Module 2: Sampling Strategies, Replicates & Experimental Design
Module 2: Sampling Strategies, Replicates & Experimental Design
Rigorous environmental field sampling: simple random, stratified, systematic, and cluster sampling, biological vs. technical replicates, gradient sampling along pollution transects, and eliminating pseudoreplication.
2. Environmental Sampling Strategies & Eliminating Pseudoreplication
No statistical test in R can compensate for flawed sampling design. When sampling natural ecosystems, spatial autocorrelation, temporal variation, and environmental heterogeneity require structured sampling designs rather than arbitrary sample collection.
The 4 Core Scientific Sampling Designs:
- Simple Random Sampling (SRS): Every sampling location in the target study area has an equal, independent probability of selection. Best for homogeneous open waters or uniform experimental mesocosms.
- Stratified Random Sampling: The study area is divided into discrete ecological strata (e.g., Riparian Zone, Forest Interior, Wetland), and random samples are drawn within each stratum proportionally. Ensures rare microhabitats are represented.
- Systematic Grid / Transect Sampling: Samples are collected at fixed spatial or temporal intervals (e.g., every 50 meters along a river transect from source to estuary). Crucial for mapping pollution gradients and spatial kriging.
- Cluster Sampling: Naturally clustered units (e.g., 5 forest patches chosen at random, with multiple soil cores collected within each patch). Reduces fieldwork logistics across vast geographical landscapes.
The Danger of Pseudoreplication (Hurlbert, 1984)
Pseudoreplication occurs when an experimenter treats subsamples from the same experimental unit as independent replicates ($N$), artificially inflating the degrees of freedom and generating falsely significant $p$-values ($p < 0.05$):
Classic Environmental Pseudoreplication Example:
A researcher studies the impact of a fertilizer spill by taking 10 soil cores around one contaminated pond and 10 soil cores around one reference pond. They analyze the data in R as $N=10$ per treatment ($N=20$ total).
→ Flaw: The sample size is actually $N=1$ pond per treatment! The 10 soil cores are technical subsamples. Any observed difference could simply be due to pond depth, local bedrock, or shading, not the fertilizer.
Hands-On R Script: Simulating Random vs Stratified Environmental Sampling
library(tidyverse)
# Simulate 300 coordinate points across 3 ecological zones
set.seed(42)
landscape <- tibble(
plot_id = 1:300,
habitat = rep(c("Wetland", "Forest_Canopy", "Grassland"), each = 100),
soil_carbon = c(rnorm(100, mean = 4.8, sd = 0.6),
rnorm(100, mean = 2.4, sd = 0.4),
rnorm(100, mean = 1.6, sd = 0.3))
)
# 1. Simple Random Sampling (n = 15 total)
srs_sample <- landscape %>% slice_sample(n = 15)
table(srs_sample$habitat) # Note unequal habitat representation!
# 2. Stratified Random Sampling (n = 5 per habitat stratum)
strat_sample <- landscape %>%
group_by(habitat) %>%
slice_sample(n = 5) %>%
ungroup()
table(strat_sample$habitat) # Perfectly balanced design across strataCommon Pitfalls & Troubleshooting Checklist
- Java Heap Exhaustion (GATK / Picard): If processes terminate with
java.lang.OutOfMemoryError, set your environment variable explicitly:export _JAVA_OPTIONS="-Xmx16g -Djava.io.tmpdir=/tmp". - Index Timestamp Mismatch: If tools complain that a BAM index is missing or out of date, regenerate with
samtools index -@ 4 sample.bam. - Conda Channel Conflicts: Avoid ABI incompatibilities between bioconda and conda-forge by setting strict priority:
conda config --set channel_priority strict. - Chromosome Naming Contamination: Always verify whether reference FASTA headers use
chr1or1; divergent naming causes silent alignment drops.