Access to this full-text is provided by IOP Publishing.
Content available from Environmental Research Letters
This content is subject to copyright. Terms and conditions apply.
Environ. Res. Lett. 17 (2022) 034005 https://doi.org/10.1088/1748-9326/ac4cf3
OPEN ACCESS
RECEIVED
6 August 2021
REVISED
28 December 2021
ACC EPT ED FOR PUB LICATI ON
19 January 2022
PUBLISHED
21 February 2022
Original content from
this work may be used
under the terms of the
Creative Commons
Attribution 4.0 licence.
Any further distribution
of this work must
maintain attribution to
the author(s) and the title
of the work, journal
citation and DOI.
LETTER
Windows into the past: lake sediment phosphorus trajectories act
as integrated archives of watershed disturbance legacies over
centennial scales
Ruchi Bhattacharya1, Simon G M Lin2and Nandita B Basu1,2,3,∗
1Department of Earth and Environmental Sciences, University of Waterloo, Waterloo, ON, N2L 3G1, Canada
2Department of Civil and Environmental Engineering, University of Waterloo, Waterloo, ON, N2L 3G1, Canada
3Water Institute, University of Waterloo, Waterloo, ON, N2L 3G1, Canada
∗Author to whom any correspondence should be addressed.
Keywords: watershed disturbance legacy, legacy phosphorus, sediment phosphorus, fertilizer inputs, watershed and lake morphology
Supplementary material for this article is available online
Abstract
Historic land alterations and agricultural intensification have resulted in legacy phosphorus (P)
accumulations within lakes and reservoirs. Internal loading from such legacy stores can be a major
driver of future water quality degradation. Yet, little is known about the magnitude and spatial
patterns of legacy P accumulation in lentic systems, and how watershed disturbance trajectories
drive these patterns. Here, we used a meta-analysis of 113 paleolimnological studies across 124
lakes and four reservoirs (referred here on as lakes) in 20 countries to quantify the linkages between
the 100 year trajectories of P concentrations in lake sediments, watershed inputs, and lake
morphology. We find five distinct clusters for lake sediment P trajectories, with lakes in the
developing and developed world showing distinctly different patterns. Lakes in the developed
world (Europe and North America) with early agricultural intensification had the highest sediment
P concentrations (1176–1628 mg kg−1), with a peak between the 1970–1980s and a decline since
then, while lakes in the developing world, specifically China, documented monotonically
increasing sediment P concentrations (857–1603 mg kg−1). Sediment P trajectories reflected
watershed disturbance patterns and were driven by a combination of anthropogenic drivers
(fertilizer input and population density) and lake morphology (watershed to lake area ratio).
Specifically, we found the largest legacy accumulation rates to occur in shallow lakes experiencing
long-term land-use disturbances. These links between land-use change and P accumulation in
lentic systems can provide insights about inland water quality response and help to develop robust
predictive models useful for resource managers and decision-makers.
1. Introduction
Exponential increase in population and food
demands in the past century have led to agricul-
tural intensification and land use change (Tilman
et al 2011). Long term changes in cropping pat-
terns, excessive fertilizer applications, and livestock
operations have shifted the global phosphorus (P)
cycle outside its natural historic range of variability
(Vollenweider 1971) and contributed to eutrophica-
tion and hypoxia in inland and coastal waters. Algal
blooms and hypoxia is a long standing concern in
lakes around the world (Schindler 2012, Michalak
et al 2013, Nürnberg et al 2013, Ho et al 2019).
Decades of efforts to manage watershed P loading
have yielded limited results. Despite reduction in total
P loads observed in few systems (Sharpley et al 1996,
Baker et al 2014), most lake P concentrations seem to
have either attained stasis or continued to increase in
various regions around the world (Sharpley et al 2013,
Oliver et al 2017). It is now recognized that this lack
of success in water quality improvement can be par-
tially attributed to the build-up of excess P in soils,
groundwater, rivers, and lake sediments over decades
© 2022 The Author(s). Published by IOP Publishing Ltd
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
of agricultural intensification and increasing popula-
tion densities (MacDonald and Bennett 2009, Keatley
et al 2011, Haygarth et al 2014, Goyette et al 2016,
2018, Powers et al 2016, Stackpoole et al 2019, van
Meter et al 2021). These legacies contribute to elev-
ated concentrations and often decadal lags in eco-
system improvements even when inputs have ceased
(Jarvie et al 2013, Sharpley et al 2013, van Meter et al
2018, Basu et al 2022).
Lake sedimentary deposits can serve as integrated
archives of P accumulation, and watershed disturb-
ance, and in turn provide insights on the deviations
from baseline ecosystem conditions and the tem-
poral trends in water quality degradation (Smol 1992,
Dearing et al 2008, Bhattacharya et al 2016, Edlund
et al 2017). Such long-term perspectives can be a
critical benchmark for validating watershed nutri-
ent transport and retention models (Anderson et al
2006, van Meter et al 2017), and for effective man-
agement of inland and coastal aquatic ecosystems
(van Meter et al 2018). Typically, studies of lake sed-
iment P trajectories have focused on reconstructing
ecological evolution of lakes using sediment strati-
graphic records of ecological indicators (e.g. diatoms
and algal pigments), along with radio-active isotopes
(e.g. 210Pb and 137 Cs) used to establish the chrono-
logical history (Appleby 2008). These studies have
mostly focused on understanding the drivers of sedi-
ment P dynamics in individual lakes (Engstrom et al
2006, Otu et al 2011, Levine et al 2018). Regional-
scale studies that explore P trajectories are limited,
albeit a few exist, such as a study across two dozen
lakes in Finland (Tammeorg et al 2018), or a range of
shallow lakes in Southern Florida, USA (Kenney et al
2016). Using a meta-analysis approach, Carey and
Rydin (2011) show that P burial patterns are repres-
entative of lake trophic status, alluding to the linkage
between lake sediment and watershed inputs. Keat-
ley et al (2011) used trajectories of sediment diatom
assemblages to reconstruct lake water P concentra-
tions and relate them to watershed disturbances. The
study largely focused on the lakes in North America
and Europe, and further warrants an exploration of
sediment P trajectories across a wider geographical
area.
Here, we build on this body of work and use
a meta-analysis approach to explore century scale
trajectories in lake sediment P across the world,
and relate these trajectories to watershed disturbance
legacies. First, we use a machine learning approach
to identify dominant clusters of sediment P traject-
ories, and analyze the relationship between the tem-
poral trajectories of the clusters and the correspond-
ing land use trajectories at the cluster level. The cluster
level analysis is complemented by lake-level analysis
where we explore the relationships between the total
P accumulation in the sediment, and cumulative P
inputs from the watershed. The cluster level analysis
allows us to explore the temporal trajectories, while
lake-level analysis allows us to better elucidate dom-
inant controls on total accumulation.
2. Methods
2.1. Data synthesis for meta analysis
We used the following keywords ‘paleolimnology’,
‘phosphorus’, ‘lake’, ‘sediment’, ‘core’ to search the
Scopus database (2020) for publications related
to lake sediment cores. We applied additional fil-
ters based on subject areas (i.e. Earth and Planet-
ary Science, Environmental Science, Biological and
Agricultural Science and their related subject areas),
document type (article), and language (English).
Availability of 210Pb chronology was used as addi-
tional key criteria for paper selection. Chronologic-
ally constrained sedimentary records were essential
for our study as it allowed the comparison of sediment
P trajectories with watershed disturbance legacies.
Our Scopus database search resulted in a total of 600
papers, and additional screening for 210Pb chronology
led to final 200 papers. Then, we used the ‘digitize’
R package (Poisot et al 2016) to extract the sediment
P trajectories and age-depth chronology from these
papers to be used for further analyses.
The P trajectories for each sediment core was
obtained by plotting the digitized depth-specific sed-
iment P concentrations (Psed; mgP kg−1dry sedi-
ment) against its depth-chronology extracted from
the publications. Age chronologies are developed
using radiometric dating techniques (e.g. 210Pb iso-
topes; Appleby 2008) that inform about the age/year
at which the sediment P was deposited at the lake bot-
tom. For lakes with multiple sediment cores, a basin-
wide single sediment P trajectory was estimated by
aggregating the individual cores chronologically.
2.2. Lake and watershed attributes
Lake coordinates and mean lake depth were extrac-
ted from the published journal articles. Lake area
was estimated from the lake polygon database Hydro-
LAKES (Lehner et al 2008). We delineated the drain-
age areas of the lakes using the digital elevation model
from the global hydrographic database HydroSHEDS
(Lehner et al 2008). We used six metrics to describe
watershed disturbance: (a) Cper (%) =crop per-
cent, (b) Pfert (kg ha−1watershed area/year) =P
fertilizer input per unit watershed area, (c) Pfertn
(kg ha yr−1)=Pfert ×W:Larea, (d) Np(per-
sons ha−1)=population density, and (e) Npn(per-
sons ha−1)=Np×W:Larea. The normalization of
the fertilizer inputs and the population density by
the watershed to lake area ratio allows us to integ-
rate both watershed inputs and lake morphology.
The annual trajectory (1900–2007) of crop percent-
age (Cper) was obtained from a historical cropland
dataset (Ramankutty and Foley 1999) aggregated to
the watershed scale by multiplying the percent crop to
watershed area. P fertilizer trajectories (1900–2013)
2
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
were estimated from the dataset of P fertilizer use
on croplands (kg ha−1cropland yr−1) (Lu and Tian
2016), and the corresponding cropped area of each
watershed. In addition to fertilizers, manure applic-
ation to croplands can be an important source of P
to the lakes, though fertilizer still constitutes between
65% and 70% of the inputs (Bouwman et al 2017,
van Puijenbroek et al 2019, Glibert 2020). Current
long term data availability limits the use of the
manure dataset, but this should be explored in future
work. Finally, gridded population density trajector-
ies (1900–2010) were obtained from the HYDE 3.2
dataset (Goldewijk et al 2010). It is important to note
that these reconstructed data products, though widely
used, have uncertainties that might confound our
interpretations; however, these are some of the most
widely used datasets and the uncertainties are much
less in the more recent timeframe (1960s to current).
2.3. Statistical methods
To explore the spatio-temporal variability in lake
sediment P trajectories, we utilized a two-pronged
approach. First, we used a hierarchical cluster-
ing method to identify the major groups of lakes
with similar temporal patterns (section 2.3.1). The
median sediment P concentration trajectories for
each cluster were then compared with the potential
drivers described above in section 2.2. The cluster
level analysis was complemented by a lake-level ana-
lysis where we analyzed the dominant controls of
lake-sediment P accumulation using ordination and
regression methods (section 2.3.2). The lake level
analysis presents temporally aggregated metrics and
allows us to explore spatial variability and domin-
ant controls. In contrast, the cluster level analysis is
aggregated spatially but allows us to explore temporal
trajectories.
2.3.1. Hierarchical clustering of sediment P trajectories
and changepoint analyses
Agglomerative hierarchical clustering—an unsuper-
vised machine learning algorithm that utilizes a
‘bottoms up approach’ (Kaufman and Rousseeuw
2005), was used to find lake groups with similar
lake sediment P trajectories across all our study lakes
(n=128). Hierarchical clustering is a well established,
and robust technique that does not require prede-
termined number of clusters as an initial parameter,
and can work with time series data with varying
levels of similarities, which makes the method more
appropriate for the real world time-series data (Agh-
abozorgi et al 2015). As such, hierarchical clustering
has been successfully used for clustering long-term
water quality and sediment time series (e.g. (Astel
et al 2006, Legendre et al 2012, Dugan et al 2017,
Byrnes et al 2020). To allow for cross-site comparison,
we developed z-score trajectories by standardizing
Psed values in each core to a mean of zero and vari-
ance of one. The z-score trajectories were smoothed
using a moving average (span =3) (TTR package v.
0.24.2, Ulrich 2016) to remove data noise or abrupt
inter-annual variability (Bigler and Hall 2003). The
smoothed z-score trajectories were clustered using
Ward’s linkage hierarchical clustering method based
on squared Euclidean distances that minimises within
group dissimilarity at each linkage (Ward 1963). We
explored Psed trajectories between 1900 (±5 years)
to 2010 (±5 years), which resulted in some traject-
ories of unequal length. To account for unequal tra-
jectories and an additional check for our clusters, we
used the dynamic time warping as the distance meas-
ure in our hierarchical clustering (Aghabozorgi et al
2015). All the clustering outcomes were evaluated by
observing the cluster dendrograms and using the sum
of squared errors (SSE) between clusters to test the
cluster ‘coherence’, with better clusters having low SSE
(Aghabozorgi et al 2015). Finally, changepoint ana-
lysis was also conducted for each trajectory (change-
point package v. 2.2.2, Killick and Eckley 2014),
to statistically detect any prominent shift in Psed
trajectories.
2.3.2. Exploring lake level variability in sediment P
trajectories and their controls
To understand spatial controls on sediment P traject-
ories at an individual lake level for all the study lakes
(n=128), we used the Principal Component Ana-
lysis (PCA), an ordination based analysis technique
(Ter Braak and Šmilauer 2002). We used two met-
rics to integrate across the temporal trajectory in sed-
iment P: (a) the rate of sediment P accumulation (PR;
mg kg yr−1) calculated as the difference in P concen-
trations between 1900 (±5 years)–2007 (±5 years)
divided by the duration of P trajectory, and the (b)
cumulative sediment P concentration totals over the
duration of 1960 (±5 years)–2007 (±5 years) (PM;
mg kg−1). These two metrics were used to assess the
spatial variability in P trajectories and allow compar-
ison with watershed and lake variables, including lake
surface area (Larea; km2), ratio of watershed to lake
area (W:Larea), average lake depth (D; m), and water-
shed disturbance indicators averaged over the dura-
tion of 1960–2007, including crop percent (¯
Cper; %),
and population density (¯
Npn; persons ha−1). We also
used the cumulative fertilizer inputs (¯
Pfert; kg) and
cumulative normalized fertilizer input (¯
Pfertn), as the
summation of the Pfert and Pfertnmetrics (section 3.4)
over the timeframe 1960–2007.
3. Results and discussion
3.1. Metadata synthesis
Our final database consisted of 124 lakes and four
reservoirs (from here on both lakes and reservoirs will
be referred to as lakes) spanning across 20 countries
and seven continents (figure 1(A); table S1 available
online at stacks.iop.org/ERL/17/034005/mmedia).
Largest number of lakes in our study were in China
3
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
Figure 1. (A) Global distribution of study lakes on a world map. Study sites are color coded based on lake depth. Lakes deeper
than 25 m are depicted in grey color. Inset map shows the location of lakes in the UK and Ireland, eastern USA, and eastern
China. Figures 1(B)–(E) The distribution (histogram) of the Lake area (Larea; km2), ratio of watershed to lake area (W:Larea), mean
lake Depth (D; m), and percent crops (Cper; %) in the year 2007 for all the study sites is also presented.
(n=43), followed by the USA (n=37), UK (n=23),
Canada (n=6), and Japan (n=5). The rest of
the 14 countries had less than 5 lakes in each. As
such, the lakes located between 20 ′N and 60 ′N are
over-represented, while the arctic and tropics are
under-represented in our dataset. The distribution
of lake surface area (Larea; km2), watershed-lake area
ratio (W:Larea), lake depth (D; m), and %watershed
that is cropped (Cper; %) (figures 1(B)–(E)) across
our study sites represent a gradient of lake mor-
phological and watershed characteristics. In terms
of lake morphology, 50% of the sampled lakes were
shallow (<5 m), and another 25% of the lakes had
depths ranging between 5 and 10 m, while 25% were
deeper lakes (figure 1(D)). The surface area spanned
between 0.022–6.8 ×104km2, with 40% of the lakes
having a surface area <10 km2(figure 1(B)). Land
use across the lake basins varied significantly (Cper
ranged from 0.3% to 85%), with 50% of lakes located
in pristine watersheds (Cper < 20%), 15% of lakes
in agriculture-dominated watersheds (Cper > 50%),
while in the remaining 35% of the lakes cropped area
varied between 20% and 50% (figure 1(E)). Thus,
our lake database over-represents pristine lakes in the
northern hemisphere, with less number of studies
from reservoirs and agriculturally impacted lakes in
the developing world, and thus may create gaps in
our understanding of global P accumulation.
3.2. Sediment P trajectories: cluster analysis
Sediment P concentration trajectories (Psed, mg
P kg−1), spanning over the last 100 years, exhibit
significant variability in space and time (figure 2).
Five distinct clusters were identified that described the
behavior of the 128 sediment P cores (figures 2(A)–
(F); table S1). Cluster 1 (C1) includes 51 lakes,
4
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
Figure 2. (A) Median sediment P concentration (Psed; mg P kg−1) trajectories for the five clusters; (B)–(F) shows the median
z-score and the inter quartile ranges of Psed trajectories for the five clusters. The red dashed line indicates the median changepoint
(CPT) for each cluster, while nindicates the number of lakes in each cluster, and (G) shows a bar plot of % of lakes from five
countries with >5 lakes in each cluster. The trajectories of clusters 4 and 5 were truncated in the year 2000 due to low (<20% of
total observations) post-2000s.
and is characterized by minimal temporal trend and
no significant changepoint over the 100 year time-
frame (figures 2(A) and (B)), and the lowest Psed
(figure 2(A)). Majority of the C1 lakes are located
in the USA (51%), followed by China (37%), and
Canada (7%) (figures 2(G) and 3). In contrast to the
C1 cluster, the other four clusters all show a clearly
defined changepoint after which there is a signific-
ant increase in Psed. The median changepoint for
cores in cluster 2 (C2) is 1978, while it is 1946 for
cluster 3 (C3), 1937 for cluster 4 (C4) and 1960 for
cluster 5 (C5) (figures 2(B)–(F)). C2 lakes have a
much later changepoint than the other clusters, and
exhibit a monotonically increasing concentration tra-
jectory beyond the changepoint. Lakes in this cluster
are located mostly in China (68%), followed by the
UK (18%), and USA (10%) (figures 2(G) and 3).
Agricultural intensification occurred much later in
China compared to the other regions of the world,
and this possibly explains the later changepoint.
C3 and C4 lakes show a peak around the 1990s
(figure 2(D)) and 1970s (figure 2(E)), respectively,
and a declining trend after that. The peak in C4 lakes
is earlier, while the decline is over a longer time-
frame and more dramatic than in C3 lakes. Lakes
in C3 (n=26; figures 2(C), (D) and (G)) and C4
5
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
Figure 3. Shows the global location of our study lakes color coded by the cluster group. Inset maps of the UK and Ireland, eastern
USA, and, eastern China show the study lake locations in more detail.
(n=12; figures 2(E) and (G)) belong primarily to the
developed world (C3: ∼40% lakes in USA and UK,
14% in China; C4: 63% in UK, in USA, and 13% in
Japan) (figures 2(G) and 3). Finally, C5 included only
11 lakes spread mostly across the US (43%) and UK
(29%), with few lakes in Finland. These lakes recorded
a steep increase in sediment P concentrations from
the 1960s till 2000s (figures 2(F) and 3).
We further find that the Psed between 1960 and
2010 in C3 (median: 19 210 mg kg−1) and C4
(median: 33 780 mg kg−1) lakes to be higher than
in C2 (median: 12 340 mg kg−1) lakes, alluding to
the higher degree of historical impact on lakes in the
developed world. The high Psed and the distinct peak
in the 1970s–1990s in C3 and C4 lakes are indicat-
ive of highly impacted systems that might be slowly
recovering. Indeed, the peak and decline response
in C3 and C4 lakes, is parallel to peak and decline
response that was observed in the P surplus trajector-
ies for the watersheds in the UK and USA (Powers et al
2016). C5 lakes also belong to the developed world,
but in contrast to C3 and C4 lakes, Psed is monotonic-
ally increasing, indicating no distinct recovery beha-
vior (figure 2(A)).
3.3. Land use and P input trajectories for the
clusters
To better understand the drivers of sediment P accu-
mulation, we then compared the Psed trajectories
(figure 2) with land use trajectories of their water-
sheds (figure 4). We found C1 lakes to have the lowest
percent cropland (Cper), which comprised the major-
ity of pristine lakes in our dataset, with no substan-
tial change over time (figure 4(A)), and this is con-
sistent with the corresponding lack of changepoint
in the Psed trajectories. There is also no significant
trend in Cper in both C3 and C4 lakes, although C4
lakes do show a decrease in percent cropland since
the 1940s that correspond to the changepoint of 1937
for their Psed trajectories (figure 2(E)). This is in con-
trast to C2 and C5 lakes where Cper increases over
time, although the increases are of a much smaller
magnitude than the corresponding Psed trajectories
(figures 2(C) and (F)). It is further interesting to note
that C2 lakes have the highest Cper but lower Psed than
C3 and C4 lakes (figures 2(A) and 4(A)). This suggests
that land use trajectories are possibly not the best pre-
dictors of lake sediment P accumulation. Indeed, fer-
tilizer application and population dynamics might be
a better indicator of Psed, as they more accurately cap-
ture the lake inputs. For example, agricultural activ-
ities in China (C2 lakes) started early contributing to
the high Cper, but the extensive use of P fertilizers is
a more recent phenomenon than in North America
and Europe (C3 and C4 lakes), and this possibly is
responsible for the more recent changepoint of the
1970s in the C2 lakes.
To explore the possible effects of watershed dis-
turbance further, we compared Psed trajectories with
P fertilizer use (Pfert; kg ha−1watershed area/year)
and population density (NP; persons ha−1) trajector-
ies in watersheds draining into the lake (figures 4(B)
and (E)). We find C1 lakes to have the lowest Pfert
and NP; this is consistent with their Psed trajectories
(figure 2(A)). In contrast, C2 lakes have a monoton-
ically increasing Pfert trajectory since the 1960s, while
C3, C4 and C5 lakes have had a peak Pfert between
1970 and 1990, and a declining trend since then. This
corresponds well with the monotonically increasing
Psed for the C2 lakes, and a peak and declining trend
6
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
Figure 4. Watershed disturbance trajectories for the five clusters. (A) Percent of the watershed area that is cropped (Cper ; %); (B) P
fertilizer inputs (Pfert; kg ha−1watershed area/year); (C) P fertilizer inputs normalized by the W:Larea ratio (Pfertn; kg ha yr−1).
Secondary Y-axis corresponds to cluster 4 trajectory (filled green circles). (D) Inset of fertilizer inputs for clusters 1–5, (E)
population density (Np; persons ha−1), and (F) population normalized by the W:Larea (Npn; persons ha−1). For each of the
variables we have presented the median trajectory for the cluster.
for the C3 and C4 lakes. Parallel also to the Psed tra-
jectories the peak in Pfert in C4 lakes precedes the peak
in C3 lakes. For C5 lakes, even though Pfert decreased
after the 1980s, the continued increase in population
density from 1960s-late 2000s likely contributed to
the increase in sediment accumulation (figure 4).
While temporal patterns in Pfert trajectories are
similar to Psed trajectories the absolute magnitudes
7
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
of Psed trajectories tell a different story. For example,
while C2 lakes have the largest Pfert, and Np,Psed in
these lakes is significantly lower than C4 and C5 lakes
(figure 2). Indeed, lakes with the highest Psed traject-
ories are C4, and C5 lakes followed by C3, C2 and C1
lakes, while the Pfert magnitudes are the largest for the
C2 lakes followed by C4, C3, C5 and C1 lakes. We
hypothesize that this mismatch occurs because Pfert
(figure 4(B)), which is the fertilizer inputs normalized
to the watershed area, might not appropriately cap-
ture the interaction between the lake and its water-
shed. For example, the same magnitude of fertilizer
inputs might impact a smaller lake differently than a
larger lake.
To explore this further, we compared the ranges
in the watershed to lake area ratio (W:Larea) between
the various clusters. We found that C4 lakes have the
highest W:Larea (median: 71; IQR: 41–135), while C1
and C2 lakes have the smallest W:Larea (for C2 median:
14; IQR: 8–43; figure S1). A higher W:Larea indicates
a larger watershed that contributes to greater inputs,
and a much smaller lake area to buffer the larger
input; this is possibly what leads to the higher Psed
magnitudes. In fact previous studies on watershed
and lake linkages have also established the unique
impact of W:Larea on nutrient loading and their sub-
sequent retention within lakes (Schindler 1971). To
explore the morphometric mediation of watershed
disturbances on Psed, we compared the W:Larea nor-
malized fertilizer inputs and population (Pfertnand
Npn) to the Psed trajectories (figures 4(C) and (F)).
Consistent with the Psed trajectories, C4 lakes have
higher magnitudes of Pfertncompared to C2 lakes. C5
lakes are an exception with Psed among the highest
and Pfertnmuch lower; however it is important to note
that C5 lakes have an extremely large range of W:Larea
ratios and the smallest number of lakes in its cluster,
making it difficult to make cluster-specific generaliza-
tions (figure S1). To summarize, we find that lake Psed
trajectories are driven by watershed inputs mediated
by watershed and lake morphology, and thus the Pfertn
metric that explicitly considers both these controls is
best able to describe the Psed trajectories.
3.4. Lake level assessment of sediment P
magnitudes and accumulation rates
We then complemented the cluster-level analysis with
lake-level analysis by exploring spatial patterns in sed-
iment P accumulation magnitudes and rates (PM;
mg kg−1and PR; mg kg yr−1) across the world. Our
samples were dominated by lakes in North Amer-
ica, Europe and China, with high accumulation mag-
nitudes apparent in eastern US, UK and eastern China
lakes (figure 5(A)). The UK lakes have the highest
PR(figure 5(B)) and PM(figure 5(C)), despite hav-
ing lower fertilizer inputs (¯
Pfert; cumulative fertilizer
inputs from 1960 to 2007) than lakes in North Amer-
ica or China (figure 5(D)). These are also the shallow-
est lakes, with the smallest depth (figure 5(E)), and
the largest watershed to lake area ratio (figure 5(F)),
leading to the largest normalized fertilizer inputs
(figure 5(G)). These results are consistent with pre-
vious studies that report the link between external P
loading and morphology on P accumulation in lakes.
For instance, elevated accumulation of Psed in smaller
lakes was also reported in Finnish (Tammeorg et al
2018) and other European lakes (Keatley et al 2011).
Lakes with higher W:Larea experience greater water-
shed inputs per unit lake area (Schindler 1971) that
potentially leads to higher sediment P accumulation.
We further use the PCA to understand the drivers
of P accumulation across all lakes (figure 6(A)). We
find that PCA axis 1 and 2 can together explain
53% of the variance of the PMand PRmetrics. We
found lake depth (D) and cumulative normalized fer-
tilizer inputs (¯
Pfertn, cumulative normalized fertilizer
inputs from 1960 to 2007) to emerge as two of the
strongest controls on accumulation (figure 6(A)). The
closer correspondence between PMand ¯
Pfertncom-
pared to PMand ¯
Pfert is consistent with the cluster
level analysis that argued for considering both water-
shed inputs and lake morphometry to explain lake
sediment P accumulation (figure 6(A)). Indeed, while
there exists a significant relationship between PMand
¯
Pfertn(figure 6(C) and table S2), with higher PM
in lakes with higher cumulative normalized fertil-
izer inputs (figure 6(B)), no significant relationship
exists between PMand ¯
Pfert (table S2). A significant
negative relationship also exists between PMand D
(table S2) with deeper lakes having smaller PMval-
ues (figure 6(D)). Depth emerges as a stronger con-
trol for the shallow UK lakes, with a long history of
development, while cumulative fertilizer inputs, aver-
age cropped area, and population appear to have a
greater influence for the lakes in China (figure 6(A)).
Recent fertilizer intensive agricultural practices have
been documented in several Asian countries, includ-
ing China (Lu and Tian 2016) that have caused the
sediment P enrichment (Rose et al 2011, Wu et al
2013). Our results highlight the differences in lake
PMbetween the developed countries and developing
countries, with developing country PMbeing more
strongly driven by cumulative fertilizer inputs, while
PMin developed countries (like UK) are driven by the
morphometry (W:Larea) (figure 6(A)).
The lake level analysis thus confirms our cluster
level analysis that found the normalized fertilizer
application rate (¯
Pfertn) to be the best predictor of
sediment P accumulation. This metric takes into
account both the cumulative fertilizer inputs to the
lake (¯
Pfert), and the lake morphometry that modulates
the response, with smaller, shallower having greater
productivity and thus accumulation magnitudes. The
finding is important and is interesting and import-
ant and suggests that in absence of lake sediment core
data, it might be possible to estimate lake sediment P
accumulation magnitudes as a function of watershed
P inputs and lake morphometry.
8
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
Figure 5. (A) Map showing the variability in sediment P accumulation between 1960 and 2007 (PM; mg kg−1) in all the study
lakes. Sites have been color coded based on PRranges, (B) rate of sediment P accumulation between 1900 and 2007 (PR;
mg kg yr−1), (C) sediment P concentration total between 1960 and 2007 (PM; mg kg−1), (D) cumulative of the W:Larea
normalized fertilizer input over years 1960–2007 (¯
Pfertn; kg ha yr−1), (E) cumulative fertilizer input over years 1960–2007 (¯
Pfert;
kg), (F) lake depth (D; m), and (G) ratio of watershed to lake area (W:Larea). Box and whiskers plots of these distributions are
presented for countries with >5lakes. See figure 1and table S1 for details about all the study lakes.
3.5. Conclusion and implications
As inland waters are experiencing increased
eutrophication and algal blooms, it is increas-
ingly important to understand the role of lakes in
mediating watershed scale nutrient export. Here, we
use a meta analysis approach to synthesize lake sedi-
ment P concentration trajectories in 128 lakes across
the world, and relate them watershed disturbance
9
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
Figure 6. (A) PCA bi-plot shows the relation between rate of sediment P accumulation between 1900 and 2007 (PR; mg kg yr−1)
and sediment P concentration total between 1960 and 2007 (PM; mg kg−1) with watershed and lake morphological variables,
including lake depth (D; m), and ratio between watershed and lake area (W:Larea), average percent cropland (¯
Cper; %), average
population density (¯
Nppersons ha−1) for the year 1960–2007, cumulative fertilizer input between 1960 and 2007 (¯
Pfert; kg), and
cumulative of W:Lare normalized fertilizer input between 1960 and 2007 (¯
Pfertn). The lakes were color coded based on their
location. Countries with more than five study lakes (n> 5) in our database, namely the UK, USA, Canada, China, and Japan are
indicated in the legend. Rest of the countries (n< 5 lakes) were shown by gray filled circles and the name of the country was
added next to the symbol (B)–(D). Plots show the variability in average PMacross ranges of cumulative ¯
Pfert, cumulative
normalized Pfertn, and depth (D).
legacies, specifically histories of agricultural intensi-
fication and urbanization over the last century.
Using a machine learning approach, we identify
five dominant lake sediment trajectory types, with
distinct trajectories observed for the developed and
developing regions of the world. We found 50% of
the lakes sampled to be relatively pristine, with the
lowest sediment P concentrations and no significant
temporal trends (cluster 1 lakes). A second category
of lakes had sediment P trajectories characterized by
a peak in the 1970s to 1990s followed by a more
recent decline (clusters 3 and 4). These lakes were
characterized by the highest P concentrations, and are
mostly located in the developed world (North Amer-
ica and Europe). In contrast, lakes in the developing
world (China) had lower P concentrations in lake sed-
iments, but an exponentially increasing accumulation
pattern (cluster two lakes). The cluster level analysis
10
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
was complemented by lake level PCA analysis to
identify dominant drivers of lake sediment P accumu-
lation. Both lake and cluster level analysis revealed the
importance of watershed inputs (cumulative fertilizer
inputs) and lake morphometry (depth and watershed
to lake area ratio) on sediment P accumulation. Spe-
cifically, we observed a positive correlation between
lake sediment P accumulation and cumulative fertil-
izer inputs to the watershed normalized by the water-
shed to lake area ratio, highlighting the combined role
of watershed inputs and lake morphology on sedi-
ment accumulation patterns.
Our findings highlight the effect of wastewater
management and agricultural development on the
timing of sediment P accumulation and eutroph-
ication across developing and developed countries.
Fertilizer-driven increase in Psed in developed coun-
tries occurred earlier (1960s–1980s), and the current
declining trends are driven by a decrease in cropped
area and fertilizer applications, as well as detergent P
bans and improvement of wastewater treatment prac-
tices that occurred across Europe and North America
in the 1970s. These observations are typical for the
developed world that had an early onset of watershed
anthropogenic alterations followed by an attempt
to manage and mitigate water quality issues in the
1970s–1980s (Müller et al 2007, Blumentritt et al
2013). In comparison, developing or in-transition
countries, such as China, experienced a later onset
of agricultural intensification, with extensive land
reclamation for rice production, and the associated
peaks in fertilizer application (Yousaf et al 2017). Des-
pite the delay in agricultural intensification, the expo-
nential rates of increase have led to rapid build-up
of sediment P stores and have turned these regions
into hotspots of eutrophication and coastal hypoxia
(Bouwman et al 2009, Fink et al 2018).
Our study has three novel contributions that have
implications for lake monitoring and management.
First, our datasets highlight a focus on pristine lakes,
and relatively fewer studies on reservoirs and more
impacted lakes, which limits our ability to manage
impacted systems. Further, the majority of study lakes
are located in the northern hemisphere and developed
countries, which creates gaps in our understanding of
global P accumulation from the more agriculturally
intensive, developing regions of the world. Second,
our analysis highlights that lake sediments are effect-
ive integrators of watershed disturbance patterns, and
can potentially act as a long term archive of water-
shed P loads. Indeed, given the lack of monitoring
data over long time scales, it might be possible to util-
ize lake sediment data to calibrate coupled watershed-
lake models. Finally, we find that the spatio-temporal
variability in lake sediment P accumulation is driven
by a combination of anthropogenic drivers (fertilizer
input and population density) and lake morphology
(lake depth, watershed to lake area ratio), with shal-
low lakes in regions of intensive agriculture having
the largest accumulation magnitudes. This highlights
the role of past watershed P inputs that can poten-
tially build up as legacy P in lakes, and act as a long
term source of P through internal loading, fuelling
eutrophication and algal blooms for years after water-
shed inputs have ceased (Jeppesen et al 2005, Orihel
et al 2017, van Meter et al 2021). Such insights about
inter-region variability in legacy P accumulation in
lakes is critical for the development of robust predict-
ive models that can be used for resource managers
and decision makers for the management of aquatic
resources.
Data availability statement
Papers used in the meta analysis are listed in the sup-
plementary table S1.
The source for fertilizer usage, percent crops, and
population data are included in the manuscript.
The data that support the findings of this study are
available upon reasonable request from the authors.
Acknowledgments
The research published in this paper is funded
through (1) the Lake Futures project which is a part
of the Global Water Futures program, with fund-
ing from the Canada First Research Excellence Fund,
and (2) Legacies of Agricultural Pollutants project
with funding from the Natural Sciences and Engin-
eering Research Council of Canada as a part of the
Water Joint Programming Initiative. We acknowledge
the help of Sydney Dolick, Celina Mohni, and Isabel
Drummond for their efforts in literature search and
data digitization.
ORCID iDs
Ruchi Bhattacharya https://orcid.org/0000-0001-
5657-9603
Simon G M Lin https://orcid.org/0000-0001-
5261-4977
Nandita B Basu https://orcid.org/0000-0002-
8867-8523
References
Aghabozorgi S, Seyed Shirkhorshidi A and Ying Wah T 2015
Time-series clustering—a decade review Inf. Syst. 53 16–38
Anderson N J, Bugmann H, Dearing J A and Gaillard M J 2006
Linking palaeoenvironmental data and models to
understand the past and to predict the future Trends Ecol.
Evol. 21 696–704
Appleby P G 2008 Three decades of dating recent sediments by
fallout radionuclides: a review Holocene 18 83–93
Astel A, Biziuk M, Przyjazny A and Namie´
snik J 2006
Chemometrics in monitoring spatial and temporal
variations in drinking water quality Water Res. 40 1706–16
Baker D B, Confesor R, Ewing D E, Johnson L T, Kramer J W and
Merryfield B J 2014 Phosphorus loading to Lake Erie from
the Maumee, Sandusky and Cuyahoga rivers: the
importance of bioavailability J. Great Lakes Res. 40 502–17
11
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
Basu N B et al 2022 Managing nitrogen legacies to accelerate water
quality improvement Nat. Geosci. 15 97–105
Bhattacharya R, Hausmann S, Hubeny J B, Gell P and Black J L
2016 Ecological response to hydrological variability and
catchment development: insights from a shallow oxbow lake
in Lower Mississippi Valley, Arkansas Sci. Total Environ.
569–570 1087–97
Bigler C and Hall R I 2003 Diatoms as quantitative indicators of
July temperature: a validation attempt at century-scale with
meteorological data from northern Sweden Palaeogeography,
Palaeoclimatology, Palaeoecology 189 147–160
Blumentritt D J, Engstrom D R and Balogh S J 2013 A novel
repeat-coring approach to reconstruct recent sediment,
phosphorus, and mercury loading from the upper
Mississippi River to Lake Pepin, USA J. Paleolimnol.
50 293–304
Bouwman A F, Beusen A H W and Billen G 2009 Human
alteration of the global nitrogen and phosphorus soil
balances for the period 1970–2050 Global Biogeochem.
Cycles 23
Bouwman A F, Beusen A H W, Lassaletta L, van Apeldoorn D F,
van Grinsven H J M, Zhang J and Ittersum Van M K 2017
Lessons from temporal and spatial patterns in global use of
N and P fertilizer on cropland Sci. Rep. 71–11
Byrnes D K, van Meter K J and Basu N B 2020 Long-term shifts in
US Nitrogen sources and sinks revealed by the new
TREND-nitrogen data set (1930–2017) Global Biogeochem.
Cycles 34 1–16
Carey C C and Rydin E 2011 Lake trophic status can be
determined by the depth distribution of sediment
phosphorus Limnol. Oceanogr. 56 2051–63
Dearing J A, Jones R T, Shen J, Yang X, Boyle J F, Foster G C,
Crook D S and Elvin M J D 2008 Using multiple archives to
understand past and present climate-human- environment
interactions: the lake Erhai catchment, Yunnan Province,
China J. Paleolimnol. 40 3–31
Dugan H A et al 2017 Salting our freshwater lakes Proc. Natl Acad.
Sci. USA 114 4453–8
Edlund M B, Schottler S P, Reavie E D, Engstrom D R,
Baratono N G, Leavitt P R, Heathcote A J, Wilson B and
Paterson A M 2017 Historical phosphorus dynamics in Lake
of the Woods (USA–Canada)—does legacy phosphorus still
affect the southern basin? Lake Reserv. Manage. 33 386–402
Engstrom D R, Schottler S P, Leavitt P R and Havens K E 2006 A
reevaluation of the cultural eutrophication of Lake
Okeechobee using multiproxy sediment records Ecol. Appl.
16 1194–206
Fink G, Alcamo J, Flörke M and Reder K 2018 Phosphorus
loadings to the world’s largest lakes: sources and trends
Global Biogeochem. Cycles 32 617–34
Glibert P M 2020 From Hogs to HABs: Impacts of Industrial
Farming in the US on Nitrogen and Phosphorus and
Greenhouse Gas Pollution vol 150 (Springer International
Publishing) pp 139–80
Goldewijk K K, Beusen A and Janssen P 2010 Long-term dynamic
modeling of global population and built-up area in a
spatially explicit way: HYDE 3.1 The Holocene
20 565–73
Goyette J O, Bennett E M, Howarth R W and Maranger R 2016
Changes in anthropogenic nitrogen and phosphorus inputs
to the St Lawrence sub-basin over 110 years and impacts on
riverine export Global Biogeochem. Cycles 30 1000–14
Goyette J O, Bennett E M and Maranger R 2018 Low buffering
capacity and slow recovery of anthropogenic phosphorus
pollution in watersheds Nat. Geosci. 11 921–5
Haygarth P M et al 2014 Sustainable phosphorus management
and the need for a long-term perspective: the legacy
hypothesis Environ. Sci. Technol. 48 8417–9
Ho J C, Michalak A M and Pahlevan N 2019 Widespread global
increase in intense lake phytoplankton blooms since the
1980s Nature 574 667–70
Jarvie H P, Sharpley A N, Spears B, Buda A R, May L and
Kleinman P J A 2013 Water quality remediation faces
unprecedented challenges from ‘legacy phosphorus’
Environ. Sci. Technol. 47 8997–8
Jeppesen E et al 2005 Lake responses to reduced nutrient
loading—an analysis of contemporary long-term data from
35 case studies Freshw. Biol. 50 1747–71
Kaufman L and Rousseeuw P J 2005 Finding Groups in Data: An
Introduction to Cluster Analysis (Wiley) (available at:
www.wiley.com/en-us/Finding+Groups+in+Data%3A+
An+Introduction+to+Cluster+Analysis-p-
9780471735786)
Keatley B E, Bennett E M, MacDonald G K, Taranu Z E and
Gregory-Eaves I 2011 Land-use legacies are important
determinants of lake eutrophication in the anthropocene
PLoS One 6e15913
Kenney W F, Brenner M, Arnold T E, Curtis J H and Schelske C L
2016 Sediment cores from shallow lakes preserve reliable,
informative paleoenvironmental archives despite
hurricane-force winds Ecol. Indic. 60 963–9
Killick R and Eckley I A 2014 Changepoint: an R package for
changepoint analysis J. Stat. Softw. 58 1–19
Legendre P, John H, Birks B, Birks H J B and Legendre P 2012
Clustering and Partitioning vol 5 (Dordrecht: Springer)
Lehner B, Verdin K L and Jarvis A 2008 New global hydrography
derived from spaceborne elevation data Eos, Trans. Am.
Geophys. Union 89 93–94
Levine S N, Lini A, Ostrofsky M L, Burgess-Grant H, Lami A,
Collyer-Gilles E, Reuter D, Schwarting-Miller L and
Kamman N 2018 The relative roles of point and nonpoint
phosphorus sources in the eutrophication of Lake
Champlain as recorded in sediment cores J. Great Lakes Res.
44 1043–56
Lu C and Tian H 2016 Global nitrogen and phosphorus fertilizer
use for agriculture production in the past half century:
shifted hot spots and nutrient imbalance Earth Syst. Sci.
Data Discuss. 91–33
MacDonald G K and Bennett E M 2009 Phosphorus accumulation
in Saint Lawrence River watershed soils: a century-long
perspective Ecosystems 12 621–35
Michalak A M et al 2013 Record-setting algal bloom in Lake Erie
caused by agricultural and meteorological trends consistent
with expected future conditions Proc. Natl Acad. Sci. USA
110 6448–52
Müller B, Finger D, Sturm M, Prasuhn V, Haltmeier T, Bossard P,
Hoyle C and Wüest A 2007 Present and past bio-available
phosphorus budget in the ultra-oligotrophic Lake Brienz
Aquat. Sci. 69 227–39
Nürnberg G K, Molot L A, O’Connor E, Jarjanazi H, Winter J and
Young J 2013 Evidence for internal phosphorus loading,
hypoxia and effects on phytoplankton in partially polymictic
Lake Simcoe, Ontario J. Great Lakes Res. 39 259–70
Oliver S K, Collins S M, Soranno P A, Wagner T, Stanley E H,
Jones J R, Stow C A and Lottig N R 2017 Unexpected stasis
in a changing world: Lake nutrient and chlorophyll trends
since 1990 Glob. Change. Biol. 23 5455–67
Orihel D M, Baulch H M, Casson N J, North R L, Parsons C T,
Seckar D C M and Venkiteswaran J J 2017 Internal
phosphorus loading in Canadian fresh waters: a critical
review and data analysis Can. J. Fish. Aquat. Sci. 74 2005–29
Otu M K, Ramlal P, Wilkinson P, Hall R I and Hecky R E 2011
Paleolimnological evidence of the effects of recent cultural
eutrophication during the last 200 years in Lake Malawi,
East Africa J. Great Lakes Res. 37 61–74
Poisot A T, Poisot M T, Allows R D, Gpl L and Repository R C
2016 Package ‘digitize’ pp 1–4
Powers S M et al 2016 Long-term accumulation and transport of
anthropogenic phosphorus in three river basins Nat. Geosci.
9353–6
Ramankutty N and Foley J A 1999 Estimating historical changes
in global land cover: croplands from 1700 to 1992 Global
Biogeochem. Cycles 13 997–1027
Rose N L, Morley D, Appleby P G, Battarbee R W, Alliksaar T,
Guilizzoni P, Jeppesen E, Korhola A and Punning J M 2011
Sediment accumulation rates in European lakes since AD
12
Environ. Res. Lett. 17 (2022) 034005 R Bhattacharya et al
1850: trends, reference conditions and exceedence J.
Paleolimnol. 45 447–68
Schindler D W 1971 A hypothesis to explain differences and
similarities among Lakes in the experimental Lakes Area,
Northwestern Ontario J. Fish. Res. Board Canada 28 295–301
Schindler D W 2012 The dilemma of controlling cultural
eutrophication of lakes Proc. R. Soc. B279 4322–33
Sharpley A, Jarvie H P, Buda A, May L, Spears B and Kleinman P
2013 Phosphorus legacy: overcoming the effects of past
management practices to mitigate future water quality
impairment J. Environ. Qual. 42 1308–26
Sharpley A, Smith S J, Zollweg J A and Coleman G A 1996 Gully
treatment and water quality in the southern plains J. Soil
Water Conserv. 51 498–503
Smol J P 1992 Paleolimnology: an important tool for effective
ecosystem management J. Aquat. Ecosyst. Heal. 149–58
Stackpoole S M, Stets E G and Sprague L A 2019 Variable impacts
of contemporary versus legacy agricultural phosphorus on
US river water quality Proc. Natl Acad. Sci. USA 116 20562–7
Tammeorg O, Haldna M, N˜
oges P, Appleby P, Möls T, Niemistö J,
Tammeorg P and Horppila J 2018 Factors behind the
variability of phosphorus accumulation in Finnish lakes J.
Soils Sediments 18 2117–29
Ter Braak C J F and Šmilauer P 2002 CANOCO Reference Manual
and CanoDraw for Windows User’s Guide: Software for
Canonical Community Ordination, Version 4.5 (Ithaca, NY:
Microcomputer Power)
Tilman D, Balzer C, Hill J and Befort B L 2011 Global food
demand and the sustainable intensification of agriculture
Proc. Natl Acad. Sci. USA 108 20260–4
Ulrich J 2016 Package ‘TTR’: Technical Trading Rules 59
van Meter K J, Basu N B and van Cappellen P 2017 Two centuries
of nitrogen dynamics: legacy sources and sinks in the
Mississippi and Susquehanna River Basins Global
Biogeochem. Cycles 31 2–23
van Meter K J, McLeod M M, Liu J, Tenkouano G T, Hall R I, van
Cappellen P and Basu N B 2021 Beyond the mass balance:
watershed phosphorus legacies and the evolution of the
current water quality policy challenge Water Resour. Res.
57 1–22
van Meter K J, van Cappellen P and Basu N B 2018 Comment on
‘legacy nitrogen may prevent achievement of water quality
goals in the Gulf of Mexico’ Science 360 427–30
van Puijenbroek P J T M, Beusen A H W and Bouwman A F 2019
Global nitrogen and phosphorus in urban waste water based
on the shared socio-economic pathways J. Environ. Manage.
231 446–56
Vollenweider R A 1971 Scientific Fundamentals of the
Eutrophication of Lakes and Flowing Waters, with Particular
Reference to Nitrogen and Phosphorus as Factors in
Eutrophication (Organisation for Economic Co-operation
and Development Paris)
Ward J H 1963 Hierarchical Grouping to Optimize an Objective
Function J Am Stat Assoc 58 236
Wu J, Ma L, Yu H, Zeng H, Liu W and Abuduwaili J 2013
Sediment geochemical records of environmental change in
Lake Wuliangsu, Yellow River Basin, north China J.
Paleolimnol. 50 245–55
Yousaf M, Li J, Lu J, Ren T, Cong R, Fahad S and Li X 2017 Effects
of fertilization on crop production and nutrient-supplying
capacity under rice-oilseed rape rotation system Sci. Rep.
71–9
13
Content uploaded by Simon Lin
Author content
All content in this area was uploaded by Simon Lin on Feb 22, 2022
Content may be subject to copyright.
