Alert thresholds for monitoring environmental variables: A new approach applied to seagrass beds diversity in New Caledonia Simon Van Wynsberge a,b,⇑ , Antoine Gilbert c , Nicolas Guillemot d , Claude Payri a , Serge Andréfouët a a UR-227 CoRéUs, IRD (Institut de Recherche pour le Développement), Laboratoire d’Excellence CORAIL, Noumea, New Caledonia b UMR-241 EIO, UPF (Université de la Polynésie Française), Faa’a, Tahiti, French Polynesia c Ginger Soproner, Noumea, New Caledonia d Nicolas Guillemot Consultant, Noumea, New Caledonia article info Keywords: Indicators Power analysis Alert threshold Seagrass Monitoring abstract Monitoring ecological variables is mandatory to detect abnormal changes in ecosystems. When the stud- ied variables exceed predefined alert thresholds, management actions may be required. In the past, alert thresholds have been typically defined by expert judgments and descriptive statistics. Recently, approaches based on statistical power were also used. In New Caledonia, seagrass monitoring is a priority given their vulnerability to natural and anthropic disturbances. To define a suitable monitoring strategy and alert thresholds, we compared a Percentile Based Approach (PBA) and a sensitivity analysis of power (SAP). Both methods defined statistically relevant alert thresholds, but the SAP approach was more robust to spatial and temporal variability of seagrass cover. Moreover, this method characterized the sensitivity of threshold values to sampling efforts, a useful knowledge for managers. Ó 2013 Elsevier Ltd. All rights reserved. 1. Introduction Resource exploitation, pollution, climate change, as well as nat- ural disturbances, can affect ecosystemic variables of different types (Bellwood et al., 2004). Their monitoring is of critical impor- tance to detect abnormal trends and environmental degradations (Hughes et al., 2003; Halpern et al., 2008). Before explaining causality of changes, monitoring plans have to detect changes. Recently, several management frameworks have called for fast and useable protocols to define ‘‘alert’’ or ‘‘quality’’ thresholds (e.g., European Water Framework Directive, 2009; Australian and New Zealand Guidelines for Fresh and Marine Water Quality, 2000). In practice, a portion of the available dataset is used to define reference values, and ‘‘alert state levels’’ are pro- claimed as soon as the monitored variable exceeds these reference values. This monitoring scheme has been recently applied for ben- thic quality assessment in both temperate (Borja et al., 2003, 2007; Nilson and Rosenberg, 1997; Rosenberg et al., 2004) and tropical (Bigot et al., 2008) regions, and for monitoring seagrass beds in Australia (McKenzie, 2009). Many studies are based on descriptive statistics approaches (e.g., percentiles or medians), mostly because of their simplicity and robustness to extreme events. In the past decades, techniques using the power of statistical tests have also been developed. The statistical power reflects the capacity to detect a significant difference between ‘‘reference’’ and ‘‘tested’’ datasets when a change actually occurred between them. Power is a function of sample size (N), the probability of type I error (a), and the magnitude of the difference between the null hypothesis and reality (the ‘‘effect size’’ or ‘‘standardized effect size’’; Cohen, 1988, Fig. 1). When, for a given sampling effort, the difference between the ‘‘tested’’ dataset and the ‘‘reference’’ data- set is high compared to natural variability, the probability to be right by concluding to a difference between the ‘‘tested’’ and ‘‘ref- erence’’ datasets, using a statistical test, is high (i.e. statistical power is high). A useful application of statistical power is the sensitivity analy- sis, which is commonly used in various science fields (e.g., Cohen, 1988; Faul et al., 2007, 2009). Sensitivity analyses of power aim at determining the effect size required to conclude on the reality of a difference for a given power. Guillemot and Ducrocq (2011) used a sensitivity analysis of power for commercial fisheries management in New Caledonia’s South Province to determine Catch Per Unit of Effort (CPUE) thresholds on the basis of historical dataset and re- lated trends. Logically, Guillemot and Ducrocq (2011) defined afterwards for managers ‘‘pre-alert’’ and ‘‘alert’’ states in reference to historical levels of exploited fish stocks. In New Caledonia, fish- eries are not the only source of concerns. Coastal environments pay a toll to ongoing mining projects throughout the country. Nickel mining is a large driver of the local economy, and massive projects lead to visible environmental impacts on terrestrial and marine realms (e.g., Fichez et al., 2010; Morat, 1993). The critical habitats 0025-326X/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.marpolbul.2013.09.035 ⇑ Corresponding author at: UMR-241 EIO, UPF (Université de la Polynésie Française), Faa’a, Tahiti, French Polynesia. Tel.: +689 76 36 35. E-mail address: simon.vanwynsberge@gmail.com (S. Van Wynsberge). Marine Pollution Bulletin 77 (2013) 300–307 Contents lists available at ScienceDirect Marine Pollution Bulletin journal homepage: www.elsevier.com/locate/marpolbul