meadow deadwood left standing five woodland) confidently applied area. landscape region, scope perhaps, appropriate caveats, widely. n = 1 (as described hand column 2) heresy scientists, reality lessons learned water quality introduction European beavers Castor fiber Puttock 2017), reintroducing wolves Canis lupus vegetation structure Yellowstone Park, USA (Mao 2005) unplanned ‘natural experiments’ impossible otherwise, responses eruption Mount St Helens Washington, (Swanson & Crisafulli, 2005). examples before-and-after comparisons (see below), community accepts despite replication. higher levels less generalizable studies, elucidate processes formal experiments. Their discussed context long-term system, interpretation aided underlying mechanism previous variation. though unreplicated weak low direction effect, usefully combined similar meta-analysis. An eradicating mammals seabirds 61 oceanic (where had 1) found seabird populations increased, mean recovery rate 1.12 (Brooke 2018). Here, very author group studies average impact action. was various single-island properly documented shared literature. Large-scale habitat restoration (Davies Gray, 2015), meta-analysis projects able quantify biodiversity ecosystem services provided (Rey Benayas 2009). Similarly, protected areas normally studied sites, generalize measure overall area (Coetzee 2014). general, organizations large-scale For RSPB, UK organization owns nature reserves own science staff, developed novel via reserves, co-designed scientists Hancock 2009; Malpas 2013). manages collaborate collect comparable multiple sites. reflects think creatively adapt organizational constraints (illustrated top overlap Figure Governments major funders role coordination organizations, support financially. are, Conversely, reducing sample variance required. Sample affected measurement errors. several ways reduced, detected. Firstly, every aspect variability. After rigour sampling error. achieved ensuring methodology consistent replicates, equipment measuring length calipers tape) training collecting data. Increasing either taking samples larger, extensive, estimates replicate. look excluding deer structure, diameter trees measured true low, between-individual substantial. ability fencing low. Taking measurements 20 straightforward obtain diameter. standard error decline investigated plotting size) find compromise level sampling, reduces acceptable level. Another covariates confound account statistically. above, edge varies trees’ diameter, plot accounted fencing. Note added analysis, allow relationships power. As unit variance. avoid pseudoreplication: treating (Hurlbert, 1984). dog-walking 1), treated points pseudoreplication, independently. mistakes incorrectly inflate study, invalidating conclusions. this, averaged analysed generalized mixed models random factor). reliability vital clearly interpreted appropriately Choosing show bigger particular cannot altered, shorter chain links stronger responses. education programme aims herbivores killed snares, people's frequency trips set snares) resultant mammal abundance species), influenced factors years noticeably. risk surrogate ultimate variable. snare setting declined areas, animals correspondingly. Including comparison allows attributed (known unknown) variables. conventional approach compare (a control-impact design). fish marine compared unprotected contain Rakitin Kramer, 1996). unclear designated protection, protection place, differ commonly (before-after design), comparing designation confounded, climatic tourists time. favoured combine before-after-control-impact (BACI) monitored Examples 2 3 shown designs Does modifying internal swift nest boxes occupancy? One controlled trial Action Swifts (Newell addition artificial, molded ‘form’ swifts Apus apus occupancy rate. consisting 142 four present. Nest forms alternate breeding season began checked end subsequent seasons. Across association presence box occupancy. demonstrated stratification future boxes. prescribed burning native tree regeneration? regeneration Scots pine Pinus sylvestris Scotland (Hancock al. randomised, reserve; 100 m2 burned unburned. seedlings ten unburned areas. led ultimately regeneration. manipulation browsing (using exclude half site: burnt unburnt) second factor. Deer exclusion minor influence establishment, supporting markedly removal predatory snails predation pressure threatened corals effort required? Florida, USA, tested removing coral-eating Coralliophila abbreviata coral Acropora palmata causing tissue loss (Williams Twelve (each 150 m2) six reefs assigned treatments: (1) removed A. only, (2) species, (3) (control). baseline snail established began. Divers took approximately 30 remove 51 feeding scars reduced Given treatments, revealed effectively targeted only. variants versus (Smith Different subject modifications aspects frequency, timing, application Example plant rapidly spreading herbicide spraying late, Marushia help budget deliver desired outcome. Comparing circumstances immediate needs leaving spread plant. step allocate treatment(s) units. Doing carefully confounded background (Johnson, 2002). Conventional scientific thinking best randomly avoids problems selective bias occur experimenter assigns nearest road visitors arrive morning. Randomization works patterns sizes large, characteristics overall. hard accept, selecting circumstances. argument suitable Alternatively, exist, composition altitudinal gradient travelled further tend later, recommended. Stratification attempts achieve representation split into, altitude hour-long period, equal group. paired stu