Application of spatial statistics and crop models to agroclimatic zoning

for rice cultivation in North Korea


Jin I. Yun

Department of Agronomy / Institute of Life Science and Natural Resources Kyung Hee University, Suwon 449-701, KOREA





Agroclimatic zoning was done for rice culture in North Korea, where access to any information is extremely limited, based on a simulation experiment. CERES-rice, a rice growth simulation model, was tuned to accommodate agronomic characteristics of major North Korean rice cultivars based on field observations in South Korea. The model was run with randomly generated 30 sets of daily weather data (from planting to physiological maturity) for 183 counties in North Korea to simulate the growth and yield response to the interannual climate variation. Weather datasets for each county were prepared through 3 consecutive steps; spatial interpolation based on topoclimatological relationships, zonal summarization of grid cell values, and conversion of monthly climate data to daily weather data. Results were analyzed with respect to spatial and temporal variation in yield and maturity, and used to evaluate the suitability of each county for rice cultivation. Among the 183 counties in North Korea, 20 were classified as "fail" and 39 as "limited" after the maturity date evaluation. Suitability scores were assigned to the remaining 124 counties. The results may be utilized as decision aids for agrotechnology transfer to North Korea, for example, germplasm evaluation, resource allocation and crop calendar preparation. 



Key words: Spatial interpolation, Crop models, Climate, North Korea, Rice

            Agroclimatic zoning





Climate is a single most important factor that should be considered before agricultural technology transfer to geographically different regions. North Korea has been suffering from a nationwide famine due to the decline of food crop production in recent years. It is a consensus of many organizations providing famine aids to North Korea that agricultural technology transfer such as provision of seeds, fertilizers, machineries and management skill is more effective than supplying foods to help the people solve the famine trouble. Since little information is available on agricultural climate in North Korea, however, the effectiveness of the technology transfer seems to be questionable.

It is a challenging task for agricultural climatologists to figure out agroclimatic features of a region which is most isolated on the earth. There is no way to access climatological data of this 120,000 km2 country with a complex topography, except the 27 standard weather station reports via GTS(Global Telecommunications System). It is practically impossible to obtain crop status information except some estimates based on satellite remote sensing data. After consideration of the political and technical environments, we decided to make use of crop simulation modelling and spatial data management techniques. This approach is one of the very few choices that we can select from, and possibly the most efficient way of obtaining products with practical meaning.

Probably, the best way of agroclimatic zoning for a crop will be a long term cultivation of the crop, because crop growth and yield summarizes the climatological features of the land during the crop season. Practical alternatives to this ideal method include statistical analysis of climatic variables, calculation of agroclimatic indexs, and empirical expression of crop-climate relationships. Recently, crop models based on major physiological processes of plants have been extensively used to simulate the responses of legumes (Hoogenboom et al., 1992), maize (Kiniry et al., 1997), rice (Yajima, 1996) and other crops to environmental variability. If spatial data of soils, weather, and other input variables to the model are available, it is possible to figure out the spatial variation of the crop status over geographical areas, enabling the agroclimatic crop forecasting feasible (Rosenthal et al., 1998). With the weather data alone, we may estimate the spatial variation of the potential productivity and make use of this information in agroclimatic zoning.

Diversified local climate or climatic variation across geographic areas with a complex terrain feature like North Korea hinders utilizing the traditional macro-scale climate atlas in agricultural technology transfer. Although North Korea has 27 standard weather stations in operation, it is far less than enough to cover his complex land area exceeding 120,000 km2. Possibility for North Korea seems to be very low, even in the near future, to establish a meso-scale observation network for agriculture as in advanced regions (Nakai, 1990; Brock et al., 1995). It is a common practice especially for many regions with complex terrain features to produce surfaces of climatological variables by spatial interpolation of point observations. Digital elevation model(DEM) and Geographic information systems(GIS) technology have been playing major role in this respect (Seino, 1993; Daly et al., 1994). Climatological precipitation surface over North Korea was produced by using this technology and the elevation - precipitation relationship found in South Korea where a dense rainfall network exists (Yun, 2000). While it can be used to produce the extensive climate data for crop model input, interfacing simulation models with GIS provides a powerful tool for spatial modelling in diversified fields (Petersen et al., 1995).

This study was conducted to figure out the agoclimatic potential for rice production at 183 counties in North Korea based on the simulated crop response obtained by a crop model and simulated weather data.



Materials and Methods


1. Overview


Major rice cultivars currently grown in North Korea were planted at two experimental farms in South Korea during the 1995 to 1998 crop seasons. The observed growth and yield data were used to adjust the genetic coefficients of CERES-rice, a rice growth simulation model.

Daily weather data for each county necessary to run the crop model were obtained by the following 3 steps. Because historical climate data are not available for each county, regression models for monthly climatological temperature estimation were derived from a statistical procedure using monthly averages of 51 standard weather stations in South and North Korea (1981-1994) and their spatial attributes such as latitude, altitude, distance from the coast, sloping angle, and aspect-dependent field of view (openness). Selected models were applied to generate monthly temperature surfaces over the entire North Korean territory on a 1km by 1km grid spacing. Monthly precipitation surfaces were generated by applying aspect-dependent regression models derived from the South Korean rainfall network consisting of 277 observation stations. Daily solar radiation data for 27 North Korean stations were reproduced by applying an empirical relationship between the measured daily solar radiation and the meteorological variables on the same day found in South Korea. The reproduced data of 27 points were converted to monthly solar irradiance surfaces by the inverse distance squared weighted (IDSW) interpolation.

The grid cell values of monthly temperature, solar radiation, and precipitation were aggregated into corresponding 183 county values, since the county serves as a land unit for the growth simulation and for the agroclimatic zoning. Finally, we randomly generated daily maximum and minimum temperature, solar irradiance and precipitation data for 30 years from the monthly climatic data for each county.

Daily weather data were fed into the CERES-rice, tuned to have agronomic characteristics of major North Korean rice cultivars, to simulate the crop status at 183 counties in North Korea for 30 years. Results were analyzed with respect to spatial and temporal variation in yield and maturity, and used to score the suitability of the county for rice culture.


2. CERES-rice model


The CERES-rice model simulates the effects of atmosphere, soil, water, nutrients and management practices on the growth and development of rice (Godwin et al., 1992). Assuming all other factors constant, the model can be used to figure out the potential effect of daily weather on the rice growth. At least four weather variables must be prepared to run the model: daily minimum and maximum temperatures, solar irradiance, and precipitation. In order to apply this model to any region, the genetic coefficients relevant to the rice varieties adapted to that specific region. This can be done by adjusting 4 growth- and another 4 phenology- related parameters based on multiple years' field observations.

Field experiments were carried out at two experiment farms in South Korea during the 1995 to 1998 period in order to collect observation data for three major cultivars in North Korea, on the growth, phenological phase, yield components and yield. Rice seedlings at 35 days old were planted on May 20 each year with the row spacing of 0.3m by 0.2m, 5 plants per hill, and 120-120-130 kg ha-1 rates of N-P2O5-K2O.

The observed data were used to adjust the model parameters following Hunt et al.(1994). SB9 was found to be an extremely early maturing cultivar with the average heading date of July 12 under the experiment condition. Heading dates of AK72 and PY15 were around early to mid August, making them comparable with early to medium maturity cultivars in South Korea.


3. DEMs, derived grids and climate data


A digital elevation model with 30 arc second grid spacing was obtained from the United States Geological Survey and the portion of the Korean Peninsula was extracted and stored in an ARC/INFO (Release 7.2.1, ESRI, USA) grid with 1km by 1km cell spacing on Transverse Mercator projection (origins: 38N and 127E). Spatially averaged elevation (ELEV), slope aspect (ASPECT), distance from the nearest seashore (CODI), slope angle (SLOPE), and openness (shade index) toward 8 azimuthal directions (OPEN_N, OPEN_S, . . . , OPEN_W) were derived from the original grid (DEM1) and the smoothed grids with various grid spacings (DEM3, DEM5, . . . , DEM31). In addition, the differences between the actual elevation of weather stations, which are explained below, and the DEMs were stored as DEV grids.

Monthly averages of the daily maximum and minimum temperatures, precipitation, and the number of days with measurable precipitation were calculated from the 14 years (1981-1994) daily data at 27 North Korean standard weather stations. The same procedure was applied to 24 standard weather stations in South Korea to obtain the monthly climate data, making 51 point observations available for subsequent analyses. In addition, daily precipitation data for 10 years (1986-1995) were collected from 277 rain gauge stations in South Korea (Fig. 1).

Figure 1  Geographical location of the study area and the 51 standard weather stations (solid circle) in North and South Korea and the 277 rain gauge stations in South Korea (empty circle).

4. Generation of climatological surfaces


4.1  Temperature

Topographical and geographical variables of the grid cells containing the 51 observation points were extracted from the DEM derived grids. A correlation matrix was prepared for these variables and the monthly maximum and minimum temperature to select candidate variables of the regression models for predicting temperature at grid cells with no observation stations. A multiple regression procedure using SAS/REG (SAS Institute, USA) with STEPWISE option (stay level = 0.15) was performed with the selected topo- and geographical variables as the independent, and the 51 climatological temperature data as the dependent variable for each month. Obtained models were applied to the DEM derived grids to produce monthly temperature estimates for each grid cell.


4.2  Precipitation

Two hundred and seventy seven rain gauge stations of South Korea were classified into 8 different groups depending on the aspect of the region they are located. Monthly precipitation averaged over the 10 year period was regressed to topographical variables of the station locations. A "trend precipitation" for each gauge station was extracted from the precipitation surface interpolated from the monthly precipitation data of 24 standard stations of the Korea Meteorological Administration and used as a substitute for y-axis intercept of the regression equation. These regression models were applied to the corresponding regions of North Korea, which were identified by slope orientation, to obtain monthly precipitation surface for the aspect regions. "Trend precipitation" from the 10 year data of 27 North Korean standard stations was also used in the model calculation. Output grids for each aspect region were mosaicked to form the monthly and annual precipitation surface with a 1km1km resolution for the entire territory of North Korea.

The number of days with measurable precipitation at 27 North Korean stations were interpolated by IDSW to obtain the monthly estimates over the entire region. This information is necessary to generate daily precipitation from the monthly climatic normals by using 'weather generators'.


4.3  Solar radiation

There is no historical data for solar radiation and sunshine duration over North Korea. We had to have restore monthly solar irradiance data at 27 standard weather stations. In South Korea, there are 20 locations where daily solar radiation data are available in addition to the standard meteorological observations. We decided to utilize an empirical relationship between daily solar radiation and the other meteorological variables in South Korea, as was done in the precipitation case, to restore the North Korean data. Daily data were collected for solar radiation, relative humidity and cloud amount at 20 standard weather stations in South Korea for the 1984 - 1997 period. We also calculated the daylength and extra-terrestrial solar radiation for these locations, and extracted the openness value (shade index) toward south direction for each location from the DEM derived grids. A regression equation was formulated from these data to estimate daily solar irradiance of any locations without measurement facilities. The equation is given by:


[Solar Radiation, MJ m-2 day-1]

  = 0.344 + 0.4756 [Extraterrestrial Solar Irradiance]

    + 0.0299 [Openness toward south, 0 - 255]

    - 1.307 [Cloud amount, 0-10] - 0.01 [Relative humidity, %]   

                                    (r2 = 0.92, RMSE = 0.95 MJ m-2 day-1)


The equation was used to simulate daily solar irradiance at the 27 North Korean stations during the 1981-1994 period. The monthly climatological solar irradiance at corresponding stations were obtained from the restored data, and were interpolated by IDSW scheme to generate the climatological solar radiation surface over North Korea for each month.


5. Simulation of daily weather


The climatological surfaces generated by the above mentioned method consist of 12 monthly grids with a 1km by 1km spatial resolution. Major climatology of the whole country can be viewed as 120,000 grid cell values. We aggregated these cell values into 183 county averages by a zonal summarization, because it was not practical to run the crop model 120,000 times on our computing facilities. That is, the land unit for crop simulation or agroclimatic zoning for North Korea was assumed to be the county level (Fig. 2). A county boundary map digitized into a polygon featured vector format was overlaid on the climate grids and the grid cells falling within each polygon were summarized to get the spatial mean and standard deviation. Average number of cells consisting of a county was 662 ranging from 45 to 2,197. Calculated zonal statistics were recorded in the attribute tables (database file) of the county map for further applications. We assumed that the zonal average value is representative of the rice fields in each county. But the temperature values were increased by one standard deviation, considering the relative location of rice fields within a county.

The CERES-rice needs daily weather values as input data. The climatological monthly data have to be transformed to daily values of multiple years to represent the interannual variation of the climate. Stochastic weather generators have been used for this purpose (Richardson and Wright, 1984; Geng et al., 1988; Wallis, 1996). We simulated daily weather at 183 counties for 30 years based on the monthly climate data by using the method of Pickering et al.(1994).




Figure 2.  County map coverage (polygon vector) overlaid on the 1km1km digital elevation model (raster grid) of North Korea. Each number on the polygon identifies crop simulation zone.


6.  Growth simulation and data analysis


The CERES-rice model was driven by the simulated 30 years' daily weather data for 183 counties and the simulation results were obtained for the extreme early-, the early-, and the medium- maturing cultivars, respectively. Soil condition was fixed for all the counties to have 1.3m soil depth with the texture of sandy loam. The same management practices were applied to each county, i.e., 35 days old seedlings, planting density of 25 plants/m2, transplanting on May 20 each year, automatic nitrogen application, and no irrigation. The output includes heading date, physiological maturity, top dry matter weights, grain yields, and so on.

Among the simulated growth, development and yield characteristics, we selected 3 variables as the criteria for agroclimatic zoning: the interannual variation in physiological maturity, the 30 year average grain yield and the interannual variation in the grain yields. Since the selected variables show a continuous distribution, the output values were grouped into 4 classes. Variation in physiological maturity was categorized as "stable", "quasi-stable", "variable", and "unstable", depending on the standard deviation values of "less than 6 days", "7-9 days", "10-15 days", and "over 16 days", respectively. The simulated yields were grouped into "high", "medium", "low", and "poor" categories, depending on the 30 year average grain yields in tons/ha of "over 6.5", "6.2-6.4", "5.9-6.1", and "less than 5.9", respectively. The interannual variation in the grain yield was expressed as the coefficients of variation (CV), that is, the ratio of the standard deviation to the average yield. They were assigned to the same category as the maturity variation, depending on the CV values of "0.05-0.11", "0.12-0.18", "0.19-0.25", and "over 0.26", respectively.

To each category of the 3 criteria was given a "suitability score" of 3, 2, 1, and 0, respectively, to express the integrated performance of the crop model at each county. Hence, any county making total of 9 points should be considered as the best place for rice cultivation in regard of the climatological condition alone.



Results and Discussion


1. Monthly climatological normals


Selected regression models for the monthly averages of daily maximum and minimum temperatures consist of the variables like latitude, distance from the coastline, elevation, slope and openness. All the models except those for daily maximum in summer months showed higher than 0.9 for the coefficients of determination (R2), and their RMSE ranged from 0.4 to 1.6 (Table 1).





Table 1.  Regression coefficients and RMSE of the climatological temperature estimation model.



Minimum Temperature


  Month      Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec 



[INTERCEPT]  50.7  43.3  58.2  17.7  19.1  36.6  41.4  36.7  44.2  41.7  44.1  47.6

[LATITUDE]   -1.52 -1.27 -0.95 -0.75 -0.58 -0.54 -0.54 -0.56 -0.75 -0.86 -1.10 -1.35

[ELEV3]/100  -0.95       -1.06 -0.80

[ELEV5]/100                                            -0.82

[ELEV7]/100        -1.03             -0.81 -0.87 -0.82       -0.89 -0.89 -0.86 -0.92

[SLOPE3]      0.64  0.48  0.23                                            0.45  0.58

[SLOPE7]                             -0.31                                     

[CODI]/100   -4.01 -2.51 -0.77              1.01  0.99       -0.89 -1.98 -2.25 -3.39

[OPEN_E]                        0.09  0.08              0.03


[OPEN_W]                 -0.12



  RMSE(mm)    1.6   1.2   0.8   0.9   0.7   0.6   0.5   0.4   0.8   1.2   1.2   1.4


  R-SQUARE    0.93  0.95  0.96  0.91  0.94  0.94  0.95  0.97  0.95  0.91  0.93  0.93


Maximum Temperature


  Month      Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec 



[INTERCEPT]  53.6  60.7  47.7  76.7  79.7  10.5  19.7  19.8  63.6 100.8  71.0  58.4  

[LATITUDE]   -1.37 -1.15 -1.01 -0.87 -0.65 -0.73 -0.61 -0.54 -0.54 -0.81 -1.17 -1.41

[ELEV3]/100  -0.65 -0.73 -0.80                                           -0.73 -0.64

[ELEV5]/100                    -0.73 -0.70       -0.60     

[ELEV7]/100        -1.03             -     -0.61       -0.68 -0.78 -0.74           

[SLOPE3]      0.43                                                              0.41


[CODI]/100   -1.72        1.56  2.89  3.24  3.97  3.03  2.27  1.32  0.53 -0.82 -1.94

[OPEN_E]                                                     -0.10 -0.14

[OPEN_S]                       -0.15 -0.19

[OPEN_W]           -0.07                                           -0.14 -0.08

[OPEN_NW]                                   0.23  0.17  0.16 


  RMSE(mm)    1.3   0.9   0.7   1.0   1.5   1.6   1.0   0.8   0.6   0.4   0.8   1.3


  R-SQUARE    0.95  0.94  0.95  0.88  0.81  0.68  0.80  0.86  0.94  0.98  0.95  0.92






These models were applied to more than 120,000 grid cells comprising the North Korean land area to obtain the temperature surfaces. Figure 3 is the average temperature of the July - September period, which is critical to rice growth in Korea, produced from the relevant temperature grids.




Figure 3.  Seasonal mean temperature pattern during July to September period in the climatological normal year estimated from the model calculation.






The restored solar radiation data at 27 locations in North Korea were used to produce the monthly climatological solar radiation surfaces by the IDSW interpolation. Figure 4 is the annual solar irradiance surface, which is the accumulation of monthly values.




Figure 4.  Annual solar irradiance pattern over North Korea estimated by interpolation of the restored data at 27 weather stations.










Figure 5 is the estimated number of rainy days interpolated by the IDSW interpolation.


Figuge 5.  Spatial variation of the number of days with measurable precipitation produced by an IDW interpolation of 27 station data.



Among the potential precipitation controls, elevation related variables were most frequently selected in the precipitation model (77 out of 96 models). About half of these were the elevation of the site relative to the smoothed elevation surface (DEV). The openness of the site was selected in 53 models showing the importance of interaction between the sloping aspect and the moving direction of weather systems. "Trend precipitation" was added to all the models regardless of the statistical significance. Forty six among the 96 models showed a significance at 5% level for their coefficients of determination (Table 2). Our results are comparable with Nalder and Wein (1998), where 55% of the regression models developed for the northern Canada showed a statistical significance with the average R2 of 0.43.

Table 2.  Coefficients of determination(R2) for the precipitation - topography regression models.


    Aspect          N      NE      E      SE       S      SW      W      NW 

Month   (# OBS.)    5      15     24      35      33      57      71     31


  January         .99*    .94    .55*    .26*    .19*     .08     .28    .43

  February        .99*    .68*    .64     .52     .28**   .25**   .55    .54*

  March           .99     .87    .55     .51     .56**   .30*    .20*   .16*

  April           .95**    .22    .77*    .88*    .85**    .74*    .34*   .77

  May             .98     .79    .30     .69*    .45*     .22    .16    .10

  June            .60     .83    .50**    .62*    .66     .53**   .32**  .59 

  July            .57     .96    .66     .76*    .55**    .68     .78**  .39

  August          .94     .65    .43**    .25*    .22     .27*    .48**  .41**

  September        -      .87    .73     .23     .37     .59     .49    .61 

  October         .99*    .98    .51     .40*     -       .33*    .19    .41

  November        .99     .64*   .14     .64*     .50      .35**   .54*   .45*

  December        .82*    .90*   .41*     .08     .23      .30**   .38    .28*

  Annual          .76     .89    .52     .47**    .41*     .50*    .58    .38


* Significant at 0.05 level

** Significant at 0.01 level

- No variable selected at 0.15 stay level during the STEPWISE procedure



We may expect that these models, which consist of only the "trend precipitation" and the topographical variables, can explain about half of the total variation in precipitation across any regions in North Korea, assuming the similarity in topographical features between South and North Korea. Spatially averaged annual precipitation of North Korea is 938mm with the standard deviation of 246mm according to the results. Figure 6 shows the spatial variation of the April to July rainfall total, which is critical to the timely transplanting of rice seedlings in Korea (Fig. 6).





Figure 6.  A sample precipitation map of North Korea showing the spatial distribution of April to June rainfall sum, which is critical to the timely transplanting of rice seedlings. Insert is an enlarged view of the rectangle area.


2. Crop performance


When the CERES-rice model was run with the genetic coefficients of SB9, the extremely early maturing rice variety in North Korea, the simulation was successful in all the counties except in the 20 northeastern counties. "Successful" means that the rice crop reached the physiological maturity by the end of the crop season in all 30 years. Thermal condition during the crop season (below 20C average during the summer) prevents even the extremely early maturing cultivar to finish the life cycle at least once in 30 years. Many of the failed counties are located in the Kaema Plateau, where the mean elevation is around 1,500m above the mean seal level.

The model failed at 59 and 63 counties, respectively, when the genetic coefficients were replaced by those of AK72, the early maturing cultivar, and PY15, the medium maturing cultivar. The additional counties failed by these cultivars appear as an extended region surrounding the 20 counties (Fig. 7). We may conclude that rice cultivation is absolutely impossible in the 20 counties and may be conditionally possible in the additional 39 to 43 counties, depending on the cultivar selection and management practices.





Figure 7.  North Korean counties classified into the regions of limited cultivation possible (gray) and those of cultivation impossible (black) based on 30 year growth simulation by CERES-rice with genetic coefficients of SB9 and AK72.





When we calculated the spatial statistics for the 124 counties successful in growing AK72, average dates for 30 years of the physiological maturity ranged from September 8 to October 8, depending on the county (Table 3). When the average variation within a county is expressed as the standard deviation, it shows about 10 days. This implies that we might expect more than 10 days early or late harvests even in the same county, depending on the interannual climate variation.



Table 3.  Spatial statistics for growth simulation results of an early to medium maturity cultivar (AK72) obtained by feeding randomly generated daily weather data for 30 years. No irrigation and an automatic nitrogen fertilizer application assumed.



                     Min.  Max.   Avg.    S.D. 



Anthesis      Mean  209.6  222.4  214.6   2.9 


(day of year) S.D.    2.1    4.7    3.4   0.5 


Maturity      Mean  251.0  280.7  260.2   6.2 


(day of year) S.D.    4.2   63.7    9.9  11.0 


Tops          Mean  11718  15397  13464   786   


(kg/ha)       S.D.    723   3646   1800   291 


Seed          Mean   5428   6913   6339   291 


(kg/ha)       S.D.    348   2261    971   357   


Rainfall      Mean    327    952    591   131 


(mm)          S.D.    106    324    190    41   


Evapo-        Mean    288    402    317    15 


(mm)          S.D.     16    121     27    11 






We assigned a suitability score, which is the point sum of the model performance in 3 decision criteria, to the 124 counties (Fig. 8). Highest scores are found in the southwestern region close to the Yellow Sea. We found even nearest two counties show significantly different scores. This kind of spatial precision cannot be expected from any other zoning products based on macro scale climate data. For the accuracy, unfortunately, we can neither judge the goodness of this zoning method, nor validate the performance score due to the inability to access necessary informations from North Korea.


Figure 8.  Overall performance score for rice cultivation in North Korea. The score is sum of three individual scores representing the annual variation in maturity date, grain yield, and the annual variation in grain yield, respectively.




In this study, we sought an operational framework for agroclimatic zoning in North Korea by interfacing crop simulation modelling and GIS technologies. Though we applied this framework to rice crop only, the framework may be extended to other major crops in North Korea such as corn and potato, given relevant simulation models. The only problem at present is that there is no way to get the necessary information to validate the results.

While the conventional criteria for agroclimatic zoning such as agroclimatic indexs evaluate only a part of the climatic resources available, crop simulation evaluates the integrated effect of all the model imbedded elements. In addition, we can control the spatial resolution of the evaluation unit. Although the current study was conducted on the county level, it is possible to carry out the same experiment on a finer scale, because the climatological surfaces were prepared on the 1km by 1km resolution. The feature of grid cell based climatological surfaces may also facilitate utilization of the satellite remote sensing data.





This work was supported by grant No. 981-0601-003-2 from the Basic Research Program of the Korea Science and Engineering Foundation (KOSEF).





Brock, F. V., K. C. Crawford, R. L. Elliott, G. W. Cuperus, S. J. Stadler, H. L. Johnson, and M. D. Eilts, 1995: The Oklahoma Mesonet : A technical review. Journal of Atmospheric and Oceanic Technology, 12, 5-19

Daly, C., R. P. Neilson, and D. L. Phillips, 1994: A statistical-topographic model for mapping climatological precipitation over mountainous terrain. Journal of Applied Meteorology, 33, 140-158.

Geng, S., J. S. Auburn, E. Brandsetter, and B. Li, 1988: A program to simulate meteorological variables: Documentation for SIMMETEO. Agronomy Progress Report No. 204. Dept. of Agronomy and Range Science., Univ. of California, Davis, CA.

Godwin, D., U. Sinȣ, J. T. Ritchie, and E. C. Alocilja, 1992: A User's Guide to CERES-Rice. International Fertilizer Development Center, Muscle Shoals, AL, USA.

Hoogenboom, G. J., J. W. Jones, anf K. J. Boote, 1992: Modelling growth, development and yield of grain legumes using SOYGRO, PNUTGRO, and BEANGRO:  a review.  Transactions of American Society for Agricultural Engineers, 35, 2043-2056.

Hunt, L. A., S. Pararajasingham, J. W. Jones, G. Hoogenboom, D. T. Imamura, R. M. Ogoshi, 1993: GENCALC: Software to facilitate the use of crop models for analyzing field experiments. Agronomy Journal, 85, 1090-1094.

Kiniry, J. R., J. R. Williams, R. L. Vanderlip, J. D. Atwood, D. C. Reicosky, J. Mulliken, W. J. Cox, H. J. Mascani, Jr., S. E. Hollinger, and W. J. Wiebold, 1997: Evaluation of two maize models for nine U.S. locations. Agronomy Journal, 89, 421-426.

Nakai, K., 1990: Japanese system of the meteorological information service to user communities including education and training. In A. Price-Budgen(ed.) Using Meteorological Information and Products. Ellis Horwood, UK, 257-274.

Nalder, I. A. and R. W. Wein, 1998: Spatial interpolation of climatic normals : test of a new method in the Canadian boreal forest. Agricultural and Forest Meteorology 92, 211-225.

Petersen, G. W., J. C. Bell, K. McSweeney, G. A. Nielsen, and P. C. Robert, 1995: Geographic information systems in agronomy. Advances in Agronomy 55, 67-111.

Pickering, N. B., J. W. Hansen, J. W. Jones, H. Chan, and D. Godwin, 1994: WeatherMan: a utility for managing and generating daily weather data. Agronomy Journal, 86, 332-337.

Richardson, C. W. and D. A. Wright, 1984: WGEN: A Model for Generating Daily Weather Variables. USDA-ARS, ARS-8, Washington, DC.

Rosenthal, W. D., G. L. Hammer, D. Butler, 1998: Predicting regional grain sorghum production in Australia using spatial data and crop simulation modelling. Agricultural and Forest Meteorology, 91, 263-274.

Seino, H., 1993: An estimation of distribution of meteorological elements using GIS and AMeDAS data. J. Agricultural Meteorology(Japan), 480, 379-383.

Wallis, T. W. R. and J. F. Griffiths, 1996: Simulated meteorological input for agricultural models. In Preprints of 22nd Conference on Agricultural and Forest Meteoroloy (Jan. 28-Feb. 2, 1996, Atlanta, Georgia), American Meteorological Society, 358-361.

Yajima, M., 1996: Monitoring and forecasting of rice growth and development using crop-weather model. In: R. Ishii and T. Horie (eds.), Crop Research in Asia: Achievements and Perspective. Asian Crop Science Association, 280-285.

Yun, J. I., 2000: Estimation of climatological precipitation of North Korea by using a spatial interpolation scheme. Korean Journal of Agricultural and Forest Meteorology, 2, 16-23. (In Korean with English summary)