Ecological Indicators (Jan 2024)
Sensitivity analysis to reduce sampling effort of a long-term monitoring program
Abstract
Long-term ecological monitoring programs are valuable but expensive. Over time, programs may need to reduce sampling due to funding limits or shifting research priorities. To maintain data streams, monitoring programs must be proactive about choosing how to reduce effort while preserving the broader goals of the program. We used data from an ongoing long-term monitoring program (13 + years) to develop a conceptual framework for guiding reduced sampling for programs that measure multiple responses. The program measures nine ecological responses to evaluate whether an artificial reef is functioning similarly to two nearby natural reference reefs (n = 82 transects per reef). We first developed goals for reduced sampling, informed by trends in historical data that showed consistent differences in the three reefs over time (i.e., priors). We determined the minimum number of locations (n = 15 transects per reef) for reduced sampling that would maintain the desired spatial distribution used in full sampling. With the decrease in statistical power associated with the reduced sampling design (45 % reduction in power; 82 % reduction in sampling effort), we changed our evaluation process from inferential statistics to a range-test approach. Subsetting from existing data (2009–2021), we found that full and reduced sampling responses were similar over time, suggesting that the 15 locations chosen for reduced sampling were representative of the reefs and full sampling. We simulated a scenario where the three reefs were ecologically identical and compared the intrinsic error rate (i.e., the likelihood of falsely concluding that a reef is similar or dissimilar) of the range test approach on the full versus reduced sampling designs. The intrinsic error rate for both sampling designs in this simulated scenario ranged from 25 to 29 %, a high but acceptable error rate considering priors. To evaluate the sensitivity of the range test approach in detecting a 20 % difference in any of the nine measured response variables, we evaluated reef-level performance when one of the reefs had 3, 4, or 5 responses degraded by 20 % using full or reduced sampling. The intrinsic error rate decreased substantially with the number of responses degraded, suggesting that we could detect differences of 20 % or more between reefs. Together, the results of these analyses increased confidence in reduced sampling and our ability to detect ecological differences among the reefs. In testing this conceptual framework for reduced sampling of long-term monitoring programs, we validated its practical application and demonstrate its usefulness in guiding how to reduce sampling.
