Mushroom fruiting and climate change
Håvard Kauserud*, Leif Christian Stige
, Jon Olav Vik
, Rune H. Økland
, Klaus Høiland*, and Nils Chr. Stenseth
*Microbial Evolution Research Group and
Centre for Ecological and Evolutionary Synthesis, Department of Biology, University of Oslo,
P.O. Box 1066 Blindern, NO-0316 Oslo, Norway; and
Department of Botany, Natural History Museum, University of Oslo, P.O. Box 1172 Blindern,
NO-0318 Oslo, Norway
Edited by Hans R. Herren, Millennium Institute, Arlington, VA, and approved January 22, 2008 (received for review September 23, 2007)
Many species of fungi produce ephemeral autumnal fruiting bodies
to spread and multiply. Despite their attraction for mushroom
pickers and their economic importance, little is known about the
phenology of fruiting bodies. Using ⬇34,500 dated herbarium
records we analyzed changes in the autumnal fruiting date of
mushrooms in Norway over the period 1940 –2006. We show that
the time of fruiting has changed considerably over this time period,
with an average delay in fruiting since 1980 of 12.9 days. The
changes differ strongly between species and groups of species.
Early-fruiting species have experienced a stronger delay than late
fruiters, resulting in a more compressed fruiting season. There is
also a geographic trend of earlier fruiting in the northern and more
continental parts of Norway than in more southern and oceanic
parts. Incorporating monthly precipitation and temperature vari-
ables into the analyses provides indications that increasing tem-
peratures during autumn and winter months bring about signiﬁ-
cant delay of fruiting both in the same year and in the subsequent
year. The recent changes in autumnal mushroom phenology coin-
cide with the extension of the growing season caused by global
climate change and are likely to continue under the current climate
phenology 兩 global warming 兩 herbarium data 兩 fungi 兩 agarics
henological changes are among the most sensitive ec ological
responses to changing climate (1–3). The observed extension
of the average annual growing season in Europe by nearly 11
days since the early 1960s (4) has been followed by rapid and
recent changes in plant flowering time (5–8) and earlier spring
migration in several bird species (9). In a recent study from the
Un ited Kingdom, it was reported for a set of mushroom species
that fruiting on average started earlier and ended later in the
season in recent years than 20 years ago, i.e., that the fruiting
period has been greatly extended (10). These changes were
linked to increased temperature and rainfall in August and
October, respectively (10).
Most fungi produce ephemeral fruiting bodies that can be
observed only for a few days each year, which makes phenolog-
ical dat a difficult and time-consuming to obtain. However,
because of the short endurance of the fr uiting bodies, collection
time is a good estimate of fruiting time. A potential source of
phenological information for this group of organisms is therefore
herbarium c ollections, which, although sampled in a nonsystem-
atic manner, share properties with random sampling processes.
Herbarium data can enable us to understand and predict
climate-induced ec ological changes in the future by understand-
ing how climate has affected ec ological processes in the past.
Several studies have already documented that herbarium col-
lections may represent a valuable source of long-term and
reliable phenological information (e.g., refs. 7 and 8).
Our study of temporal trends in fruiting phenology is based on
⬎34,500 herbarium records collected in Norway during the
period 1940–2006 and representing 83 agaricoid (mushroom)
species [supporting infor mation (SI) Table 1]. By thorough
analyses of these data we aim to est ablish quantitative relation-
ships between climate (and climate change) and fungal autumnal
f ruiting time in Norway.
Results and Discussion
In an analysis of variance, the observed variation in fr uiting dates
can be partitioned into bet ween-species differences (15.8%),
variabilit y within species between years (25.9%), and variability
within species within years (58.3%). Using generalized additive
models (GAMs) (11) (see below and SI Table 2 for model
descriptions), we find that geographic differences in fruiting time
(across all species) explain 3.5% of the tot al variation, that
temporal trends (across groups of species) explain 3.9% of the
variation, and that 7.2% of the variation can be attributed to
shared responses of species to interannual variability in temper-
ature and precipitation (no temporal trend term in the model)
(P ⬍ 0.001 in all cases; bootstrap tests).
Mushroom fr uiting date changed considerably during the
period 1940–2006, with earlier fruiting in the early years (1940–
1950) and later fruiting in the last 15 years (Fig. 1A). On average
across the period (and across all species) there has been a delay
in fr uiting of 13.3 ⫾ 1.2 days [linear rate of change per 60 years ⫾
bootstrap standard error (P ⬍ 0.001); GAM, correcting for
location and species effects]. Most of the shift took place
bet ween the 1980s and the 2000s [12.9 ⫾ 1.2 days per 20 years
(P ⬍ 0.001); analysis of data 1980–2006]. The displacement of
f ruiting date parallels the delay of other autumn events, such as
leaf coloring being delayed by 4.8 days in Europe during the
period 1959–1993 in response to climate change (4). The delay
of mushroom fruiting does, however, contrast with the general
climate-induced advance in plant fruiting and ripening (1),
suggesting that constraints on fr uiting differ between fungi and
Fr uiting was more strongly delayed for the early autumnal
f ruiters, as revealed by a continuous interaction term fitted
bet ween initial fruiting day (1940–1959) and year (Fig. 1B).
Whereas fruiting of early f ruiters was delayed by ⬎30 days over
the entire period, late fruiters had no fr uiting-time delay. This
ac cords with a highly sign ificant linear relationship between the
in itial (1940–1959) mean day of fruiting for each species and the
displacement of fruiting time from 1940 to 2006 (Fig. 1C).
Studies of spring phenology i n plants have also shown differences
bet ween ‘‘early’’ and ‘‘late’’ species in their response to climate
change (5, 12). The stronger delay for early fruiters compared
with late fruiters implies that the start of the fruiting season has
been delayed while the end has remained more or less un-
changed. Thus, the mushroom fruiting season in Norway has
bec ome progressively more compressed into late autumn in
Nor way. A decrease in residual variation (most at the within-
species within-year level; restricted maximum-likelihood analy-
sis) with time also indicated that the length of the overall fruiting
Author contributions: H.K. and L.C.S. contributed equally to this work; H.K., L.C.S., K.H., and
N.C.S. designed research; H.K., L.C.S., J.O.V., R.H.Ø., and N.C.S. performed research; H.K.
and L.C.S. analyzed data; and H.K., L.C.S., J.O.V., R.H.Ø., K.H., and N.C.S. wrote the paper.
The authors declare no conﬂict of interest.
This article is a PNAS Direct Submission.
To whom correspondence should be addressed. E-mail: firstname.lastname@example.org.
This article contains supporting information online at www.pnas.org/cgi/content/full/
© 2008 by The National Academy of Sciences of the USA
March 11, 2008
season has been c ompressed (Fig. 1D and SI Text). Our results
c ontrast with Gange et al.’s recent finding that the mushroom
f ruiting season has ex panded in both directions in the United
K ingdom (10). We see no obvious reason why first and last
f ruiting dates in the United Kingdom should show trends
dif ferent from the early and late quantiles of Norwegian mush-
room fruiting dates (Fig. 1D), nor are we aware of differences in
the mechanisms controlling mushroom f ruiting between the
Un ited Kingdom and Norway. Studies of the cues and con-
straints that govern fungal fruiting might clarify this issue. Cues
might relate to autumnal events that occur later than before,
whereas constraints on resource acquisition and achieving ‘‘fruit-
ing potential’’ might be fulfilled earlier when the climate is
milder. In many mushrooms, f ruiting can be induced experi-
ment ally after veget ative growth by reducing the temperature by
at least 5°C (13), and this might be an important environmental
cue that has been delayed because of global warming.
Geographic location was also found to be highly import ant for
f ruiting date. Fruiting bodies typically appear considerably ear-
lier (in the range of 10–20 days) in northern, continental, and
alpine regions of Norway compared with more southern and
ocean ic regions (Fig. 2). This latitudinal pattern follows general
trends well known f rom plant phenology (14). However, we did
not find significant dependencies of temporal changes in fr uiting
date on location (SI Table 2).
We found no relation between trends in fruiting time and
fungal feeding mode, i.e., whether fungi live in symbiosis with
plants (ectomycorrhizal mode) or feed on dead organ ic matter
(saprotrophic mode) (SI Table 2). Gange et al. (10) found
delayed fruiting only for myc orrhizal fungi living in association
with deciduous trees but not for fungi associated with conifers
(which lack a concentrated period of leaf shedding). Unfortu-
nately, our herbarium data do not allow a similar comparison.
Temperature and moisture are exogenous key factors known
to influence the production of autumnal fruit bodies (15). We
analyzed the effects on f ruiting time of monthly regional anom-
alies in temperature and precipitation from June the preceding
year to November the immediate year. Our results suggest that
high temperature during November the preceding year, high
precipit ation during July the preceding year, and high temper-
atures during February, August, and October (the current year)
are associated with delayed fruiting (Fig. 3). Furthermore, high
temperatures during May and June, high precipitation during
June and October, and intermediate amounts of precipitation
during November (all factors referring to the current year) are
associated with earlier fruiting (Fig. 3). Worth noticing is that the
overall year effect (Fig. 1 A) was no longer statistically significant
when added to the best model with climatic predictors (SI Table
2). Hence, the documented overall year effect is explainable as
a direct effect of significant increases of winter and autumn
temperatures in Norway during the last decades (SI Tables 3
It might be claimed that our results have been derived f rom
biases in the data. However, we find it most unlikely that
sampling bias, for which herbarium data may be criticized on
theoretical grounds, can acc ount for the observed delay in
f ruiting. For our data to be biased with respect to fruiting time,
a temporal shift in collecting ef fort toward later in the autumn
1940 1960 1980 2000
Normalized fruiting date
1940 1960 1980 2000
Normalized fruiting date
220 240 260 280
Initial mean fruiting day
1940 1960 1980 2000
Fig. 1. Temporal variation in fruiting during the period 1940–2006. (A)
Diagram showing temporal changes in seasonal fruiting time during the
period 1940–2006. Lines indicate ﬁtted year effect with 95% bootstrap con-
ﬁdence limits from a GAM in which species and location effects were ac-
counted for. The trend was modeled as a smooth effect of time and was
signiﬁcantly positive (P ⬍ 0.001, bootstrap test). Points indicate partial resid-
uals averaged for each species and year combination. (B) Changes in fruiting
during the period 1940 –2006, partitioned on early (red; 28 species), middle,
(black; 27 species) and late (blue; 28 species) fruiters. Lines indicate ﬁtted year
effects with 95% bootstrap conﬁdence limits (for the median of the initial
mean fruiting dates within each group) from a GAM in which species and
location effects were accounted for. The trend was modeled as a tensor-
product smooth function of time and initial fruiting day. This model provided
signiﬁcantly better ﬁt to the data than the model with a common trend for all
species (P ⬍ 0.001, bootstrap test). Points indicate partial residuals averaged
for each species and year combination. (C) Displacement in fruiting time
during the period 1940–2006 for the 83 fungal taxa related to initial (1940–
1959) mean day of fruiting of each species. Displacement in fruiting (per 60
years) was calculated for each species separately by using GAMs with linear
time effects and the geographic effects accounted for by smooth functions of
longitude and latitude (thin-plate regression spline with maximally 11° of
freedom). Filled points indicate statistically signiﬁcant effects (P ⬍ 0.05,
bootstrap tests for each species). Lines indicate linear regression line ⫾ 1.96
standard error (across species). (D) Residuals from the model shown in B. Lines
indicate quantiles (5%, 10%, 25%, 50%, 75%, 90%, and 95%) as estimated by
quantile regression (23, 24). The standard deviations of the residuals within
years were negatively correlated with year (Pearson’s correlation coefﬁcient,
r ⫽⫺0.46, 95% bootstrap c.i. ⫽⫺0.64, ⫺0.26), as were the standard deviations
of the raw observations (r ⫽⫺0.47, c.i. ⫽⫺0.63, ⫺0.28).
5 1015202 0
Fig. 2. Spatial patterns in mean day of fruiting of 83 fungal species in
Norway. The isoclines (lines of different colors) represent iso-lines with 95%
bootstrap conﬁdence limits. The effect of geographic location was estimated
as smooth functions of longitude and latitude (thin-plate regression spline
with maximally 11° of freedom) by using GAM also accounting for effects of
species and temporal trends (Fig. 1B).
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0709037105 Kauserud et al.
would have to have taken place during the period 1940–2006,
most dramatically during the last 20 years when the most
dramatic changes in f ruiting time have taken place. It seems
dif ficult to imagine any rational reason for such a change of
behavior among fungus collectors. Furthermore, we are unable
to see how sampling bias possibly c ould lead to significant
associations between monthly climate variables and fruiting
time, especially with time lags. Consequently, we think that the
delay in fruiting is most parsimoniously explained through the
documented effects of climate.
Through our analysis we have demonstrated changes in the
temporal pattern of fungal fruiting in Norway during the period
1940–2006 that most likely are responses to climate change.
Worth noticing is that the accelerated delay of fruiting in the last
20 years has coincided with dramatic global warming (16). We
predict that the projected rise of global temperatures by up to
4°C by 2100 (16) will have drastic ef fects on fungal fruiting
phenology. Because fruit bodies function as habitat and diet for
many organ isms, these changes may have profound side ef fects.
Materials and Methods
Data. Data for agaricoid species (mushrooms) with at least 250 entries in the
Norwegian Mycology Database (Natural History Museum, University of Oslo)
and with mainly autumnal fruit bodies were included in the study (n ⫽ 83
species) (SI Table 1). Only herbarium records with a proper dating (day) and
geographic localization (municipality) were used for analyses. The approxi-
mate geographic positions of the records were obtained by allocating geo-
graphic coordinates for the municipalities to the records. Categorical infor-
mation for the species feeding mode (saprotrophic or ectomycorrhiza) was
also included. The number of records analyzed for each species ranged from
226 to 945; the total number of records was 34,528. Climate data for monthly
temperature and precipitation anomalies in Norway’s ﬁve main regions, for
the period 1939–2006, were obtained from the Norwegian Meteorological
Statistical Analyses. We used the GAM implementation of the ‘‘mgcv’’ library of
R (11). To compare competing models we computed genuine cross-validation
(CV) errors, models with lower CV having higher out-of-sample predictive power.
Because of within-year correlations in the response variable, CV was calculated by
leaving data for 1 year out at a time. Outlier observations (n ⫽ 60), identiﬁed by
using Grubb’s test (17) on the residuals from a GAM accounting for species and
location effects, were removed before ﬁnal analyses. All outlier observations
were records made before day 155 (June 4). A total of 34,468 records (after outlier
exclusion) for all of the 83 species were analyzed in one model with collection
time (reﬂecting the time of fruiting) as response. Thus, we did not use ﬁrst/last
observations as, e.g., Gange et al. (10) did, but instead included all records
throughout the entire fruiting season, revealing trends both in mean fruiting
date and in the variability around these trends. Differences between species in
mean fruiting time were accounted for in all models, as were location effects.
Location effects (as shown in Fig. 2) were assumed to be similar across species and
were modeled by a thin-plate regression spline of longitude and latitude (max-
imally 12 knots, i.e., 11° of freedom). Temporal trends were modeled as either
linear or smooth effects (natural cubic splines with maximally 4 knots) of year
(1940–2006). Possible differences in temporal trends between (i) different re-
gions (central, east, north, south, and west Norway), (ii) species groups deﬁned by
feeding mode (saprotrophic and ectomycorrhiza), or (iii) species groups charac-
terized by the initial (1940 –1959) mean day of fruiting (the 28 earliest-fruiting
species, the 27 intermediate-fruiting species, and the 28 latest-fruiting species)
were accounted for in models with group-speciﬁc year terms (i.e., with 5, 3, or 2
smooth year terms instead of 1). A continuous interaction between initial fruiting
day and year was modeled by a tensor-product smooth function constructed
from linear combinations of terms that were cubic regression spline basis func-
tions of the two variables (each with maximally 4 knots) (11). CV showed that only
the interaction terms including initial fruiting day and year improved the model
(SI Table 2). Accordingly, results from the continuous interaction model, which
had the lowest CV prediction error, are shown in Fig. 1B. Data points are colored
by intervals of initial fruiting date as detailed in the ﬁgure legend.
A separate model was ﬁtted to identify climate variables accounting for
interannual variation in fruiting. Linear effects of 36 different climate vari-
ables were considered (monthly regional anomalies in temperature and pre-
cipitation from June the preceding year to November the current year).
Starting with the full model, terms were removed until the model with the
lowest CV prediction error was found. CV prediction error was reduced further
by substituting two of the selected linear climatic effects with smooth terms
and subsequently removing two more linear terms. We also explored the
possible direct or lagged effects of the North Atlantic Oscillation (NAO) using
the PC-based NAO index described in refs. 18 and 19 (www.cgd.ucar.edu/cas/
jhurrell/indices.data.html#naopcdjfm), but we found this to have lower ex-
planatory power than the precipitation and temperature indices (results not
shown). Finally we tested for linear interaction effects between the selected
climatic variables and feeding mode or initial fruiting time and for the
combined effects of climatic variables and year. We found no signiﬁcant
interactions between the climatic variables and feeding mode. We found
evidence for a negative interaction between initial fruiting day and temper-
ature in February and October (SI Table 2), suggesting a stronger response of
early fruiters to these climate variables. However, these interaction effects did
not fully explain the stronger year effect of early fruiters, because adding an
interaction term between year and initial fruiting day improved the predictive
power of the model (SI Table 2). In contrast, the model with only climatic
variables was not improved by adding a smooth year effect (SI Table 2),
suggesting that the overall temporal trends are indeed explainable by the
measured climatic variables.
The statistical signiﬁcance of terms and conﬁdence intervals were com-
puted by using a modiﬁed wild bootstrap approach (20, 21), as described in
ref. 22, which accounted for both heteroscedasticity and within-year correla-
tion of residuals. When calculating the signiﬁcance of terms, bootstrap data
sets were constructed from residuals and ﬁtted values from models without
the given terms, and the increase in variance explained (R
) by including the
given terms was used as test criterion. When calculating the signiﬁcance of the
interaction between initial fruiting day and year, initial fruiting day was
recalculated for each bootstrap sample, thus accounting for bias resulting
from one of the predictor variables being derived from the response variable.
A quantile regression analysis, using the Frisch–Newton interior point
−4 −2 0 2 4
Temp. Nov. t − 1
−8 −4 0 4 8
Temp. Feb. t
−2 0 2
Temp. May t
−2 0 2 4
Temp. June t
−2 0 2 4
Temp. Aug. t
−6 −2 0 2 4
Temp. Oct. t
50 150 250
Precip. July t − 1
(% of normal)
50 100 200
Precip. June t
(% of normal)
50 100 200
Precip. Oct. t
(% of normal)
50 150 250
Precip. Nov. t
(% of normal)
Fig. 3. Climatic effects on interannual changes in fruiting of 83 fungal species. Temperature and precipitation variables are referring to either the preceding
year (t ⫺ 1) or the same year (t) as fungal fruiting. The climatic effects shown were estimated as linear or smooth terms in one GAM also accounting for location
and species effects (SI Table 2). Whole and broken lines indicate ﬁtted partial effects with 95% bootstrap conﬁdence limits. The tick marks on the x axis show
the location of the covariates (see also SI Fig. 4, showing partial residuals).
Kauserud et al. PNAS
March 11, 2008
method and cubic splines with 8 knots, was performed with the residuals from
model shown in Fig. 1B as response (23, 24). In a separate analysis (SI Text),
residual variance structure was analyzed by restricted maximum-likelihood
methods (25) using the nlme library of R (26), selecting the most parsimonious
model based on Akaike’s information criterion.
ACKNOWLEDGMENTS. We acknowledge all who have contributed with spec-
imens to the Norwegian mycological herbaria, Einar Timdal for making the
data available through the web interface, the Climate Division at the Norwe-
gian Meteorological Institute for providing climate data, and two anonymous
reviewers for valuable comments on earlier versions of the manuscript.
1. Menzel A, Sparks TH, Estrella N, Koch E, Aasa A, Ahas R, Alm-Ku¨ bler K, Bissolli P,
Braslavska´ O, Briede A, et al. (2006) Glob Change Biol 12:1969 –1976.
2. Stenseth NC, Mysterud A, Ottersen G, Hurrell JW, Chan KS, Lima M (2002) Science
3. Walther GR, Post E, Convey P, Menzel A, Parmesan C, Beebee TJC, Fromentin JM,
Hoegh-Guldberg O, Bairlein F (2002) Nature 416:389 –395.
4. Menzel A, Fabian P (1999) Nature 397:659.
5. Bradley NL, Leopold AC, Ross J, Wellington H (1999) Proc Natl Acad Sci USA 96:9701–9704.
6. Fitter AH, Fitter RSR (2002) Science 296:1689 –1691.
7. Lavoie C, Lachance D (2006) Am J Bot 93:512–516.
8. Primack D, Imbres C, Primack RB, Miller-Rushing AJ, Del Tredici P (2004) Am J Bot
9. Jonze´ n N, Linde´ n A, Ergon T, Knudsen E, Vik JO, Rubolini D, Piacentini D, Brinch C, Spina
F, Karlsson L, et al. (2006) Science 312:1959 –1961.
10. Gange AC, Gange EG, Sparks TH, Boddy L (2007) Science 316:71.
11. Wood SN (2006) Generalized Additive Models: An Introduction with R (Chapman and
Hall/CRC, Boca Raton, FL).
12. Post E, Stenseth NC (1999) Ecology 80:1322–1339.
13. Ku¨ es U, Liu Y (2000) Appl Microbiol Biotechnol 54:141–152.
14. Ovaska JA, Nilsen J, Wielgolaski FE, Kauhanen H, Partanen R, Neuvonen S, Kapari L,
Skre O, Laine K (2005) Ecol Stud 180:99 –115.
15. Eveling DW, Wilson RN, Gillespie ES, Bataille A (1990) Mycol Res 94:998 –1002.
16. Intergovernmental Panel on Climate Change (2007) Climate Change 2007: Impacts,
Adaptation and Vulnerability. Contribution of Working Group II to the Fourth
Assessment Report of the Intergovernmental Panel on Climate Change, eds Parry ML,
Canziani JP, Palutikof JP, van der Linden PL, Hanson CE (Cambridge Univ Press,
17. Grubbs F (1969) Technometrics 11:1–21.
18. Hurrell JW (1995) Science 269:676 – 679.
19. Stenseth NC, Ottersen G, Hurrell JW, Mysterud A, Lima M, Chan KS, Yoccoz NG,
Adlandsvik B (2003) Proc R Soc London Ser B 270:2087–2096.
20. Liu RY (1988) Ann Stat 16:1696–1708.
21. Mammen E (1993) Bootstrap and wild bootstrap for high dimensional linear models in
resampling. Ann Stat 21:255–285.
22. Stige LC, Ottersen G, Brander K, Chan KS, Stenseth NC (2006) Mar Ecol Prog Ser
23. Koenker R (2006) QUANTREG: Quantile Regression. R package (Vienna University,
Vienna), Version 4.01.
24. Koenker R, Bassett G (1978) Econometrica 46:33–50.
25. Pinheiro JC, Bates DM (2002) Mixed-Effects Models in S and S-PLUS (Springer, New
26. Pinheiro JC, Bates D, DebRoy S, Sarker D (2006) nlme: Linear and nonlinear mixed
effects models. R package (Vienna University, Vienna), Version 3.1-73.
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0709037105 Kauserud et al.