Europ. J.

Agronomy 40 (2012) 102–111

Contents lists available at SciVerse ScienceDirect

European Journal of Agronomy

journal homepage: www.elsevier.com/locate/eja

Why farmers’ sowing dates hardly change when temperature rises

P.A.J. van Oort a,∗ , B.G.H. Timmermans b , A.C.P.M. van Swaaij c
a
Wageningen University, Chair Crop and Weed Ecology, PO Box 430, 6700AA Wageningen, The Netherlands
b
Wageningen University, Chair Crop and Weed Ecology, 6700AA Wageningen, The Netherlands
c
Dutch Institute of Sugar Beet Research (IRS), PO Box 32, 4600 AA Bergen op Zoom, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: Previous studies have shown that temperature rise leads to an earlier onset of spring in wild plant species
Received 18 November 2011 and that farmers are not keeping track of climate change. Crop growth models and experiments show yield
Received in revised form 6 February 2012 gains to be obtained from earlier sowing. Why do farmers not sow earlier? We propose simple models
Accepted 15 February 2012
on the relation between several weather variables and farmers’ sowing dates and test these models with
392 observed weekly sugar beet sowing percentages in the Netherlands. Data showed no correlation
Keywords:
between spring temperature and sowing dates. However when we corrected for number of frost days in
Sowing date
spring, we clearly saw earlier sowing at higher temperatures. Frost causes crumbling of the soil resulting
Planting date
Adaptation
in a nice seedbed; farmers sow later when there have been less frost days. Temperature rise “makes”
Temperature farmers sow earlier, but the number or frost days simultaneously decreases and this “makes” farmers
Climate change sow later. The combined effect is that temperature rise leads to only a small net advance in sowing date,
Global warming 2 days per 1 ◦ C according to our model. Many climate change studies have assumed advances in sowing
Sugar beet date are often larger than likely based on historical trends and models developed in this paper. Such
Beta vulgaris L studies may well be overly optimistic in projected yield gains that follow from earlier sowing. Our results
also revealed that sowing dates were correlated with the sum of solar radiation over the preceding 7–10
months. This correlation remained significant also when other weather variables (rain, temperature, frost
days) were accounted for. Our work raises the question what is the process causing this correlation and
does such correlation with radiation sum also show up in other time series of phenological events or
human behaviour.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction Ewert found that sowing dates of barley advanced by 0.2 days per
decade over the last 50 years. Kucharik found that in the Central
There is more and more evidence of global warming on the “bio- USA corn planting did advance at 4.8 days per decade which is more
logical” start of spring (Menzel et al., 2006a; Schwartz et al., 2006; than in the other studies cited above, but Kucharik also found that
Cleland et al., 2007). In a hotter climate, plant species emerge ear- this advancement of sowing period could not be explained from
lier and develop faster through their phenological stages. This trend temperature increases, but rather from management and techno-
is only partly reflected in agricultural crops. Menzel et al. (2006b) logical changes. In agricultural research it is widely known that
show that farmers are sowing earlier (2.1 day per decade) but that earlier sowing enables the crop to capture more of spring radiation
crops are responding even faster to climate change (growth stages and so produce higher yields. In sugar beet this has been shown
advancing 4.4–7.1 d decade−1 ). There has been much more research experimentally (Scott et al., 1973; Lauer, 1997; Öztürk et al., 2008;
on plant response to climate change than on human response to Hoffmann and Kluge-Severin, 2010, 2011) and with crop growth
climate change. More recent studies of farmers’ sugar beet sowing models (Smit, 1993; Jones et al., 2003; Richter et al., 2006; Jaggard
dates over 30 or more years confirm the trend noted by Menzel et al., 2007). It is known for a long time already. For example, in the
et al.: sugar beet development is speeding up but there is no or Netherlands Wind (1960) reviewed literature, referring to sowing
only a weak trend towards earlier sowing (Jaggard et al., 2007; date experiments in the Netherlands and Germany in 1930s and in
Kaukoranta and Hakala, 2008). Multi-decadal studies in other crops 1950s in US, all showing that earlier sowing gives higher yields for
suggest that crop phenology is advancing faster than farmers’ sow- a range of crops.
ing dates (Kucharik, 2006; Siebert and Ewert, 2012). Siebert and The above observations are difficult to reconcile. Agricultural
science tells us that yield gains can be obtained with earlier sow-
ing. Nature is responding to temperature rise and there is no reason
∗ Corresponding author. Tel.: +31 317481357. why agricultural crops should be an exception. Then why are farm-
E-mail address: [email protected] (P.A.J. van Oort). ers not, or only weakly, responding to temperature rise? Why are

1161-0301/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.eja.2012.02.005
 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111 103

not farmers sowing earlier? Our objective is to investigate why of the values of startr,y,doy . In the third step our aim was to find
farmers do not show a response to climate change and sow ear- out how much of the variation in sowing dates can be explained
lier in spring. Could it be that farmers’ sowing dates are not at all by weather variables, without prior knowledge on start dates. We
restricted by low temperatures, because another factor is critically first simulated startr,y,doy and given those simulated values we then
preventing successful early sowing? Or is there a temperature effect determined suitabler,y,doy and percentsownr and used both to cal-
after all that is hidden from our sight by the presence of other more culated SIMSOWr,y,doy . Parameters for startr,y,doy were estimated
important factors? To answer such questions we develop simple through linear regression, the parameters for suitabler,y,doy and
models for predicting sowing dates, feed these models with differ- percentsownr were estimated by maximising the goodness of fit
ent weather variables and compare their accuracies in predicting of the overall model, which was calculated as the sum of squared
farmers’ sowing dates. errors (SSE):

2. Materials and methods SSE = (SIMSOWr,y,w − OBSSOWr,y,w )2 (2)
r,y,w

2.1. Data where SIMSOWr,y,w is the simulated percentage in the week

preceding day w, obtained by adding up the daily percentages
The Dutch Institute of Sugar Beet Research IRS reports for 12 SIMSOWr,y,doy (Eq. (1)) over the preceding 7 days and OBSSOWr,y,w
regions in the Netherlands the weekly percentage of the total area is the observed weekly percentage sown.
of the crop that is sown, for the years 1996 to now (15 years). Of course farmers’ sowing dates depend not only on weather
The larger regions have different soil types and weather may vary variables but also on other factors such as availability of staff,
within the regions. Therefore we limited our analysis to six small machinery, seeds, logistics of other farm operations, experience
regions with predominantly one soil type per region and where it and awareness of the benefits of early sowing. A perfect predic-
is valid to assume that one weather station in the centre is rep- tion of farmers sowing dates is therefore impossible from weather
resentative for the weather in whole the region. Four of these variables alone. But, as we show below, there are good reasons to
regions have clayey soils: Oost-en Zuid Flevoland (Flevo), Noordeli- expect that the decision on when to sow depends in part on the
jke klei (NK), Noordoostpolder (NOP) and Zeeuws Vlaanderen (ZV). weather:
The other two regions have sandy soils: Noordelijk Zand (NZ) and
Oost-Brabant (OB). For each region we selected a weather station Hypothesised factors affecting the start date
in the centre of the region. Sowing is concentrated in a period of H1. Temperature in the preceding weeks or months: it must be
2–5 weeks, within which occasionally weeks with no sowing occur sufficiently warm for healthy growth of the young plants and to
and of which the start date is in some years 4 weeks later than avoid bolting risk (Milford et al., 2010).
in other years. If we select for each year and region the start of
the sowing period and the end of the sowing period and count the H2. Frost days in the preceding weeks or months. Forst causes
weeks in between then we have 392 observed weeks. We used this crumbling of the soil which results in a good seedbed; this effect is
number of 392 as the number of observations (n) in the statistical stronger on clay than on sand (Benoit, 1973; Bullock et al., 1988;
tests in Sections 3.2 and 3.3. For each of the regions, the IRS dataset Arvidsson et al., 2000; Oztas and Fayetorbay, 2003).
contains the percentage sown of the total sugar beet area for each H3. Trafficability. We use three proxies for trafficability of fields:
week of the year, recorded on each Monday. We define OBSSOWr,y,w rainfall, rainfall minus potential evapotranspiration (rain − ET0 )
as the observed cumulative percentage sown in region r, year y, and groundwater level (Jaggard, 1977; van der Neut et al., 1995;
recoded on day w. We use a w to indicate that this percentage is Earl, 1997; Raper, 2005).
the sum over the preceding week, recoded on day of the year w.
Thus OBSSOWNOP,1996,71 is the percentage of sugar beet that was H4. Radiation in the preceding months, through some process
sown in the Noordoostpolder region, from day 65–71 in the year yet not understood.
1996. Recording every Monday implies that every year, the weekly Hypotheses regarding sowing beyond start date
summary was made on a different date, for example day 10 March
in 1997 (w = 69), 9 March 1998 (w = 68), 8 March 1999 (w = 67), etc. H5. Weather in the preceding months (as defined in H1–H4) is
important in determining the annual distribution of weekly sowing
2.2. Methods percentages.
H6. Farmers prefer sowing at a certain target sowing date which
We simulate the daily percentage sown for day of the year doy, is very early. The annual distribution of weekly sowing percentages
for each region r (1–6) and each year y (1996–2010), SIMSOWr,y,doy , is caused only by unfavourable weather in the days at and after the
using Eq. (1): target sowing date (as defined in H7 and H8).
SIMSOWr,y,doy = startr,y,doy ∗ suitabler,y,doy ∗ percentsownr (1) H7. Workability. What makes a day suitable depends only on
where startr,y,doy and suitabler,y,doy are binaries indicating whether rainfall in the preceding days. A too wet top soil, in interaction with
a day is suitable or not and percentsownr is a constant percent- the machinery and soil type, leads to irregular and thus sub-optimal
age for how much of the area is sown in case startr,y,doy = 1 and sowing (e.g. Rounsevell, 1993; Earl, 1997).
suitabler,y,doy = 1. The binary startr,y,doy marks the start of the sowing H8. What makes a day suitable depends on minimum tempera-
period of each region and each year. Startr,y,doy is set to 0 before the ture in the preceding or upcoming days; too cold weather leads to
first suitable day of the year and to 1 for this day and beyond. Beyond poor emergence and bolting risk (Milford et al., 2010).
the start date suitabler,y,doy determines whether a day is suitable for
sowing or not. We first determined the weather variables explain- Hypotheses regarding regional variation
ing the variation in startr,y,doy . In a second step we set startr,y,doy to H9. The same parameter sets for calculating startr,y,doy apply in
the observed date for each year and region and then determined each region or soil.
the weather variables explaining the variation in suitabler,y,doy . This
second step informs us on which weather variables determine H10. The same parameter sets for calculating suitabler,y,doy apply
suitabler,y,doy without being troubled by uncertainty in estimates in each region or soil.
104 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111

H11. The same parameter sets for calculating percentsownr apply

in each region or soil.
All hypotheses were tested with the F-test.
We do not think farmers observe weather variables as they are
listed above and based on that decide when to sow. Rather, we con-
sider the weather variables as scientific indicators that are likely to
be linked to farmers’ perception of their environment. The terms
trafficability and workability are briefly discussed here. Earl (1997),
following Campbell and O’Sullivan (1991), defines trafficability as
the ability of soil to (1) provide adequate traction for vehicles and
(2) withstand traffic without excess compaction or structural dam-
age. Workability is a combination of trafficability and the ability
of soil to be manipulated in a desired way without causing sig-
nificant damage or compaction. Both depend on strongly on soil
moisture status. Trafficability depends on soil moisture in the com-
plete soil profile up to one or two metres deep so we expect this
depends on accumulated rainfall over a longer period of time. The
additional workability criterion for sowing is that the top soil must
be sufficiently dry, otherwise seeds and clumps of soil stick to the Fig. 1. Averaged daily percentage sown in the null model.
machinery and lead to irregular sowing densities, both too high and
too low. We expect that up to some days after a rainfall event, the
Neut et al., 1995; Raper, 2005) and therefore postponed sowing.
topsoil may be temporarily too difficult to work.
No good data are available on drainage, therefore we use three
The definition of the actual (observed) start date is somewhat
proxies for trafficability: rainfall, rainfall minus ET0 (the rainfall
arbitrary. It can be set to the date at which the first 1% was sown,
surplus) and groundwater level (which is the result of rainfall, ET
or the first 5% or the first 10% or even higher. We tried different
and drainage). If trafficability is indeed the main factor determining
thresholds and found that the lowest SSE (Eq. (2)) was obtained
interannual variation in start dates, then we would expect to find
by setting the threshold to 10%. Here, we limit the presentation of
high R2 values for these three proxies (hypothesis 3). Groundwater
results to the 10% case. The day at which 10% was sown was linearly
levels were extracted from between 19 and 98 wells located in agri-
interpolated from weekly observed sowing percentages, hence the
cultural production areas within each region. From these wells we
interpolation error for obtaining this day is at most 3.5 days, which
calculated the daily average level (averaged across the wells). Then
is small in comparison with the spread in start dates (∼30 days).
we searched for correlation between observed earliest (10%) sow-
Following hypothesis 6 we identify for each region a target sowing
ing date and groundwater for any given date or averaged over any
date, which we set to the earliest out of 15 observed start dates of
previous period defined by s1 and e1 . We also checked if response
each region (Eq. (3)):
to weather variables was different on sand and clay.
tardoye10r,y = miny=1996,2005 (obsdoye10r,y ) (3) For the simulations beyond the start date we assumed that the
first 10% of the total sugar beet area was sown on the day just before
simdoye10r,y = b0 + b1 ∗ W 1r (s1 , e1 ) (4) the start date. In reality these 10% are spread over a period of 1–2
The binary startr,y,doy in Eq. (1) is set to 0 for any day of the weeks before the start date. We do not understand well what deter-
year earlier than tardoye10r,y or simdoye10r,y and otherwise to 1. mines when these first 10% are sown. Assigning them to the day just
Hypotheses 1–5 lead to Eq. (4) where simdoye10r,y is the simulated before simdoye10r,y or tardoye10r,y is the simplest possible solution.
day on which the first 10% of the area is sown, b0 and b1 are regres- As it concerns only 10% of the sowing dates, this simplification has
sion coefficients, W1r is a weather variable from a station in region little effect on the calculated SSE value (Eq. (2)). For the remaining
r, s1 is the start of the period over which W1r is calculated and e1 90% we simultaneously calibrated suitabler,y,doy and percentsownr .
is the end day of the period over which W1r is calculated. We var- A day is considered suitable (suitabler,y,doy = 1) if the rainfall sum
ied s1 and e1 on a time step size of 10 days, with a wide range: over the preceding drydays is less than maxmm, if the average of
s1 from −240 to 80 and e1 from 80 to 120. We looped through minimum temperatures over the preceding or upcoming tempdays
all possible sets of s1 and e1 , for each set s1 ,e1 we calculated b0 is higher than minoC, or a combination of these two criteria. We
and b1 through linear regression and retained the parameter set tested using minimum (TN) and maximum temperature (TX), TX
s1 ,e1 ,b0 ,b1 with the highest R2 value. We also tested multivariate gave the most accurate results.
models and used the F-test to find out whether multivariate models Let SSEmodel be the SSE value calculated with Eq. (2) from pre-
were significantly (˛ = 0.05) more accurate than models with just dicted sowing dates generated with a model as parameterised
one weather variable. As weather variables we used daily average based on the above hypotheses. We calculated R2 values for each
temperature, number of frost days, rainfall, radiation and poten- model as R2 = 1 − SSEmodel /SSEnull , which meant we needed to
tial evapotranspiration (ET0 ) as explanatory variables. A frost day establish a null model. We defined the null model as the model
was defined as a day having minimum (air) temperature below in which the daily percentage sown is the average daily percentage
0 ◦ C. Theoretically it would be better to use soil temperature rather sown, averaged across regions and 15 years (Fig. 1).
than air temperature at 2 m above soil, but as soil temperature
is not routinely collected and also very strongly correlated with 2.3. Validation
minimum air temperature we preferred to use the latter. For the
potential evapotranspiration (ET0 ) we used values as provided in For validation we used independent data from the same regions.
the weather data files, which the Royal Netherlands Meteorological We used observed daily sowing dates in the period of 1985–2005
Institute (KNMI) calculates using the Makkink equations (Makkink, from a sample of around 90 farmers per region, individual farmers
1957, 1959) based on radiation and temperature. Excess rainfall in who filled in surveys for a sugar producing company (Suiker Unie,
spring, in combination with insufficient capacity for draining the Sjaak Kolff, pers. comm.). We consider two types of errors: (1) no
fields could lead to limited trafficability (Jaggard, 1977; van der sowing occurred on a day suitable according to our model (false
 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111 105

Fig. 2. Average monthly weather data. Solid lines: minimum and maximum temper-
ature. Dashed line: radiation. Dashed dotted line (left y-axis): potential evaporation
(ET0 , mm). Dotted line (left y-axis): rainfall (mm). Calculated from daily weather Fig. 3. Relation between observed earliest (10%) sowing day, average temperature
data 1996–2010 from the station De Bilt in the centre of the Netherlands. over the period from day of year 50 to 110 (cf. model 12, Table 1) and number of frost
days (from day 50 to 100). To get the temperature sum as used in Table 1, multiply
average temperature (x-axis) with 110–50 (=60). Looking at all the data (x), there is
positive) and (2) sowing occurred on a day unsuitable according to
only a very weak correlation between temperature sum and start day. But for any
our model (false negative). Type 1 errors are difficult to quantify. given number of frost days ( and ), it is clear that sowing is earlier if temperature
There can be different reasons for why no sowing occurred on a spe- is higher. Temperature and frost days are correlated: when temperature is higher
cific day, which not necessarily means that if no sowing occurred (moving to the right) there are less frost days ().
that our model was wrong. It is possible that by pure chance, none
of the farmers in a validation dataset sowed. It is possible that most
farmers had not yet started sowing or that most farmers had already predicted with temperature (hypothesis 1, R2 = 0.13, Table 1, model
completed sowing. As a result of this, false positives are likely to be 5) or from (night) frost days (H2, R2 = 0.11, Table 1, model 6).
overestimated and overall accuracy is likely to be underestimated. Temperature sum and night frost are negatively correlated. A
To overcome the problems noted above in quantifying false posi- higher temperature sum in March means earlier sowing due to
tives we validated using only data during each region’s main sowing a higher temperature sum. But the number of frost days is less
period. Furthermore in the validation data set we considered day when temperature sum is high and sowing is later with less frost
j suitable if on day j sowing occurred, but also if sowing did occur days. When these opposed effects are included in one model
on both day j − 1 and day j + 1, the day just before and the day just (Table 1, model 12) a large part of the variation in sowing dates
after. In total, we used 2520 sowing dates for validation, of which can be explained (R2 = 0.49). The low correlation with tempera-
1331 dates were suitable according to the validation data and above ture only (R2 = 0.13) is the consequence of underlying correlation
definitions. between temperature and frost days and the fact that temper-
ature sum and frost days work in opposite directions. This is
3. Results illustrated in Fig. 3 where it is clear that there is no relation
between temperature and start day. But if we look closer at a
First an impression is given of the weather in the study area. given number of frost days, we do find that sowing is earlier
Fig. 2 shows monthly averages for a number of weather variables, when temperature from day 50 to 110 is higher. The effect of
for a station in the centre of the Netherlands. Temperatures are night frost in regions with predominantly clayey soils is stronger
higher in the south. The temperature difference between the south- than in regions with predominantly sandy soils (see models
ernmost region ZV and the northernmost region NK is in the order 12–14).
of +2 ◦ C during daytime (warmer in the south) and +0.3 ◦ C during On first sight it seems unlikely that the start of sowing period
night time in the period of January–February–March. The average depends on limited trafficability (caused by excessive winter or
day on which 50% is sown in the Netherlands was day of year 98 (8 spring rainfall, hypothesis 3), as the R2 value of the rain–ET0 model
April). The southernmost region ZV is consistently earlier (50% day (R2 = 0.32) and the rain model (R2 = 0.31) are relatively low. Also
is 91) than the northernmost region NK (50% day is 104). Within correlation between the start of the sowing period and ground-
each region there are large interannual differences in the 50% date, water table was low (R2 < 0.1, not shown in table) even when
up to 30–40 days. In the Flevoland region 1999 was 34 days later calibrated at a regional level. When the main effect of radiation
than 2003 (50% day 117 vs 83). These differences are mostly due (R2 = 0.52) is accounted for, a significant increase in accuracy can
to variation in start of the sowing period and to a lesser extent be gained by including rainfall (R2 = 0.59), indicating that traffica-
due to unsuitable days beyond the start of the sowing period. For bility does play a role, though minor in comparison with radiation.
Flevoland, the difference in the 10% date closely matches the dif- Likewise, adding rainfall to the temperature + frost days model sig-
ference in the 50% date (10% day 110 vs 78, 32 days later in 1999). nificantly increases accuracy (model 12 → 13, R2 = 0.49 → 0.60). We
Figs. 3 and 4 show differences in the start of the sowing period of also checked if the sensitivity to rainfall differed between the sandy
up to 30 days. soils and clayey soils (model 4 vs 7 and 8 vs 10); this proved not to be
the case. The low impact of rainfall on the start of the sowing period
3.1. Start of the sowing period may be specific for the Netherlands. The impact of winter rainfall
may be higher in environments with a higher winter rainfall sur-
The start of the sowing period, i.e. the day on which the first plus. The impact of winter rainfall may be higher in environments
10% was sown in each year and region, could not be accurately with less active water management; the Dutch are renowned for
106 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111

Significantly (p < 0.01) more accurate than model 1

Significantly (p < 0.001) more accurate than model

Significantly (p < 0.05) more accurate than models

12; calibrated with same start and end dates as in

significantly increase accuracy (R2 = 0.16, p = 0.08)

model 8; t-test indicated intercept often does not

8; calibrated with same period for rain, temp and
Significantly (p < 0.01) more accurate than model

Not significantly more accurate than models 8 or

Not significantly (p = 0.71) more accurate than

Not significantly (p = 0.81) more accurate than

9; significantly (p < 0.01) more accurate than

The effect is stronger on clay soil but did not

frost days (s2 = s3 = s4 ; e2 = e3 = e4 )

differ between regions

model 12

model 13
model 8

5&6
Parameters
Fig. 4. Relation between radiation sum (3 September to 10 April, day −120 to 100,
Table 1, model 1) and observed start of the sowing period, for the six study regions.

12

15
4
4
4
4
4
4

6
7
9

9
their water management, which includes securing that soils are not

0.52
0.51
0.32
0.31
0.13
0.11

0.31
0.59
0.65

0.59

0.59

0.49

0.66
0.60
too wet and not too dry during the period of seedbed preparation

R2
and sowing (e.g. see van der Neut et al., 1995).
Starting with the radiation model (model 1) we find that 52%
of the variation in start of sowing dates can be explained, with
minor improvements by adding rainfall and temperature plus

166.7 − 0.35*frostdays sand(50,100) − 0.52*frostdays clay(50,100)

140.7 − 0.75*frostdays sand(50,100) − 1.06*frostdays clay(50,100)

126.9 − 0.55*frostdays sand(50,100) − 0.80*frostdays clay(50,100)

170.2 − 4.6*bin(Flevo) − 2.5*bin(NK) − 7.0*bin(ZV) − 2.9*bin(NZ)
81.7 + 0.151*rainsumSand(70,110) + 0.144*rainsumClay(70,110)
frost days (model 1 → 8 → 9, R2 = 0.52 → 0.59 → 0.65). Starting with

155.6 + 0.126*rainsumSand(70,90) + 0.101*rainsumClay(70,90)

the temperature plus frost days model (model 12, R2 = 0.49) we

183.4 − 0.030*radsum(−200,100) + 0.084*rainsum(50,100)

156.1 + 0.107*rainsum(70,90) − 0.064*radsum(−110,100)
find minor improvements by adding rainfall and radiation (model
12 → 13 → 14, R2 = 0.49 → 0.60 → 0.66). Basically this tells us that,

− 0.307*frostdays(50,100) − 0.067*tempsum(50,100)

− 0.087*tempsum(50,110) + 0.106*rainsum(60,90)

− 0.067*tempsum(50,110) + 0.084*rainsum(60,90)
due to multi-collinearity between temperature, frost days and radi-
ation we cannot determine their relative importance. The question
then arises whether there exists a radiation effect at all. If there

− 3.1*bin(OB) − 0.064*radsum(−120,100)
were no effect of radiation then we would expect that the weather
86.9 + 0.156*(rain − ET0 )sum(60,90)

variables in model 13 together explain all or almost all of the vari-

171.5 − 0.068*radsum(−120,100)
189.7–0.372*ET0 sum(−150,110)

ation in radiation sum as calculated in model 1. We find that only

110.3–0.057*tempsum(80,120)
81.7 + 0.146*rainsum(70,110)

60% of the variation in radiation sum of model 1 is explained, which

− 0.064*radsum(−110,100)

− 0.019*radsum(−200,100)
93.5–0.75*frostdays(0,10)

is relatively small when considering possible detrimental effects

− 0.10*tempsum(50,110)
of multi-collinearity on the ability to detect effects of different
explanatory variables (O’Brien, 2007). Also, we find that adding
radiation to the temperature + frost days + soil + rainfall model sig-
nificantly increases the accuracy (model 13 → 14, R2 = 0.60 → 0.66,
p < 0.01). Thus the effect of radiation is also found when other
Model

weather variables are accounted for. But since there is collinearity

between weather variables, we cannot make certain statements on
their relative importance. Fig. 4 shows that the relation between
radiation sum and earliest 10% sowing date exists in all regions.
We show in this figure the radiation sum as defined in model 1, i.e.
Temp + frostdays + soil + rain + radiation
Models for predicting the start of the sowing period.

calculated over the period of day −120 to 100 (3 Sept to 10 April),

Radiation + rain + temp + frostdays

a period of 7 months. When other weather variables were added

Radiation + regional intercept

to the radiation models, the period over which the radiation sum
Temp + frostdays + soil + rain

is calculated shifted to day −200 to 100 (15 June to 10 April), a

Temp + frostdays + soil

period of 10 months. We also checked if we were not actually look-

Radiation + rain + soil

ing at an (yet not understood) autumn radiation effect or spring

effect. In the univariate radiation model (model 1), R2 dropped from
Radiation + rain

0.52 to below 0.40 when radiation sum was calculated over autumn
Rain − ET0

Rain + soil
Frostdays
Radiation
Variables

only or spring only. Models for simulating the start date were also
Temp

regionally calibrated, we do not present the results here, as the gen-

Rain
ET0

eral pattern was the same as shown in Table 1. With the regionally
calibrated models we could explain between 72% and 96% of the
variation in start dates and on average 82% of the variation in start
Table 1

dates. We did use the regionally calibrated models in Section 3.3

No.
1
2
3
4
5
6

7
8
9

10

11

12

13

14

on the full model.

 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111 107

Table 2
Accuracy of rules for defining what makes a day suitable for sowing, identified with observed start of sowing period.

Suitabler,y,doy Percentsownr Parameters

Scale Variable Scale SSE R2 Start Suitable Percentsown Total

Regional Rain Regional 47,565 0.81 90 12 6 108

National Rain Regional 58,638 0.76 90 2 6 98
Regional Rain National 66,854 0.73 90 12 1 103
National R + TX National 80,144 0.68 90 4 1 95
National Rain National 81,644 0.67 90 2 1 93
National TX National 211,270 0.14 90 2 1 93
Null model 246,651 0

TX = maximum temperature; using daily minimum temperatures gave worse results.

3.2. Once sowing has started average 70% of the days in April and May are suitable according to
our rainfall rules. This suggests that future changes in rainfall will
Beyond the start date and with observed start dates, 67% of not have a big impact, as ample opportunities will remain if rainfall
the variation in weekly sowing percentages can be explained from increases not too much – we follow-up on this in Section 3.5. Fig. 5
rainfall (Table 2). This percentage can be further increased to 81% shows that sowing started much earlier in 2003 than in 1999. The
by taking into account regional differences in the rule for suit- figure also shows that according to our model, the whole period
able and by taking regionally different sowing percentages. Note from day 70 to 100 was suitable for sowing in both years. The dif-
that these R2 values are calculated from all the weekly observed ferences in sowing dates between these two years can therefore
and simulated sowing percentages. Accurately reproducing these not be attributed to lack of suitable (=sufficiently dry) days in 1999.
weekly percentages also guarantees that regional averages are Our model for simulating variation in start dates can explain the
correctly simulated. The 81% model of Table 2 simulates average differences between these two years, simulating the start days as
sowing dates (one for each year and region), with an accuracy of day of year 97 and 77. These numbers support the hypothesis 5
R2 = 0.92. All differences in SSE and R2 values in Table 3 were sta- that weather in the preceding months is important in determining
tistically significant, suggesting that what makes a day suitable for the annual distribution of weekly sowing percentages. It contra-
sowing differs between regions (hypothesis 10) and that the per- dicts the target date hypothesis (H6) which states that farmers will
centage sown differs between regions (hypothesis 11). The model in principle in this region start sowing on day 77 unless restricted
with (maximum) temperature as the only explanatory variable by rainfall such as in 1994. This is just one illustrative example, in
was only slightly more accurate than the null model and adding Section 3.3 we provide results of a more systematic testing of the
temperature to the rain model did not significantly increase the target date hypothesis.
accuracy. This indicates that sowing beyond the start date is not
influenced by temperature (hypothesis 8). The number of suffi- 3.3. The full model
ciently dry days required was less on the sandy soils (4–5 days
no more than 9–11 mm rain) than on the clayey soils (7–8 days, For the full model we estimated both the start date, whether a
13–18 mm), which is in line with faster drainage on the sandy soils. day was suitable or not and the percentage sown on a suitable day.
The percentage sown on any suitable day was lower on the sandy For the start date we used (1) start dates estimated with region-
soils (5%) than on the clayey soils (7–10%), we do not know the ally calibrated models, (2) start dates estimated with model 9 from
cause of this difference. Possibly sowing is less of an issue if you Table 1 and (3) the target sowing date for each region (which
have a lighter soil that is workable on more days. we set to the earliest observed over 15 years, Eq. (3)). We show
The rain rule is illustrated for the province of Flevoland (Fig. 5) only the most contrasting models, as Section 3.2 already showed
for a year with a relatively wet spring (1994) and two years with that both the parameters percentsownr and the parameters defin-
“normal” amounts of rain (1999 and 2003). The start dates for these ing suitabler,y,doy differ significantly between regions. Without any
years are, simulated with model 14 (Table 1): 104, 97 and 77. In prior knowledge 65% of the variation in regional weekly sowing
1994 it was raining a lot in the period from day 70 to 100. This percentages can be explained from weather (Table 3). The gap from
late start would be properly modelled with the rain rule alone (i.e. 65% to 81% (Table 2) is due to misestimation of the start date. The
suitabler,y,doy ), but the late start is also explained from our model remaining gap (19%) may be due to variation of weather within
for simulating the start date. Sowing in 1994 started on day 108 each region, due to simplifications made in our model and due to
and then immediately peaked (according to the black squared val- other biophysical and socio-economic factors not included in our
idation data shown in Fig. 5 and also in our calibration data which models. Accurately reproducing weekly sowing percentages per
are not shown here). In most years however there is less rainfall definition also guarantees that regional average sowing dates are
in the same period and there are ample sowing opportunities. On accurately reproduced. Fig. 6 shows that this is indeed the case, our

Table 3
Rules for defining what makes a day suitable for sowing.

Model Calibration of start Rule for starta Scale for suitable Variable for suitable Percentsown R2 Parameters

1 Regional Simulated Regional Rain Regional 0.65 46

2 Regional Simulated National Rain National 0.52 31
3 National Simulated Regional Rain Regional 0.53 27
4 National Simulated National Rain National 0.46 12
5 Regional Target Regional Rain Regional 0.43 24
6 Regional Target National Rain National 0.38 9
7 Null model National National 0
a
The target start date is for each region the earliest out of 15 observed start dates (Eq. (3)); the simulated start date is estimated with model 9 in Table 1.
108 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111

Fig. 5. Suitable days for sowing in the Flevoland region. Black line: rainfall, white squares: suitable days according to our model (8 days ≤ 18 mm), black squares: validation
data: number of individual farmers sowing on each day.

model simulates accurately and unbiasedly the regional average 3.4. Validation
sowing dates in each of the regions, with an overall R2 of 0.82 (as
opposed to 0.92 with observed start dates). Table 3 clearly shows Of the days on which farmers sowed, 75% were also suitable
that variation in the start of the sowing period is the most important according to our model (Table 4, with regionally calibrated start
factor in determining variation in interannual differences in sowing date and regional rain rule). When we use observed start dates
dates. Simulations with the target sowing date were markedly less this number increases to 86% (data not shown). On a day in which
accurate than those with simulated start dates (hypothesis H5 vs none of the farmers sowed it is impossible to tell whether this was
H6) and more accurate results were obtained when start date was due to pure chance or whether the day was actually unsuitable.
regionally calibrated (hypothesis 9). According to the validation, 499 out of 1189 (42%) of the dates on

Fig. 6. Simulated and observed average sowing dates for the six regions, simulated with Table 3, model 1. Average means averaged across all farmers in the region; each
subplot contains 15 points, one for each year (1996–2010).
 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111 109

Table 4 a reason for not sowing too early (e.g. Maton et al., 2007; Rötter
Validation: observed sowing dates versus sowing dates suitable according to the
et al., 2011) and not the positive effects of frost on soil structure
best full model (Table 3, model 1).
(Benoit, 1973; Bullock et al., 1988; Arvidsson et al., 2000; Oztas and
Simulated suitable Observed sowing Fayetorbay, 2003). We suspect that also in other long term studies
No Yes Total Accuracy (Kucharik, 2006; Jaggard et al., 2007; Kaukoranta and Hakala, 2008;
Siebert and Ewert, 2012), the frost effect may explain why no strong
0 690 336 1026 0.67
1 499 995 1494 0.67 temperature signal was found. The interaction with frost days also
Total 1189 1331 2520 provides an explanation for why the response of wild plant to tem-
Accuracy 0.58 0.75 0.67 perature rise is stronger than farmers’ response to temperature rise
(Menzel et al., 2006a,b). Farmers can await more frost days and they
can do extra spring soil tillage in the absence of frost days, both
which farmers did not sow were suitable according to our model. of which result in delayed sowing. Wild plants do not make such
Together this leads to an overall accuracy of our model of 67%, conscious decisions.
which is similar to accuracy the found in the calibration (Table 3,
model 1, R2 = 0.65). As noted in Section 2.2, this accuracy estimate 4.2. Reflection on methods
is conservative. It is possible that no sowing occurred on a day by
chance, which does not necessarily mean that a day with no sowing A strength of our work is that we systematically test hypotheses
observed was actually unsuitable for sowing. Real accuracies of on how a series of weather variables might influence farmers’ sow-
our model are probably higher than those listed in Table 4. ing dates. Our model is in set-up similar to that of Maton et al.
(2007) but our method of calibration and hypothesis testing is
3.5. Climate change more rigorous, the number of weather variables included is larger
and the datasets used for calibration and validation are larger. Our
We can use the equations derived in this paper to simulate what approach for calculating trafficability and workability is relatively
impact climate change will have on sowing dates, assuming our simple. We used rainfall, rainfall minus potential evapotranspira-
rules will stay the same in the future. Climate scenarios are par- tion, and groundwater levels, but we did not use a mechanistic
ticularly uncertain about the magnitude and direction of rainfall model of soil moisture (such as in Hammer et al., 1987; Rounsevell,
and radiation change (Solomon et al., 2007). The numbers below 1993; van der Neut et al., 1995; Earl, 1997; Raper, 2005; Maton
are presented more to get a feel for the order of magnitude of cli- et al., 2007; Rötter et al., 2011). Such a model would require addi-
mate change effects and are in no way meant to be predictive. tional data and with lack of such data we saw no real benefit gained
We calculate possible changes in the start of the sowing period, from using more mechanistic approaches. We lack data on where
using models 9 and 14 from Table 1, simply by adding changes to and when within each region sugarbeet was sown. Therefore we
our weather data. If daily average temperature increases by 1 ◦ C could not link our sowing dates to location specific groundwater
the start of the sowing period would become 3.4/4.0 days earlier data and soil texture data. Many farmers in the Netherlands have
(model 9/14); if daily minimum temperature (TN) increases by the drainage pipes in their fields to enhance drainage, but the quantity
same amount the number of frost days would decrease and so the and quality of these pipes at different locations in the country is
start of the sowing period would become 1.5/2.1 days later. The net unknown. This too makes it difficult to estimate drainage rates. As
effect of a 1 ◦ C temperature increase is a shift to 1.9 days earlier. A a result of these unknowns we preferred using a more empirical
7 ◦ C temperature increase would lead to sowing becoming 20 days approach based only on rainfall. A consequence of our more empir-
earlier. ical approach is that the rainfall rules derived here are probably not
A 10% increase in radiation would result in sowing 7.7/5.0 days directly applicable in other environments.
earlier. A 10% increase in rainfall would result in a 0.8/0.5 days later Our model could explain 65% of the variation in start dates
start of the sowing period. Next we simulated the effect of rainfall when calibrated at a national level and 82% when calibrated at
increase on sowing beyond the start of the sowing period, with regional level. Beyond the start date 81% of the sowing dates could
Table 3 model 1. This added on average a 0.8 days delay, thus the be explained from rainfall, with the effect of rainfall being differ-
combined effect of 10% rainfall increase would be sowing 1.6/1.3 ent on sand and clay, which is logical in relation to the rate at
days later. In what direction and magnitude radiation and rainfall which they dry up through drainage and evaporation. Finding a
will change is unclear. But it is clear that any change in radiation will high correlation between sowing date and weather variables, or
have a much larger effect than changes in rainfall or temperature. a statistically significant increase in accuracy, does not guarantee
that the relationship found is grounded in a real causal process.
4. Discussion The question about causality becomes especially pertinent in the
case of radiation, which we discuss in the next paragraph. Correla-
4.1. Temperature effect on sowing date tion between weather variables exists and hampers the ability to
draw conclusions on their relative importance. Due to these corre-
Previous studies and also this study have shown, contrary to lations it becomes more difficult to quantify the relative importance
expectations, little to no relation between spring temperature and of weather variables and such inferences are logically impossible
farmers’ sowing dates. Our analyses showed that farmers do sow in case of perfectly correlated weather variables. But, as we have
earlier with higher spring temperatures, but that this effect is shown, the correlations were in this study sufficiently low to allow
largely offset by the fact that they sow later if there have been us to quantify whether each weather variable had explanatory
less frost days. The start date of the sowing period in a region was power.
found to depend mostly on temperature and frost days, beyond The most surprising outcome of our work is finding significant
the start date we found that sowing depends on rainfall and soil correlation between radiation sum (August–mid April) and the start
texture and not at all on temperature. We logically expect the tem- of the sowing period. Also when other more obvious explanatory
perature and frost days effect also to be present in other cropping variables were accounted for, adding radiation to our models signif-
systems in temperate environments that have frost during late win- icantly (p < 0.01) increased the accuracy. Three new questions arise
ter or spring. Previous studies on crop yields and sowing dates have (1) are other phenological events or human decisions also corre-
considered only the detrimental effects of frost on crop growth as lated with radiation sum and (2) is there an underlying process
110 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111

responsible for the correlation between radiation sum and start of pressures and less problems from sub soil compaction. It has also
the sowing period? Or (3) is radiation correlated with yet another been suggested that farmers may be hesitant to sow earlier than
variable not included in our models, with this other variable being now out of fear for increased bolting risk. Our calculations for the
a real source of variation in start dates? The first question is rela- Flevoland region, using 15 years of weather data and the Milford
tively simple to answer. Databases of phenological events and farm et al. (2010) model with parameters of the most sensitive cultivars
management decisions do exist and it is simply a matter of fitting indicated that, when sown on day 80, 20 March (the current average
regression models for these time series in a similar manner as out- for this region is day 99, 9 April), there would still be no bolting risk.
lined in this paper. This being said it should be noted that despite The potential yield gain from earlier sowing and the apparent lack of
an intensive search, we found not a single other study reporting on unsurpassable constraints for earlier sowing is reason for the Dutch
a relation between phenological events or human behaviour and Institute of Sugar Beet Research to actively promote earlier sowing.
radiation calculated over periods as considered here. The second Many other studies worldwide have quantified with crop growth
and third questions on causes of the relation between radiation models the potential yield gains from earlier sowing. To actually
and choice of sowing date are much more difficult to answer. We realise a shift to earlier sowing we first need to better understand
have not a clue about what could cause the strong correlation and why farmers do not sow as early as seems optimal according to the
look forward to any hypotheses and suggestions from the scientific models. And then in the next step investigate options for alleviat-
community. ing constraints for earlier sowing. We recommend more research
into what factors affect farmers’ sowing dates, also in other parts of
4.3. Implications the world.

Our work has implications for climate change adaptation

5. Conclusions
research. According to White et al. (2011) adjusting sowing date
is by far the most frequently investigated climate change adapta-
Our work provides a plausible explanation for why temperature
tion option. In indeterminate crops like sugar beet earlier sowing
rise has in the past lead to only small shifts in farmers’ sowing dates.
has an immediate yield increasing effect (Jones et al., 2003; Jaggard
Farmers do sow earlier when temperature rises. They sow later
et al., 2007). In determinate crops like cereals it is the combination
when in the preceding weeks there has been less frost, which is
of earlier sowing and selection of cultivars with long duration crops
related to crumbling of the soil. And since temperature rise leads to
(with a higher temperature sum) that leads to large yield gains (e.g.
a reduction in number of days with frost, the advance in sowing due
see Olesen and Bindi, 2002; Tubiello et al., 2000; Rötter et al., 2011).
to temperature rise is to a large extent offset by the delay in sowing
The simulated advances in sowing dates are often much larger than
due to less frost days. If this relation between temperature, frost
may logically be expected based on simulations of future sowing
days and sowing dates remains the same in the future then sowing
dates such as presented in this paper and larger than may logi-
will become only slightly earlier with temperature rise, 2 days per
cally be expected based on extrapolation of historical trends (such
1 ◦ C according to our models. Simulation studies that anticipate on
as reported in Kucharik, 2006; Menzel et al., 2006b; Jaggard et al.,
earlier sowing as an adaptation option to climate change may be
2007; Maton et al., 2007; Kaukoranta and Hakala, 2008; Siebert
overly optimistic on projected changes in sowing dates and overly
and Ewert, 2012; van Oort et al., 2012). How then should we inter-
optimistic about yield gains that follow from anticipated earlier
pret previous climate change scenario studies? Should we interpret
sowing. Scenario studies on climate change adaptation can be made
them as projections of future yields, based on the best available
more realistic and action-oriented if followed up by research on
knowledge? In that case they are probably overly optimistic on
how earlier sowing can actually be realised and by research on what
expected yield gains, because projected shifts in sowing dates are
is keeping farmers from not sowing earlier.
often larger than expected based on historical trends and the model
The surprising outcome of our analyses is that the start of the
presented in this paper. Another interpretation of such scenario
sowing period is also correlated with the sum of solar radiation
studies is to consider them as “what-if” studies. They show that
over the preceding autumn plus winter plus spring period. Farmers
yields can be increased if farmers sow earlier. For example our
sow earlier if this radiation sum is higher. The correlation between
additional calculations with the SUMO model (van Swaaij et al.,
radiation sum remains present in our data also when more obvious
2003) and the WOFOST model (van Diepen et al., 1989) show that
effects of rainfall, temperature, and frost days are accounted for.
in the Netherlands, yields can be increased by 5–6% if the sow-
This outcome raises new questions: is correlation with radiation
ing date were 15 days earlier than now, relative to the long term
also found in other human behaviour and in other living organ-
national average sowing date (day 100, 10 April). In 1998 there
isms? And secondly what is the causal mechanism behind it? Why
was a large spread in farmers’ sowing dates. The national aver-
do farmers sow earlier when there has been more radiation in the
age day on which the first 10% and the first 90% of the total area
preceding 7–10 months?
was sown was day 91 and day 130 (1 April and 10 May). Accord-
ing to our calculations with the WOFOST model, the yield penalty
of this 39 days delay was a 23% lower yield. This year is the most Acknowledgements
extreme case in our series of 15 years, in other years the spread
in sowing dates is smaller and accordingly so also the late sowing The work presented in this paper was funded through the
yield loss. AgriAdapt-NL project as part of the “Climate Changes Spatial
The practical and relevant follow-up question is what is keeping Planning” programme. We thank the following institutions for
farmers from sowing earlier. Our results suggest that tempera- providing their data: the Dutch Institute of Sugar Beet Research
ture and workability play an important role. The soil conditions (IRS) for calibration data on sowing dates, which are freely avail-
depend in part on tillage by the farmers. Arvidsson et al. (2000) able from www.bietenstatistiek.nl; the Suiker Unie company for
have shown that specific tillage methods may enable earlier sow- validation data on individual farmers; the Royal Netherlands Mete-
ing. Hanse et al. (2011) in a comparison between Dutch “top” and orological Institute for weather data, which are freely available
“average” farmers found that the “top” farmers sow their sugar- from www.knmi.nl; TNO for groundwater data, which are freely
beet 5 days earlier and suggested that this may be related to the available from www.dinoloket.nl. We thank Marjon Paas for sup-
top farmers doing more effort to ensure good trafficability. They port with interpreting the groundwater data. We thank Jan Vos for
use less passes for seedbed preparation, have lower tire inflation pointing us to the positive effects of frost.
 P.A.J. van Oort et al. / Europ. J. Agronomy 40 (2012) 102–111 111

References Milford, G.F.J., Jarvis, P.J., Walters, C., 2010. A vernalization-intensity model to predict
bolting in sugar beet. Journal of Agricultural Science 148 (2), 127–137.
Arvidsson, J., Rydberg, T., Feiza, V., 2000. Early sowing – a system for reduced seedbed O’Brien, R.M., 2007. A caution regarding rules of thumb for variance inflation factors.
preparation in Sweden. Soil and Tillage Research 53 (2), 145–155. Quality and Quantity 41 (5), 673–690.
Benoit, G.R., 1973. Effect of freeze–thaw cycles on aggregate stability and hydraulic Olesen, J.E., Bindi, M., 2002. Consequences of climate change for European agricul-
conductivity of three soil aggregate sizes. Proceedings of the Soil Science Society tural productivity, land use and policy. European Journal of Agronomy 16 (4),
of America 37, 3–5. 239–262.
Bullock, M.S., Kemper, W.D., Nelson, S.D., 1988. Soil cohesion as affected by freezing, Oztas, T., Fayetorbay, F., 2003. Effect of freezing and thawing processes on soil aggre-
water content, time and tillage. Soil Science Society of America Journal 52 (3), gate stability. Catena 52 (1), 1–8.
770–776. Öztürk, Ö., Topal, A., Akinerdem, F., Akgün, N., 2008. Effects of sowing and har-
Campbell, D.J., O’Sullivan, M.F., 1991. The cone penetrometer in relation to traffica- vesting dates on yield and some quality characteristics of crops in sugar
bility, compaction, and tillage. In: Smith, K.A., Mullins, C.E. (Eds.), Soil Analysis: beet/cereal rotation system. Journal of the Science of Food and Agriculture 88 (1),
Physical Methods. Marcel Defier, New York, pp. 399–429. 141–150.
Cleland, E.E., Chuine, I., Menzel, A., Mooney, H.A., Schwartz, M.D., 2007. Shifting plant Raper, R.L., 2005. Agricultural traffic impacts on soil. Journal of Terramechanics 42
phenology in response to global change. Trends in Ecology and Evolution 22 (7), (3–4), 259–280.
357–365. Richter, G.M., Qi, A., Semenov, M.A., Jaggard, K.W., 2006. Modelling the variability of
Earl, R., 1997. Prediction of trafficability and workability from soil moisture deficit. UK sugar beet yields under climate change and husbandry adaptations. Soil Use
Soil and Tillage Research 40 (3–4), 155–168. and Management 22 (1), 39–47.
Hammer, G.L., Woodruff, D.R., Robinson, J.B., 1987. Effects of climatic variability and Rötter, R.P., Palosuo, T., Pirttioja, N.K., Dubrovsky, M., Salo, T., Fronzek, S., Aikasalo, R.,
possible climatic change on reliability of wheat cropping – a modelling approach. Trnka, M., Carter, T.R., 2011. What would happen to barley production in Finland
Agricultural and Forest Meteorology 41 (1–2), 123–142. if global warming exceeded 4 ◦ C? A model-based assessment. European Journal
Hanse, B., Vermeulen, G.D., Tijink, F.G.J., Koch, H.-J., Märländer, B., 2011. Analysis of Agronomy 35 (4), 205–214.
of soil characteristics, soil management and sugar yield on top and averagely Rounsevell, M.D.A., 1993. A review of soil workability models and their limitations
managed farms growing sugar beet (Beta vulgaris L.) in the Netherlands. Soil in temperate regions. Soil Use and Management 9 (1), 15–21.
and Tillage Research 117, 61–68. Schwartz, M.D., Ahas, R., Aasa, A., 2006. Onset of spring starting earlier across the
Hoffmann, C.M., Kluge-Severin, S., 2010. Light absorption and radiation use effi- Northern Hemisphere. Global Change Biology 12 (2), 343–351.
ciency of autumn and spring sown sugar beets. Field Crops Research 119 (2–3), Scott, R.K., English, S.D., Wood, D.W., Unsworth, M.H., 1973. The yield of sugar beet
238–244. in relation to weather and length of growing season. The Journal of Agricultural
Hoffmann, C.M., Kluge-Severin, S., 2011. Growth analysis of autumn and spring sown Science 81 (2), 339–347.
sugar beet. European Journal of Agronomy 34 (1), 1–9. Siebert, S., Ewert, F., 2012. Spatio-temporal patterns of phenological development
Jaggard, K.W., 1977. Effects of soil density on yield and fertilizer requirement of in Germany in relation to temperature and day length. Agriculture and Forest
sugar beet. Annals of Applied Biology 86, 301–312. Meteorology 152, 44–57.
Jaggard, K.W., Qi, A., Semenov, M.A., 2007. The impact of climate change on sugarbeet Smit, A.L., 1993. The influence of sowing date and plant density on the decision to
yield in the UK: 1976–2004. Journal of Agricultural Science 145 (4), 367–375. resow sugar beet. Field Crops Research 34 (2), 159–173.
Jones, P.D., Lister, D.H., Jaggard, K.W., Pidgeon, J.D., 2003. Future climate impact on Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tig-
the productivity of sugar beet (Beta vulgaris L.) in Europe. Climatic Change 58 nor, M., Miller, H.L. (Eds.), 2007. Contribution of Working Group I to
(1–2), 93–108. the Fourth Assessment Report of the Intergovernmental Panel on Climate
Kaukoranta, T., Hakala, K., 2008. Impact of spring warming on sowing times of cereal, Change. Cambridge University Press, Cambridge, United Kingdom/New York,
potato and sugar beet in Finland. Agricultural and Food Science 17 (2), 165–176. NY, USA.
Kucharik, C.J., 2006. A multidecadal trend of earlier corn planting in the central USA. Tubiello, F.N., Donatelli, M., Rosenzweig, C., Stockle, C.O., 2000. Effects of climate
Agronomy Journal 98 (6), 1544–1550. change and elevated CO2 on cropping systems: model predictions at two Italian
Lauer, J.G., 1997. Sugar beet performance and interactions with planting date, geno- locations. European Journal of Agronomy 13 (2–3), 179–189.
type, and harvest date. Agronomy Journal 89 (3), 469–475. van Diepen, C.A., Wolf, J., Van Keulen, H., Rappoldt, C., 1989. WOFOST: a simulation
Makkink, G.F., 1957. Testing the Penman formula by means of lysimeters. Journal of model of crop production. Soil Use and Management 5 (1), 16–24.
the Institution of Water Engineers 11, 277–288. van der Neut, D., van Dam, J.C., Feddes, R.A., 1995. Effecten van peilopzetten in de
Makkink, G.F., 1959. Lysimeters in the Netherlands. In: Symposium of Hannoversch- Hoeksche Waard: een evaluatie over 42 jaren van de opbrengstveranderingen
Münden.Vol. II: Lysimeters. IASH Publication No. 49. IASH, pp. 5–12. van aardappelen en bieten bij vier verschillende dieptes van de drainagebasis.
Maton, L., Bergez, J.-E., Leenhardt, D., 2007. Modelling the days which are agro- Landbouwuniversiteit, Wageningen, p. 79 (in Dutch).
nomically suitable for sowing maize. European Journal of Agronomy 27 (1), van Oort, P.A.J., Timmermans, B.G.H., Meinke, H., van Ittersum, M.K., 2012. Key
123–129. weather extremes affecting potato production in the Netherlands. European
Menzel, A., Sparks, T.H., Estrella, N., Koch, E., Aaasa, A., Ahas, R., Alm-Kübler, K., Bis- Journal of Agronomy 37 (1), 11–22.
solli, P., Braslavská, O., Briede, A., Chmielewski, F.M., Crepinsek, Z., Curnel, Y., van Swaaij, A.C.P.M., Withagen, L.M., Schiphouwer, T., Smit, A.B., 2003. SUMO: a
Dahl, Å., Defila, C., Donnelly, A., Filella, Y., Jatczak, K., Måge, F., Mestre, A., Nordli, growth model for predicting the yield and quality of sugar beet. In: Proceedings
Ø., Peñuelas, J., Pirinen, P., Remišová, V., Scheifinger, H., Striz, M., Susnik, A., of the Joint Colloquium on Sugar Beet Growing and Modelling, Lille (France),
Van Vliet, A.J.H., Wielgolaski, F.-E., Zach, S., Zust, A., 2006a. European phenolog- September 12th, 2003.
ical response to climate change matches the warming pattern. Global Change White, J.W., Hoogenboom, G., Kimball, B.A., Wall, G.W., 2011. Methodologies for
Biology 12 (10), 1969–1976. simulating impacts of climate change on crop production. Field Crops Research
Menzel, A., von Vopelius, J., Estrella, N., Schleip, C., Dose, V., 2006b. Farmers’ annual 124 (3), 357–368.
activities are not tracking the speed of climate change. Climate Research 32 (3), Wind, G.P., 1960. Opbrengstderving door te laat zaaien. Landbouwkundig Tijdschrift
201–207. 72 (4), 111–118 (in Dutch).