Reading a Stream Restoration Through a Drought
In 2019 a watershed council restored 1.2 km of Cedar Creek by adding large woody debris, removing a culvert, and reconnecting the channel to its floodplain. Monitoring compares the restored reach with two unrestored reference reaches upstream and downstream. Benthic macroinvertebrates are sampled each September with three Surber samples per reach, and the reported metric is EPT (Ephemeroptera, Plecoptera, Trichoptera) taxon richness. Results for 2017 through 2023 are: restored reach 8, 8, 15, 14, 13, 11, 11; reference reach means 9, 9, 10, 10, 9, 8, 8. The region experienced a severe drought in the summers of 2021 and 2022. The restored reach is the only one with a reconnected floodplain, so it retained standing pools while the reference reaches ran nearly dry. The council's press release states that the project increased stream health by nearly 90%.
1. Use the before-after-control-impact logic to evaluate the 'nearly 90%' claim. Calculate a defensible effect size and explain what the press release did wrong.
Read the worked answer
What the press release did. It compared the best post-restoration year (15 taxa) with the pre-restoration baseline (8 taxa): 15/8 is a 87.5% increase. That is a single-year, single-reach before-after comparison with no control. It ignores three things: the reaches were never identical, the reference reaches show their own trend, and the restored reach declined after 2020.
BACI calculation. Using before = 2017–2018 and after = 2020–2023:
- Restored: before mean 8.0, after mean (15 + 14 + 13 + 11)/4 = 13.25, change +5.25.
- Reference: before mean 9.0, after mean (10 + 10 + 9 + 8)/4 = 9.25, change +0.25.
- BACI effect = 5.25 − 0.25 = +5.0 EPT taxa.
Interpretation. The restoration still shows a real, positive effect of about 5 taxa, and that is a much stronger claim than 90%. But the raw before-after number is inflated: some of the 2019 jump reflects the fact that the restored reach started a taxon below the references, and the reference reaches drifted slightly downward, which the press release silently attributes to the project. The honest statement is a difference-in-differences of roughly five taxa, with the drought caveat discussed next.
2. Explain how the 2021–2022 drought undermines a simple causal claim, and propose how the monitoring data could be analyzed or extended to handle it.
Read the worked answer
The confound. Drought is a second treatment applied to the whole study, but not equally. The restored reach has a connected floodplain and retained pools; the reference reaches dewatered. So in 2021–2022 the reaches differ in two ways at once: restoration structures and drought refugia. Any taxa that survive in pools - lentic mayflies, dragonflies, beetles - will appear in the restored reach for reasons that have nothing to do with woody debris improving water quality. The restoration and the drought interaction are perfectly entangled.
What the data can still support. The strongest evidence is the pre-drought window: 2019 and 2020 show 15 and 14 against a reference of 10, before drought began. That is the cleanest test of the restoration effect. The 2021–2022 numbers should be reported separately, not averaged into a single post-restoration mean.
Extensions. Include drought as an explicit covariate or use an interaction term (period × reach type); add a second restored reach that lacks floodplain access, which breaks the confound; collect discharge, temperature and pool-persistence data so the mechanism is measured rather than assumed; and separate pool-dwelling from riffle-dwelling taxa so that a community shift toward lentic taxa is visible rather than hidden inside a richness total. If richness rises only through pool taxa, the claim becomes 'the project created drought refugia,' which is a legitimate but different finding.
3. The design uses one restored reach and three Surber samples per reach. Explain what this permits and what it does not, and suggest one metric change that would make the conclusion more mechanistically informative.
Read the worked answer
Replication. The three Surber samples are subsamples within a reach. They estimate local patch variability in that reach; they are not independent replicates of restoration. Treating n = 3 as if it were three independent restored streams inflates the apparent sample size and produces confidence intervals that are far too narrow. This is pseudoreplication. With one restored reach, the inference is to Cedar Creek's restored section, not to restoration as a practice.
What is permitted. A before-after comparison within this reach, and a contrast with two nearby reference reaches, support a site-specific inference: the restored section gained roughly five more EPT taxa than the references over the same period. That is worth reporting.
What is not permitted. Generalizing to other streams, or attributing the change to a specific restoration action, because the treatments were applied together (woody debris, culvert removal, floodplain connection) and cannot be separated within one reach.
Metric change. EPT richness is sensitive to sampling effort and lumps very different organisms. A more mechanistic choice is a trait- or habitat-based breakdown: EPT abundance as a percentage of total invertebrates, plus separate counts of riffle-obligate versus pool-associated taxa, or a dissolved-nutrient tracer to measure uptake velocity. These distinguish 'more taxa arrived' from 'the stream now processes nutrients and retains water differently,' which is what the council actually wants to claim.