Agricultural and Forest Meteorology 101 2000 167–186
Scaling-up the AFRCWHEAT2 model to assess phenological development for wheat in Europe
Paula A. Harrison
a ,∗
, John R. Porter
b
, Thomas E. Downing
a
a
Environmental Change Institute, University of Oxford, 1a Mansfield Road, Oxford OX1 3TB, UK
b
Department of Agricultural Sciences, Royal Veterinary and Agricultural University, Agrovej 10, 2630 Taastrup, Denmark Received 2 June 1999; received in revised form 15 November 1999; accepted 22 November 1999
Abstract
A method was developed for scaling-up the AFRCWHEAT2 model of phenological development from the site to the continental scale. Four issues were addressed in this methodology: i the estimation of daily climatic data from monthly
values, ii the estimation of spatially variable sowing dates, iii the simulation of multiple cultivars, and iv the validation of broad-scale models. Three methods for estimating daily minimum and maximum temperatures from monthly values were
compared using AFRCWHEAT2: a sine curve interpolation, a sine curve interpolation with random daily variability, and two stochastic weather generators WGEN and LARS-WG. The sine curve interpolation was selected for the continental scale
application of AFRCWHEAT2 because computational time was short and errors were acceptably small. The average root mean square errors RMSEs for the dates of double ridges, anthesis and maturity were 6.4, 2.2 and 2.1 days, respectively.
The spatial variability of European sowing dates was reproduced using a simple climatic criterion derived from the AFR- CWHEAT2 vernalization curve. The use of several cultivar calibrations enabled the broad-scale model to capture current
responses and compare responses to future climate change. Results from the continental scale model were validated using a geographically-referenced database of observed phenological dates, output from other site-based models and sensitivity anal-
ysis. The spatial model was able to emulate a similar spatial and temporal variability in phenological dates to these sources under the present climate. The predominant effect of an increase in mean temperature was a reduction in the emergence to
double ridges phase. The shift in the timing of subsequent development stages to earlier in the season meant that changes in their duration were relatively minor. Changes in inter-annual temperature variability resulted in only small changes in the
mean date of development stages, but their standard deviation altered significantly. ©2000 Elsevier Science B.V. All rights reserved.
Keywords: Site modelling; Spatial modelling; Winter wheat; Phenological development; Climate change; Scaling-up
1. Introduction
Scale is an inherent concern in resource manage- ment and planning, and hence in crop-climate im-
∗
Corresponding author. Tel.: +44-1865-281186; fax: +44-1865-281181.
E-mail address: paula.harrisoneci.ox.ac.uk P.A. Harrison.
pact assessment. Farmers are most often concerned with conditions in their fields. Even at this micro-scale
however, variability between and within fields is high Church and Austin, 1983; Russell and van Gardingen,
1997, one of the key motivations behind the adop- tion of precision agriculture. Farmers are also inher-
ently concerned with regional and global production — prices, markets and competitiveness. Commodity
0168-192300 – see front matter ©2000 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 8 - 1 9 2 3 9 9 0 0 1 6 4 - 1
168 P.A. Harrison et al. Agricultural and Forest Meteorology 101 2000 167–186
boards and market regulators are more concerned with spatially aggregated values than production at a single
site. Many policy decisions to mitigate or adapt to cli-
mate change must take place at the regional scale. For example, the European Union and member states
are concerned with ‘What will be the impact of cli- mate change on national and European wheat produc-
tion?’ rather than ‘What will be the impacts of climate change on potential wheat yields for specific fields or
typical fields in a local area?’ It is not often feasible to perform experiments on crops at any scale larger than
the field. Thus, there is a need to scale-up site scale ex- perimental and modelling observations to study such
larger scale effects.
To address these concerns, two different scales of assessment are commonly used. Experimental pro-
grammes and site-based mechanistic crop models have been used to analyse detailed physiological
responses of crops to changes in environment at indi- vidual locations e.g., Mearns et al., 1992; Miglietta
and Porter, 1992; Kocabas et al., 1993; Semenov et al., 1996; Bindi and Fibbi, 1999; Wolf, 1999. At
broader scales, simpler crop models have been used to analyse changes in aggregated response and spatial
shifts in crop suitability and productivity e.g., Kenny and Harrison, 1992; Brignall and Rounsevell, 1995;
Carter and Saarikko, 1996; Harrison and Butterfield, 1996; Olesen et al., 1999.
The clear need is to link these two scales — the detailed, process-based, local site understand-
ing and the regional, agroecology and landscape level of spatial planning. A primary concern when
scaling-up is to retain the advantages of detailed site models Harrison and Butterfield, 1996. The mech-
anistic, process-orientation of site models produces more reliable predictions of responses to possible
future changes in climate than simpler spatial mod- els whether reduced-form or empirical correlations
Carter et al., 1994. Several methods have been em- ployed to scale-up site models to regional assessments
van Gardingen et al., 1997; Downing et al., 1999. At the simplest level, results for representative sites
are aggregated to a regional value e.g., Easterling et al., 1993; Wolf, 1993; Rosenzweig and Parry, 1994.
More spatially explicit approaches involve applying a site model to input data sets which are spatially
interpolated to regular grids andor coherent polygons e.g., Brklacich et al., 1996; Rounsevell et al., 1996;
Easterling et al., 1998; Butterfield et al., 1999; Carter et al., 1999. More quantitative methodologies rely on
relating the site characteristics to its spatial domain using remotely sensed and other environmental data
sets e.g., Bindi et al., 1999; Delécolle, 1999.
At larger spatial scales, feedbacks from veg- etation and land processes to the climate itself
also become important. These issues are consid- ered in soil–vegetation–atmosphere transfer schemes
SVATs, which are used to represent land processes in global climate models Koster and Suarez, 1994.
Such feedbacks will in turn affect regional crop production. However, the feedbacks of land-surface
processes to the atmosphere are complex and still far from understood Dickinson et al., 1996 and are thus
considered to be outside the remit of this paper.
This paper
critically evaluates
methods for
scaling-up the phenology model of AFRCWHEAT2 Weir et al., 1984; Porter et al., 1987; Porter, 1993
from the site scale to a large European region. We discuss restrictions, such as the availability of input,
calibration and validation data sets, to the application of mechanistic crop models across large regions. We
then evaluate the sensitivity of broad-scale wheat de- velopment to changes in temperature, such as might
be experienced with climate change.
AFRCWHEAT2 models
wheat development
through the interaction of thermal time, photope- riod and vernalization. Dates of emergence, double
ridges, terminal spikelet, anthesis, beginning and end-of-grain filling and physiological maturity are
calculated by the model for a specific cultivar. Double ridges is an early reproductive apical development
stage; terminal spikelet marks the end of the cre- ation of grain-producing spikelet primordia. Input
data required by AFRCWHEAT2 for phenological development are daily minimum and maximum air
temperature, sowing date and latitude. Calibration and validation data for an appropriate wheat variety
at the site of interest are also required. These data are not available across large regions, such as Europe. We
address four specific questions in this paper in order to scale-up the AFRCWHEAT2 model:
1. How can daily climatic input data be estimated
from monthly values, with acceptable estimates of phenological timing and spatial errors of pheno-
logical stages?
P.A. Harrison et al. Agricultural and Forest Meteorology 101 2000 167–186 169
2. What climatic criterion can be used to determine realistic, spatially variable sowing dates across
Europe? 3. How can different wheat varieties cultivated in
Europe be represented in the broad-scale model? 4. How can the scaled-up broad-scale model be vali-
dated?
2. Methods