Agronomy Journal Grow Your Career With ASA
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via ISI Web of Science (16)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Jones, P. G.
Right arrow Articles by Thornton, P. K.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Jones, P. G.
Right arrow Articles by Thornton, P. K.
Agricola
Right arrow Articles by Jones, P. G.
Right arrow Articles by Thornton, P. K.
Related Collections
Right arrow Agroclimatology
Right arrow Other Models
Right arrow Software
Agronomy Journal 92:445-453 (2000)
© 2000 American Society of Agronomy

SOFTWARE

MarkSim

Software to Generate Daily Weather Data for Latin America and Africa

Peter G. Jonesa and Philip K. Thorntonb

a Centro Internacional de Agricultura Tropical (CIAT), AA6713, Cali, Colombia
b Int. Livestock Research Inst. (ILRI), P.O. Box 30709, Nairobi, Kenya

p.jones{at}cgiar.org


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 Program description
 Results
 Discussion
 REFERENCES
 
A software package to generate daily weather data for Latin America and Africa is described. The program is based on a stochastic weather generator that uses a third-order Markov process to model daily weather data. The model has been fitted to data from more than 9200 stations with long runs of daily data throughout the world. The climate normals for these stations were assembled into 664 groups using a clustering algorithm. For each of these groups, rainfall model parameters are predicted from monthly means of rainfall, air temperature, diurnal temperature range, and station elevation and latitude. The program identifies the cluster relevant to any required point using interpolated climate surfaces at a resolution of 10 min of arc (18 km2) and evaluates the model parameters for that point. The application currently contains surfaces for Latin America and Africa, and other regions will later be added. Use of the software is demonstrated by generating daily weather data files for running one of the DSSAT crop models.

Abbreviations: CD-ROM, compact disk–read-only memory • CGIAR, Consultative Group on International Agricultural Research • CIAT, Centro Internacional de Agricultura Tropical • CPU, central processing unit • DSSAT, decision support system for agrotechnology transfer • ICASA, International Consortium for Agricultural Systems Applications • ILRI, International Livestock Research Institute


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
Program description
 Results
 Discussion
 REFERENCES
 
THE PRINCIPAL CONSTRAINTS

to many important applied research and development activities do not lie in the methodologies and tools that are employed or that need to be developed, but in the availability and format of data. In particular, data shortages with respect to soils, weather, and appropriate socioeconomic information constitute severe impediments to many impact assessments, characterization studies, and policy analysis frameworks that use tools such as simulation models and geographic information systems (GIS).

Weather is a primary determinant of agricultural production. Weather data are needed for a wide range of problems at many different scales. At one extreme lies the issue of continental climate and agroecological classification, where available weather data (usually in the form of monthly means) are taken and interpolation is carried out between sites to form the basis of such a classification. At the other extreme, work may be done at the plot level, for example, using weather data by the minute or hour to provide information concerning the impact of technological interventions on agricultural enterprises of the individual smallholder farmer. However, for many purposes, daily values are sufficiently detailed.

Some global and regional climate databases do exist (NCDC, 1994; IIMI, 1997; Texas A&M Univ. Systems, 1998). Currently, however, such products are limited in one or more of the following ways.

  1. Lack of interpolation facilities, meaning that analysis can be performed only for sites where data exist.
  2. Limited number of weather variables in the database, precluding the running of certain types of models and analytical tools.
  3. Limited number of years of historical data in the database, severely restricting the inferences that can be drawn concerning temporal weather variability at the site in question.
  4. Inappropriate temporal scale for many research applications, where scales of a month or dekad (10-d totals) may be insufficient.

In situations where appropriate data exist, a common standard of a minimum dataset and simple ASCII formatting greatly assists transport of data between models. For agronomic experiments, one such standard is that of the Decision Support System for Agrotechnology Transfer–International Consortium for Agricultural Systems Applications (DSSAT–ICASA). This was originally set up to facilitate data and model exchange between crop modeling groups in the USA, Canada, Europe, and Australia (van Kraalingen and Hunt, 1997; Tsuji et al., 1998). The DSSAT–ICASA standards are not perfect, but they are simple and starting to be used relatively widely.

This paper describes a software package that allows the user to choose a location or pixel in Latin America and Africa and generate daily weather data that is characteristic of the location and can be used to feed a wide range of crop models and other analytical tools.


   Program description
TOP
 ABSTRACT
 INTRODUCTION
 Program description
Results
 Discussion
 REFERENCES
 
Nearly 20 yr of work at CIAT have resulted in large databases of monthly mean temperature and rainfall data. Enormous effort has been expended in collecting and collating data from thousands of sites worldwide. More recently, a third-order Markov rainfall model has been constructed and extensively tested (Jones and Thornton, 1993). A Markov model works by randomly sampling a series of events where the probability of observing an event depends on the occurrence of previous events. A third-order Markov model takes into account events occurring over the previous 3 d. Previous rainfall models such as WGEN (Richardson, 1985), SIMMETEO (Geng et al., 1988), and WEATHERMAN (Pickering et al., 1994) tend to underestimate the variance of monthly and annual rainfall for many sites in the tropics and subtropics. Being able to model outlying rainfall years satisfactorily is particularly important in studies aimed at quantifying production system risk. The third-order rainfall generator has this ability built into it by means of random sampling of some of the parameters of the model itself. The model is fitted to historical daily data using an analysis of deviance based on a probit link function and binomial error vector. We use a generalized linear model as described by Nelder and Wedderburn (1972) and implemented in algorithm GO2GBF in the NAG library (NAG, 1990). The baseline probits (normal probabilities of rain on any day following 3 dry days) are estimated for each month. Three lag parameters are estimated and combined to form a matrix of transition probabilities for the year. The matrix has 96 elements, made up of 12 months by eight wet-or-dry states covering the previous 3 d (i.e., 23). Rainfall events (wet-day amounts) are assumed to have a gamma distribution with an event mean size and shape parameter for each month (Ison et al., 1971; Buishand, 1977; Garbutt et al., 1981; Coe and Stern, 1982). Monthly standard errors for the baseline probits and a variance–covariance matrix are used in the annual resampling.

The basic model has been fitted to data from more than 9200 stations with long runs (from 14 to 100 yr) of daily data throughout the world. These data have all been checked for consistency. We used contingency tables for missing values and data run lengths, gamma distribution probabilities for rainfall event outliers, and digital analysis for checking rainfall data entry errors.

Jones and Thornton (1997) showed that patterns could be discerned in the parameter values that were typical for certain types of climate. The model can thus be used to interpolate rainfall data where they do not exist. Regression models have been developed that predict the Markov model parameters within certain restricted climate sets (Jones and Thornton, 1999). The climate normals for this calibration set of stations were grouped into 664 groups using a leader clustering algorithm to produce very tight clusters in the scaled space of the 36 climate variables (Hartigan, 1975, p. 75–78). For each of these clusters, a regression model was found to predict the model parameters from the basic climate normals for rainfall, temperature, diurnal temperature range, and the elevation and latitude of the station. Many clusters were chosen to ensure that each individual regression model would have to predict parameters in a limited range of climates. This system identifies the cluster relevant to any required point on the globe using interpolated climate surfaces, and evaluates the model parameters for that point.

The climate surfaces used are the 10 minutes-of-arc surfaces fitted at CIAT, based on the NOAA data set TGOP006 (NOAA, 1984) using inverse square distance weights for spatial interpolation, and lapse rate to correct temperature for elevation effects.

Other surface fitting techniques, such as those of Hutchinson (1989, 1997) and Daly et al. (1993), have certain advantages, especially when fitting surfaces to smaller pixel sizes. Such surfaces may be added to the MarkSim system as they become available.

Input and Output Files
Figure 1 shows the major components of the software package. It operates in two stages. The first stage creates files with the model parameter estimates. These are designated CLX files and contain all the information for subsequent processing (Table 1 shows an example of a CLX file for Palmira, Colombia). The second stage creates either simulated rainfall files in calendar format or DSSAT–ICASA standard daily weather files (described below), which can be used to run the DSSAT crop models.



View larger version (35K):
[in this window]
[in a new window]
  		Fig. 1 Components of the software package, showing the three methods of generating daily data. Shaded boxes represent input and output files

 

View this table:
[in this window]
[in a new window]
  		Table 1 Site parameter file PALM.CLX.{dagger}

 
Three options determine how the weather data are generated (see Fig. 1). The user can specify:
  1. A point defined only by latitude and longitude. For this option, the parameters of the weather generator will be estimated from the 10-min interpolated climate files derived from the CIAT climate database (version 3.37 for Latin America and version 4.00 for Africa). These have a pixel size of about 18 by 18 km. For mountainous areas especially, there is a risk that the interpolated data are not representative of a given point within the pixel.
  2. A point defined by latitude, longitude, and elevation. Processing continues as for Option 1, but better estimates of the parameters for the weather generator should result (i.e., that produce more characteristic daily weather data), because temperatures are corrected for altitude.
  3. The long-term monthly climate normals for the desired point. This results in the best estimate of the parameters of the weather generator because the error caused by the fitted surface is eliminated. The normals required are rainfall, temperature, and diurnal temperature range (obtainable from maximum and minimum temperature).

The user interface allows the user to enter the location required directly using a point-and-click system, using the keyboard, or from a run control file. Latitude and longitude are specified negative west, negative south, and can be entered in decimal degrees or degrees and minutes.

If the user has a long list of sites for which simulated weather is required, inputs can be made using a control file that contains three options:

The control file is read in free format. However, every point in the control file must have a unique CLX filename. If the ultimate output is rainfall data in calendar format, the site name can be up to eight characters long. If output files are to be produced following the DSSAT–ICASA standard, the site code must consist of four characters. For example, "PALMIRA" is a valid CLX file name to produce calendar rainfall files characteristic of Palmira, Colombia (located at 3.5°N, -76.5°W), and "PALM" is a valid CLX file name for weather files in this format.

If the user wishes to enter observed climate normals, then these may be presented to the package as a file (designated DAT) for each simulation point. The name of the DAT file will be used in the subsequent simulation. The names of a whole sequence of DAT files can be located in a control file, allowing batch processing for many points at a time. Table 2 shows the fixed format of a DAT file.


View this table:
[in this window]
[in a new window]
  		Table 2 Site file PALMIRA.DAT containing historical long-term climate normals.{dagger}

 
Once the user has created the CLX parameter file or files that are needed, simulated weather files can be produced. The type of output can then be chosen. Calendar output will produce daily rainfall files for as many years as are specified, located in an output file 1 yr after another, starting from the year 3001 (to remind the user that these are generated data and not historical). These are always given the file extension .GEN. Table 3 shows the format. In this case, the only weather generator used is the standard third-order Markov rainfall model originally described in Jones and Thornton (1993).


View this table:
[in this window]
[in a new window]
  		Table 3 Calendar format for historical and generated rainfall: file PALM.GEN.{dagger}

 
If DSSAT–ICASA files are required, then these will be produced containing daily rainfall, solar radiation, and maximum and minimum temperatures, 1 yr per file (Table 4 shows an example of the format), with the extension .WTG to signify generated data as opposed to historical data. In this case, the third-order Markov rainfall model is used to generate rainfall data. The DSSAT weather generator (Pickering et al., 1994), based on routines of Richardson (1985) and Geng et al. (1988), is then used to generate daily values of solar radiation and maximum and minimum temperatures based on whether the day is wet or dry. The parameters for generating these variables are the long-term monthly means stored in the CLX site file. The monthly values of solar radiation are generated from the temperature normals using the model of Donatelli and Campbell (1997) that modifies and improves the earlier model of Bristow and Campbell (1984). If this option is chosen, a DSSAT climate file (CLI) will be produced, even though no information will be used from it for the generation of weather data. The CLI file is produced to ensure that it exists for subsequent use by any of the DSSAT models. Table 5 shows an example of a CLI file.


View this table:
[in this window]
[in a new window]
  		Table 4 Sequential format (DSSAT-ICASA v3.5 standard) for historical and generated weather—file PALM0101.WTG.{dagger}

 

View this table:
[in this window]
[in a new window]
  		Table 5 DSSAT-ICASA v3.5 format climate file PALM.CLI.{dagger}

 
For all weather generation, the seed for random number generation is taken from the system clock by default and is written to the screen. A facility exists for the user to override this and choose a particular seed. This gives some control, either from the keyboard or the control file, over the simulated weather sequences produced, which is especially useful for model debugging purposes.

Table 6 shows a listing of all the files included on the CD-ROM.


View this table:
[in this window]
[in a new window]
  		Table 6 Model-related files on the MarkSim v.1 CD-ROM

 
The DSSAT–ICASA Data Standard
The ICASA v1.0 data standard (van Kraalingen and Hunt, 1997) was developed from the Decision Support System for Agrotechnology Transfer (DSSAT) crop modeling system, the original version of which was released in 1989 (Jones et al., 1998). The latter itself grew out of the realization that minimum data sets of biophysical variables were required if agronomic experiments were to be used for extrapolation purposes through the medium of a crop model (Nix, 1987). The standard is designed to be used both to standardize input data formats for running detailed biophysical simulation models, and as a means of storing experimental data. Recent developments regarding the standard mean that a wide variety of crop models can now be run using these input–output standards (van Kraalingen and Hunt, 1997). A great attraction of these standards is that they specify an ASCII format, precluding the need for third-party software for manipulation.

Errors in the System
The third-order Markov rainfall model fits individual station data very well (see Jones and Thornton, 1993, for comprehensive testing of the model for three sites in the tropics). However, the interpolation system is only as good as the interpolated surface. Various deficiencies can be recognized in the surfaces presented in the current system; this is largely why different options are available for simulating daily data.

One type of problem occurs where the interpolated surface is plainly wrong. This may be because of errors or gaps in the data and in the interpolation method. A mapping of the sites from the calibration data set of more than 9200 stations reveals significant gaps where sufficiently complete and long-term data are not readily available (Fig. 2) . Currently, it is difficult to gain access to data for Sudan, Uganda, Zaire, Angola, Nigeria, Venezuela, and a significant list of other countries. Some countries are represented by very few stations. For large countries, this can vastly underestimate the climatic diversity present. Although much effort has been spent on preparing the surfaces to be free of data error, in practice this is extremely difficult to achieve. Such errors can be corrected over time as they are found. Other errors may occur in situations where the underlying digital elevation model is inadequate, and estimates are produced using incorrect elevation data.



View larger version (68K):
[in this window]
[in a new window]
  		Fig. 2 Global distribution of the calibration data set of 9221 stations with at least 14 yr of daily rainfall data (from Jones and Thornton, 1999)

 
The second type of error is inherent in the interpolated surface. The values stored for a given pixel represent, at most, the modal values for the pixel. For a surface with a precision of 10 arc minutes, it will often be found that the modal values do not represent a precise point within the pixel. The major discrepancies usually arise because of local elevation changes. It is mostly for this reason that in the package the option exists for the user to override the pixel elevation with the known elevation of the required site.

A third serious source of error is in the classification of a pixel into a climate cluster for model fitting. Despite deriving a global climate classification with 664 climate clusters, areas exist for which no cluster is really representative. This is because of data gaps in the calibration data set (Fig. 2). These are particularly noticeable in Central Africa and some areas of Latin America. The software package includes IDRISI images DAFRICA and DAMERICA to show the distance (in climate measure) between each pixel and the cluster to which it is allocated. These images are standardized to the climate cluster limits. A value <1.0 denotes that the pixel is within the limit set for cluster membership in the leader cluster algorithm. A value >1.0 shows that the model will be extrapolating from the nearest cluster centroid. Note that the distances are in climate, and not geographical, measure. The cluster centroid may be very close in climate measure, but geographically on the other side of the globe.

The fourth source of error is in the estimation of the parameters within the climate cluster (details are in Jones and Thornton, 1999). Although many clusters fit very well, some do not. The latter need to be checked for sources of error, and this will be done before the next revision of the software.


   Results
TOP
 ABSTRACT
 INTRODUCTION
 Program description
 Results
Discussion
 REFERENCES
 
Rainfall data generated from the surfaces have been comprehensively tested for three diverse sites in Africa and Latin America, and the results are reported in Jones and Thornton (1999). To summarize, rainfall data generated from the surfaces were tested against historical data for Guatemala City, Guatemala; Palmira, Colombia; and Tillabery, Niger in terms of the following:

The results of this testing suggest that some statistically significant discrepancies occurred between historical and generated rainfall data for these three diverse sites, but they were relatively minor (Jones and Thornton, 1999). It was concluded that for general information concerning rainfall patterns at a site, particularly where data are sparse, the method shows considerable promise. Users should certainly test generated rainfall obtained using the software against historical data, if they exist, to ascertain the model's performance for their own particular purposes.

To illustrate the use of the software, daily weather data for three contrasting sites in Kenya were generated. Table 7 shows details of the sites. The software was used to generate CLX files for the three sites (Kilifi, Machakos, and Kitale) using Option 2 (Fig. 1), that is, longitudes and latitudes were specified and temperatures corrected for using the actual elevation of each site, rather than the value from the digital elevation model. Twenty years of daily data in DSSAT–ICASA format were then generated for each site.


View this table:
[in this window]
[in a new window]
  		Table 7 Characteristics of the sites in Kenya used in the example simulations.{dagger}

 
To give an idea of the performance of the software, Table 8 shows the annual means and variances of the generated and historical rainfall data for the period from planting to physiological maturity in each of 20 yr. For each site, the probability distribution of generated rainfall was compared with that of historical rainfall as follows. First, each distribution (generated and historical) was tested for normality using the Lilliefors and Shapiro-Wilk tests (Conover, 1980). For those distributions where the null hypothesis of normality was not rejected at the 5% level, means and variances of historical and generated rainfall amounts were compared using the t-test and the F-test. For the cases where the hypothesis of normality was rejected at the 5% level for the generated and/or the historical rainfall amount distributions, means were compared using the Mann-Whitney test and variances compared using the Squared Ranks test (Conover, 1980). The generated and historical rainfall distributions themselves were then compared for each site using the Smirnov test and Cramer–von Mises test (Conover, 1980). For the three sites studied, no statistically significant differences at the 5% level were found for mean rainfall amounts, their variances, or the distribution of rainfall amounts over 20 growing seasons.


View this table:
[in this window]
[in a new window]
  		Table 8 Results of twenty seasons of maize simulations for three sites in Kenya and two sets of daily weather data

 
The rainfall files were then used to drive CERES–Maize (Ritchie et al., 1998). For illustrative purposes, the same maize variety (Katumani Composite B) was grown at all three sites, and planted at 4.2 plants m-2. Planting dates were 1 April for Kilifi and Kitale, and 15 March for Machakos, fairly typical for these locations (Hassan, 1998). Each soil profile (FURP, 1987; Wafula et al., 1998) was assumed to have some 50 kg ha-1 of mineral N in the profile at planting. Each growing season was replicated 20 times using generated weather data. The three treatments were then repeated using 20 yr of historical daily weather recorded for the three sites, with temperatures and solar radiation generated using the DSSAT weather generator (Pickering et al., 1994).

Results are shown in Fig. 3 , in terms of simulated maize yield distributions for the six treatments (three sites by two sets of daily weather data). Table 8 also shows results in terms of simulated days from planting to maturity in these environments. The cumulative yield distributions show that the simulations using weather data generated from the interpolated surfaces compare reasonably well with the distributions obtained using historical rainfall data. Kitale is in an important maize-producing region of Kenya, and yield potential is generally high. Kilifi, in the coastal lowlands, is in a much lower potential zone, but the distributions indicate that some maize yield is forthcoming in almost all years; the probability of total failure owing to the rains is small. Machakos, in the dry transitional zone, also has limited potential for maize production, largely because of the highly variable rainfall. The simulations indicated that the probability of essentially no maize yield because of inadequate rainfall was about 1 in 5 yr, for both sets of weather data. Table 8 bears out the highly risky nature of the environment for maize production in Machakos; the coefficients of variation of both yield and seasonal rainfall are very high. The coefficient of variation for total seasonal precipitation for the lowland tropics (such as Kilifi) is 36%, for the dry transitional zone (such as Machakos) is 40%, and for the moist transitional zone (such as Kitale) is 27% (Hassan, 1998, p. 52). The figures derived in this example are comparable.



View larger version (26K):
[in this window]
[in a new window]
  		Fig. 3 Cumulative probability distributions for simulated maize yield at three sites in Kenya using generated rainfall data and historical rainfall data

 

   Discussion
TOP
 ABSTRACT
 INTRODUCTION
 Program description
 Results
 Discussion
REFERENCES
 
The software package outlined above appears to generate daily weather data that are satisfactory for many types of analysis—notably where the time period under consideration is more than about a few days. For situations where almost nothing is known about a particular location in terms of long-term monthly means, the package provides at least a first approximation to weather data. These can then be used for a wide variety of purposes, including the driving of detailed simulation models with time steps of 1 d. In situations where more is known about the location, then more confidence can be felt in the data produced. As noted above, there are almost certainly errors in Version 1 of MarkSim. We feel confident that it will work well in most places, but we welcome feedback to track down some of the inevitable errors that exist. At the same time, it is planned to collect more daily data to fill in the gaps shown in the map in Fig. 1. In future, it is hoped to expand the product to include Asia, and possibly to increase the resolution of the surfaces, based on newer digital elevation models.

Documentation
A set of user notes is distributed with the CD-ROM containing the package, together with the text of this paper. The full text of Jones and Thornton (1993, 1997, 1999) is also included, by kind permission of Elsevier Science Publishers Ltd.

Hardware and Software Requirements
The software runs on any personal computer running Windows. It has been tested under Windows 3.1, Windows 95, and Windows NT, and performs without problems so far. It has been developed and tested on machines with at least 32 Mb RAM, but it should run on as little as 16 Mb. The parameter estimation and data generation programs are written in FORTRAN and are run in DOS windows from within the Windows user interface. About 52 Mb of hard disk space are required for the software. Parameter estimation for a location takes about 2 s with an Intel 486 processor and less for faster CPUs. Weather data generation similarly takes about 1 s yr-1 on an Intel 486 processor. To view the IDRISI images provided on the CD-ROM, the IDRISI v. 4.1 DOS-based routine COLOR.EXE is provided, with kind permission of Dr J.R. Eastman (Eastman, 1993).

Availability
The CD-ROM is available from the corresponding author, Dr. Peter G. Jones, Centro Internacional de Agricultura Tropical (CIAT), AA6713, Cali, Colombia, email p.jones@cgiar.org. The price of the CD-ROM is $75 (U.S.) to nonprofit organizations. Large discounts are available. If potential users think they deserve one, then they will probably qualify, and should contact the authors at p.jones@cgiar.org and p.thornton@cgiar.org. Registered users will receive upgrades at special low prices. Commercial concerns should contact the authors for software and customized applications. Users are requested to test the software, and any feedback provided to either of the authors will be gratefully received, especially with regard to locations where clear errors occur in the interpolated climate surfaces, or where the software simply does not work properly.Numerical Algorithms Group Ltd 1990; National Oceanographic and Atmospheric Administration. 1984; Texas A&M University Systems 1998


   ACKNOWLEDGMENTS
 
We are very grateful to John Lynam of the Rockefeller Foundation in Nairobi, Kenya, for financial support of this work. We thank Paul Wilkens of the International Fertilizer Development Center (IFDC) for programming the user interface.

Received for publication April 6, 1999.
   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 Program description
 Results
 Discussion
 REFERENCES
 





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via ISI Web of Science (16)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Jones, P. G.
Right arrow Articles by Thornton, P. K.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Jones, P. G.
Right arrow Articles by Thornton, P. K.
Agricola
Right arrow Articles by Jones, P. G.
Right arrow Articles by Thornton, P. K.
Related Collections
Right arrow Agroclimatology
Right arrow Other Models
Right arrow Software

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Crop Science Vadose Zone Journal
Journal of Natural Resources
and Life Sciences Education		 Soil Science Society of America Journal
Journal of Plant Registrations Journal of
Environmental Quality		 The Plant Genome