Introduction

Northern peatlands are globally important to the global carbon cycle,
thought to contain up to 25% of global soil carbon
[@yuGlobalPeatlandDynamics2010]. The carbon accumulated in northern
peatlands has acted as a source of global cooling for the last 8 - 11
000 years [@frolkingHoloceneRadiativeForcing2007]. While definitions of
peatlands vary, they are usually defined as wetland ecosystems whose
deep soil contains high percentages of organic carbon
[@lourencoPeatDefinitionsCritical2023]. This study focuses on northern
peatlands, defined as those above 30 degrees latitude
[@yuGlobalPeatlandDynamics2010]. The Boreal-Arctic Wetland and Lake
Dataset divides peatlands into two nutrient level categories: bogs and
fens, with distinct vegetation assemblages
[@olefeldtBorealArcticWetland2021].

In addition to their importance to the carbon cycle, these wetlands are
also critical biodiversity refuges, home to a number of red-listed plant
species
[@christianiNegativeEffectsClimate2025; @lockyPlantDiversityCompositionManitoba2006; @saarimaaPredictingHotspotsThreatened2019].
While peatlands aren't typically higher in species richness compared to
surrounding areas, their principal biodiversity value is in supporting
regionally rare and threatened species, especially in regions where
peatlands are degraded or threatened with development
[@lockyPlantDiversityCompositionManitoba2006; @perrinVegetationRichnessIreland2020].

The main threat affecting the habitat potential of northern peatlands
are land use changes, specifically drainage for agriculture, forestry,
peat extraction and grazing [@loiselExpertAssessmentFuture2021].
Drainage for forestry is typically preformed by digging ditches to
convey groundwater away from the site to increase the rate of tree
growth and allow easier access for harvesting equipment
[@minkkinen2008climate]. This also increases the rate of decomposition
of the organic material in the soil, permanently altering the site to
become more forest-like [@minkkinen2008climate]. After drainage,
northern peatlands can experience "peatland forest succession", where
mosses and herbs adapted to wet conditions decline while deep-rooted
trees and shrubs increase in abundance
[@laihoDynamicsOfPlantMediatedDrawdown2003; @potvinEffectsWaterTable2015].
This is a gradual, long-term process, with most studies typically
observing changes over 15+ years [@laineLongTermEffectsWater1995] .
However, this pattern is not uniform - for example, very low-nutrient
peatlands in the United Kingdom and Ireland that didn't contain trees to
begin with might instead become more shrub-dominated
[@andrewsPlantCommunityResponses2022; @minkkinen2008climate].

Water level drawdown experiments are often used to study the effect of
peatland drainage on its plant communities, soil microbes or
geochemistry [@kosterWaterLevelDrawdown2023]. Some studies involve
placing plant cores in a tank with tightly controlled water levels
[@breeuwerDecreasedSummerWater2009], while others are simply
observational, comparing nearby drained and undrained sites after a
drainage event [@kosterWaterLevelDrawdown2023]. Other studies have
installed pumps in a hydrologically isolated part of a bog to simulate
water level drawdown caused by droughts
[@andrewsPlantCommunityResponses2022].

This study seeks to compare the results of water level drawdown
experiments and examine the effects of water level drawdown in northern
peatlands on (a) functional group abundances (trees, herbs, shrubs and
mosses) and (b) plant species richness. We expect to see an increase in
tree and shrub abundance and a decrease in herbs and mosses, consistent
with peatland forest succession, and a decrease in plant species
richness as rare bog species are out-competed by forest species.

Methods

Initial search

The Web of Science database was queried to find suitable papers for our
analysis, using all databases and collections, accessed through the
University of Helsinki network. Our initial search term, which was more
general and didn't focus specifically on water table drawdown or
rewetting, resulted in 19,505 results. After discussion, we decided to
restrict it to only water table drawdown experiments. The next query was
much narrower and only returned 50 papers. Finally, we used the
following query on April 20, 2026, which returned 368 papers:

("peat drainage" OR "drainage ditch*" OR "drained peatland*"
OR "water table drawdown" OR "water level drawdown" OR "WT*
drawdown" OR rewet* OR "hydrological restoration") AND (peatland*
OR bog* OR fen* OR mire* OR "boreal peatland*") AND ("plant
communit*" OR "community composition" OR "vegetation
composition" OR "plant functional type*" OR "functional group*"
OR "species turnover" OR "species richness")

Screening

Eleven duplicates were removed from the returned results, and the titles
and abstracts were screened to remove 257 unsuitable papers. In order to
keep the scope of the project more manageable, a decision was made to
exclude the 58 studies which examined the effect of peatland restoration
or re-wetting, restricting our scope to only water level drawdown
studies. The results of each step in this process can be seen in Figure
1.

Data extraction

Each screened paper was reviewed and relevant information was recorded
in a spreadsheet. First, the country, coordinates and elevation of the
study site were recorded. Next, we recorded the duration of the study,
or the number of years from the water table drawdown until the year of
the field observations.

Following this, the water table depth (WTD) mean, variance and number of
observations was recorded for the control and treatment sites. In some
cases, these values were inferred from a chart of WTD readings using
WebPlotDigitizer [@WebPlotDigitizer]. Some studies did not report
detailed WTD measurements, only the difference between sites. In one
paper [@hohtanenVegetationTeuravuoma1999], no WTD measurements were
made, so the WTD difference was inferred from a site restoration plan
[@niskanenTeuravuoma2017].

If the study recorded plant species richness at control and treatment
sites, then the mean species richness, number of plots and variance were
all noted. We also noted if any taxa were excluded from the species
richness measures, such as mosses.

Finally, we searched the paper for references to biomass or percent
cover of plants broken down by functional groups. While different papers
reported this in different groupings, we harmonized these into four
categories: trees, shrubs, herbs and mosses. For example, graminoids and
forbs were reported separately in [@kosterWaterLevelDrawdown2023], but
these categories were combined into "herbs" for this analysis, while
"ericoids" as reported in [@breeuwerDecreasedSummerWater2009] was
changed to "shrubs". In some cases, when no groups were reported
directly but a full dataset of measurements of individual species was
available, these were manually assigned into our categories, such as in
[@andrewsPlantCommunityResponses2022]. Some studies did not report
certain functional groups [@breeuwerDecreasedSummerWater2009], or
aggregated all vascular plants together
[@laineImpactOfLongTermDrawdownFunctionalTrait2021]; in these cases,
data was only extracted for some functional groups. WebPlotDigitizer was
again used in some cases to record abundance data from bar plots
[@WebPlotDigitizer].

Despit the techniques used for biomass measurements varying widely
between papers, an effort was made to record as much data as possible.
For example, some reported biomass productivity of new growth
[@kosterWaterLevelDrawdown2023], some reported below-ground biomass
[@murphyEffectsWaterTable2009], and others reported the change in
biomass production based on the number of hits from a sampling pin
[@breeuwerDecreasedSummerWater2009]. The measurement unit and technique
was recorded for each study.

Analysis

Statistical analysis was preformed using the "metafor" package for R,
with visualizations created using "orchaRd"
[@R2026; @orchaRd; @metafor]. After initial data cleaning using the
"tidyr" package [@tidyr], the data was split into species richness and
functional group studies. Effect size was calculated using the log
transformed ratio of means measure ("ROM") for the functional group
records. Since the functional group abundance was reported in different
units, this measure is most suitable for comparing data from very
different sources, since it's based on ratios and log-transformed
[@metafor]. For species richness, where this was not an issue, the
standardized mean difference with heteroscedastic population variances
measure ("SMDH") was used to reduce bias from assuming the same
variance in control and treatment groups.

Next, a one-variable multi-level linear mixed-effects model was used to
examine the relationship between functional groups and effect size,
using the paper ID over data point ID as nested random effects.
Functional group was used as a moderator. Following this, a similar
model which also used water table depth difference as a crossed
moderator was used to see if the effect on functional groups was
affected by the degree of drainage in a study. The same process was
repeated using the study time frame in years instead of water table
difference to assess if the duration of the study had an effect on which
functional groups became dominant.

Due the small number of studies containing species richness data, these
individual effect sizes were compared quantitatively rather than by
statistical methods.

Results

Scope of dataset

After completing the data extraction process, only nine papers were
found to have reported suitable data to use in our analysis, and only
three of these reported species richness per plot with variances (Table
1{reference-type="ref"
reference="tab:studySummary"}). However, since many papers examined
multiple sites, we extracted data for 16 cases of functional group
abundances and 5 cases for species richness. Since each functional group
case had multiple measurements, this gave us a total of 46 data points
to use in our meta-analysis (Figure
1).
This very small sample size severely limited the potential of our
statistical modelling.

Nearly all studies used a different measure for functional group
abundance, except for percent cover and tree volume per hectare which
were used in multiple studies (Table
1{reference-type="ref"
reference="tab:studySummary"}).

:::

Citation Country Cases Functional Abundance SR
Groups measure

[@andrewsPlantCommunityResponses2022] UK 1 Shrub, herb, Pin-touch N
moss counts

[@breeuwerDecreasedSummerWater2009] Sweden 1 Shrub, herb, Biomass N
moss production (g
m-2) and change
in production
(pin-touch
count)

[@hohtanenVegetationTeuravuoma1999] Finland 1 Tree Tree stand Y
volume (m3
ha-1)

[@kokkonenResponsesPeatlandVegetation2019] Finland 3 Tree Tree stand Y
volume (m3
ha-1)

[@kosterWaterLevelDrawdown2023] Finland 3 Tree, shrub, Max biomass N
herb production (g)

[@laineImpactOfLongTermDrawdownFunctionalTrait2021] Finland 3 Moss Percent cover N

[@maanaviljaImpactDrainageHydrological2014] Finland 1 Tree, shrub, Tree stand Y
herb, moss volume (m3
ha-1), percent
cover

[@murphyEffectsWaterTable2009] Finland 1 Tree, shrub, Below ground N
herb biomass (g m-2
y-1)

[@urbanovaVegetationCarbonGas2012] Czechia 2 Shrub, herb Vascular green N
area (m2 m-2)

: Summary of the number of cases, represented functional groups,
measurement used and if species richness (SR) was reported in the nine
studies used in the meta-analysis.
:::

Despite searching for papers from across the northern hemisphere, the
geographic extent of the papers used in the meta-analysis was limited to
Europe (Figure 2).
Three papers from North America were found to have suitable abstracts,
but each one had issues with data extraction. Of the nine final papers,
six were from Finland, with four from the same Lakkasuo peatland
complex, while the remaining three were from Sweden, Czechia and Wales.
Mosses had the least representation of all the functional groups, being
reported in only four papers, while herbs were reported in six of them.
All of the papers reporting species richness were from Finland.

Six of the studies used experimental methods for water table drawdown,
while the remaining three were observational studies comparing nearby
drained and undrained sites. The four experimental studies from Lakkasuo
peatland complex constructed ditches to drain part of the site and used
nearby intact sections as controls. The remaining two experimental
studies used pumps to simulate drought conditions
[@andrewsPlantCommunityResponses2022] or placed cores removed from an
intact bog into a tank with controlled water levels
[@breeuwerDecreasedSummerWater2009].

Model results

The one-variable multilevel linear mixed-effect model found a
significant difference in effect size between groups, based on the Test
of Moderators ($Q_M(df=4)=10.0036,p=0.0404$). Additionally, there
was a highly significant positive effect on tree abundance after water
level drawdown ($z=+1.3028,p=0.0020$), while all the other functional
groups had a pooled effect size close to zero and $p>0.5$ (Figure
3). Despite the very
small effect size and confidence intervals spanning zero in all of the
other functional groups, the effect size estimate for mosses was
negative ($z=-0.2323,p=0.5297$) and very slightly positive for shrubs
($z=+0.0567,p=0.8947$) and herbs ($z=+0.0384,p=0.9202$).

The variance component estimate from the Paper ID / ID nested random
effect was very high (${\sigma }_2^2=0.9104$), indicating that nearly all
the random effect related variance in the model came from differences
between papers, and our data points were not very close to each other.
The model also showed a highly significant value from the test for
residual heterogeneity ($Q_E(df=27)=256.2995,p<.0001$),
indicating our effect sizes have almost no overlap with each other. This
heterogeneity is also supported by the fact that nearly all the
confidence intervals span across zero.

No significant results were found from the two-variable model which
included water table depth (Figure 4{reference-type="ref"
reference="fig:wtd_fg"}). Similarly to the one-variable model, most of
the variance was between papers (${\sigma }_2^2=0.8727$), and the data
displayed a high level of residual heterogeneity as shown by the
significant $Q_E$ score. However, in this model no significant
difference was found between the 8 crossed moderator groups
($Q_M(df=8)=14.7521,p=0.0641$). This model also had a lower
Akaike Information Criterion (91.5 vs 96.4) than the fist one,
indicating less goodness of fit.

Herbs, shrubs and trees had negative slopes of $z=-0.0451$, $-0.1124$
and $-0.0435$ respectively, indicating that very low water tables had a
larger positive effect on their abundances. Mosses had a slope very
close to zero ($0.0096$), indicating that they were not greatly affected
by water table depth, although this interpretation is limited by the
small group size ($k=6$ in four papers).

No clear trend in effect size or significant results were apparent when
using the study duration as a modifier in a multilevel model with
functional groups, despite its large range from two years to 77.

The effect sizes for the five species richness data points were mostly
positive, indicating that species richness usually increased after water
level drawdown (Table 2{reference-type="ref"
reference="tab:srES"}). The two longer duration studies all showed an
increase in species richness after drainage, while the short term (15
year) study showed a decrease in two out of three sites.

::: {#tab:srES}
+---------------------------------------------+-------------+----------+-----------+----------+-------+
| Citation | Site | Duration | Effect | Average | $n$ |
| | description | (y) | size | variance | plots |
+:===:+:=:+:=+:=:+:=+:+
| [@hohtanenVegetationTeuravuoma1999] | Fen | 61 | $+0.7975$ | $0.4340$ | 6 |
+---------------------------------------------+-------------+----------+-----------+----------+-------+
| [@maanaviljaImpactDrainageHydrological2014] | Spruce | 77 | $+0.2839$ | $0.2528$ | 9 |
| | swamp | | | | |
+---------------------------------------------+-------------+----------+-----------+----------+-------+
| [@kokkonenResponsesPeatlandVegetation2019] | Rich fen | 15 | $-0.1998$ | $0.0516$ | 40 |
| +-------------+----------+-----------+----------+-------+
| | Poor fen | 14 | $-0.4851$ | $0.0528$ | 40 |
| +-------------+----------+-----------+----------+-------+
| | Bog | 15 | $+0.2835$ | $0.0518$ | 40 |
+---------------------------------------------+-------------+----------+-----------+----------+-------+

: Effect size (standardized mean difference with heteroscedastic
population variances) and variance on species richness per plot after
water level drawdown.
:::

Discussion

The sign of the pooled effect size estimates within the functional
groups generally matches our hypothesis. Most obvious are the trees,
which had a very clear and significant increase in abundance following
water level drawdown (Figure 3{reference-type="ref"
reference="fig:fg"}). This is consistent with our hypothesis and
established research suggesting that boreal tree growth is strongly
limited by excess soil moisture [@larsonTreeGrowthPotential2024].
However, one study showed a decrease in tree biomass production: the
rich fen in Lakkasuo after 20 years of water level drawdown
[@kosterWaterLevelDrawdown2023]. This is likely an issue with the data
extracted for this study, which was biomass production of tree
seedlings. The paper describes the rich fen as having developed a dense
tree canopy cover, but data on biomass production of all the trees was
not measured [@kosterWaterLevelDrawdown2023]. If there was a dense tree
cover in the drained site, the new tree seedlings may have had their
growth inhibited by the tree cover compared to the control site. A
future analysis should probably remove these tree seedling measurements
from consideration due to this complication, or use standing tree volume
instead of seedling biomass production.

In the other functional groups, two of the signs of the pooled effect
sizes match our hypothesis (shrubs generally increased and mosses
generally decreased). However, none of these results were significant,
and each group had data points with both positive and negative effect
sizes, indicating that there was a high degree of variability in the
types of responses between studies. The very small positive pooled
effect size in herbs and shrubs is not enough to draw any conclusion
from.

The clearest trend in effect size was in the mosses, which had the
second-lowest p-value and the most extreme z-value other than the trees.
Two of the studies which reported very small changes or increases in
moss abundance
[@andrewsPlantCommunityResponses2022; @laineImpactOfLongTermDrawdownFunctionalTrait2021]
also reported that the species of moss present shifted from
moisture-loving to other species more adapted to dry conditions, while
the overall amount of moss didn't change much. These results suggest
that the peatland-forest succession process is not as straightforward as
expected, and that changes in shrub, herb and moss biomass post-drainage
are not as simple as expected, and that perhaps another grouping would
be a more appropriate way to model this transition.

The second model did not have any significant results, but nevertheless
the trends were somewhat interesting. The slope of the water table
difference regression by functional group showed that herbs, shrubs and
trees tended to increase in abundance most in the studies with the
largest differences in water table depth between control and treatment
sites (Figure 4{reference-type="ref"
reference="fig:wtd_fg"}). This makes sense for trees and shrubs, whose
deep roots would still be limited by more moderate drainage, while
non-vascular mosses are affected the same as soon as the water has been
drained below the surface. However, this analysis is very limited by the
small sample size, and these trends are not nearly robust enough for
serious consideration.

The high amount of between-study heterogeneity in both models makes
sense, considering that nearly every study used different techniques and
measurements (Table 1{reference-type="ref"
reference="tab:studySummary"}) and was preformed under very different
conditions. A future study could reduce this heterogeneity by broadening
the search terms and being more selective with the units and techniques
of data extracted for the meta-analysis.

While the lack of data meant we could not establish any clear
statistical trends in species richness, it appears that most of the
effect sizes calculated contradict our hypothesis that there we would
see a decrease in species richness after draining peatlands. Species
richness is a poor metric for assessing the biodiversity value of
peatlands, whose value comes from containing specialized species that
are not found in other habitats
[@lockyPlantDiversityCompositionManitoba2006; @perrinVegetationRichnessIreland2020].
In fact, it makes sense that draining could actually increase the plant
species richness of bogs by making them resemble more species-rich
nearby forest habitats, or by adding forest species into surviving
peatland specialists. Instead of species richness, a future study could
assess the biodiversity impact of peatland draining by looking at beta
diversity between the site and surrounding area, assessing community
heterogeneity, or assigning a rarity coefficient to the bog species and
ranking the sites based on species richness scaled with rarity
[@perrinVegetationRichnessIreland2020].

Overall, this study was severely limited by the low sample size, spatial
bias from southern Finland (specifically Lakkasuo, the source of 27/46
data points used in the study). Removing the studies related to
restoration greatly restricted the number of papers that were reviewed,
although this was necessary in order to keep this project within its
limited scope. Publication bias was mostly avoided since most of the
papers didn't set out to examine changes to functional groups, but
usually were looking at some other related variable. Despite these
deficiencies, the trends were overall as expected, and I suspect that
restarting the analysis with a new search term and more robust data
collection could result in more statistically significant values.

If more time was available, a future study could look more into
performing outlier analysis, which was not possible in this time frame
and due to the small sample size. In addition, using a within-group
measure for effect size might be more appropriate for the studies which
compared the same site before and after treatment. Using a Bayesian
approach with prior expected probabilities for the changes in functional
group abundances could also yield interesting results.

Although we could not prove our hypothesis completely, the trends in
tree size matched what was expected. The lack of clear changes in
shrubs, herbs and mosses indicates that the peatland-forest succession
process may not be as simple as increases in these functional groups,
and more complex modelling may be required to demonstrate this process.
The data on species richness was too sparse to prove anything about
species richness, but our results suggest that perhaps water level
drawdown might actually increase species richness in peatlands, but an
investigation of rarity or beta-diversity might demonstrate the
biodiversity value of peatlands.

Acknowledgements

Topic and research question selection, data extraction and statistical
analysis were done in collaboration with my colleague Md. Rifath Ahamed.
Guidance and support on the research process was provided by Dr. Helen
Phillips. Data interpretation was primarily guided by the textbook
"Doing Meta-Analysis with R: A Hands-On Guide" [@harrer2021doing]. No
artificial intelligence tools were used in the writing of this report,
except for accidentally reading AI-generated Google search responses
when debugging R code.

Data availability

The full dataset and all code used for data extraction and meta-analysis
is available on the author's GitHub page:
https://github.com/north-ross/NorthRoss_MacroecologyFinalReport.