### Abstract

Three methods of deriving weather for use in a nitrogen decision support system with a weekly time-step are described and evaluated. The simplest of these is simply the mean (M) of the three weather variables: Rainfall (R), evapotranspiration (ET) and temperature (T). To represent the variability of weather, many sets of generated data are needed but this is not possible with mean values. Two methods of generating rainfall are described: a fully stochastic simulator (FS) and a method based on partitioning the distribution into sections (SM). Temperature, T, and evapotranspiration ET are represented in both generators by sinusoidal functions. The amount of R is modelled as an empirical distribution, rain persistence as a Markov chain. All three means of deriving weather were compared directly with actual weather from the historical record and in use with the SUNDIAL DSS. The mean values of R, ET and T were reproduced satisfactorily (r > 0.99) by both FS and SM, but variability less accurately (r > 0.80 for the standard deviations of T, r > 0.97 for R and ET). The mean values of several components of the nitrogen cycle simulated with SUNDIAL were generally reproduced well for both methods of weather generation, but less accurately for mean weather. For leaching, the root mean square errors were 2.6, 2.3 and 11.4 kg/ha for FS, SM and M, respectively. The sectioning method generally gave a poor estimate of variation, which was significantly underestimated for the majority of variables, in the case of leaching by a factor of three. Where variance is important, FS is preferred; weather data generated by this method may be used with confidence for risk assessments of denitrication and crop N uptake.

Original language | English |
---|---|

Pages (from-to) | 257-266 |

Number of pages | 10 |

Journal | Nutrient Cycling in Agroecosystems |

Volume | 73 |

DOIs | |

Publication status | Published - 2005 |

### Keywords

- Decision Support System
- nitrogen fertiliser
- stochastic model
- weather generator
- COMPUTER-SIMULATION
- DAILY PRECIPITATION
- SOIL
- CROP
- VARIABILITY
- WINTER

### Cite this

*Nutrient Cycling in Agroecosystems*,

*73*, 257-266. https://doi.org/10.1007/s10705-005-3031-3

**Weekly weather generation for a nitrogen turnover model.** / Dailey, A G ; Smith, J U ; Whitmore, A P .

Research output: Contribution to journal › Article

*Nutrient Cycling in Agroecosystems*, vol. 73, pp. 257-266. https://doi.org/10.1007/s10705-005-3031-3

}

TY - JOUR

T1 - Weekly weather generation for a nitrogen turnover model

AU - Dailey, A G

AU - Smith, J U

AU - Whitmore, A P

PY - 2005

Y1 - 2005

N2 - Three methods of deriving weather for use in a nitrogen decision support system with a weekly time-step are described and evaluated. The simplest of these is simply the mean (M) of the three weather variables: Rainfall (R), evapotranspiration (ET) and temperature (T). To represent the variability of weather, many sets of generated data are needed but this is not possible with mean values. Two methods of generating rainfall are described: a fully stochastic simulator (FS) and a method based on partitioning the distribution into sections (SM). Temperature, T, and evapotranspiration ET are represented in both generators by sinusoidal functions. The amount of R is modelled as an empirical distribution, rain persistence as a Markov chain. All three means of deriving weather were compared directly with actual weather from the historical record and in use with the SUNDIAL DSS. The mean values of R, ET and T were reproduced satisfactorily (r > 0.99) by both FS and SM, but variability less accurately (r > 0.80 for the standard deviations of T, r > 0.97 for R and ET). The mean values of several components of the nitrogen cycle simulated with SUNDIAL were generally reproduced well for both methods of weather generation, but less accurately for mean weather. For leaching, the root mean square errors were 2.6, 2.3 and 11.4 kg/ha for FS, SM and M, respectively. The sectioning method generally gave a poor estimate of variation, which was significantly underestimated for the majority of variables, in the case of leaching by a factor of three. Where variance is important, FS is preferred; weather data generated by this method may be used with confidence for risk assessments of denitrication and crop N uptake.

AB - Three methods of deriving weather for use in a nitrogen decision support system with a weekly time-step are described and evaluated. The simplest of these is simply the mean (M) of the three weather variables: Rainfall (R), evapotranspiration (ET) and temperature (T). To represent the variability of weather, many sets of generated data are needed but this is not possible with mean values. Two methods of generating rainfall are described: a fully stochastic simulator (FS) and a method based on partitioning the distribution into sections (SM). Temperature, T, and evapotranspiration ET are represented in both generators by sinusoidal functions. The amount of R is modelled as an empirical distribution, rain persistence as a Markov chain. All three means of deriving weather were compared directly with actual weather from the historical record and in use with the SUNDIAL DSS. The mean values of R, ET and T were reproduced satisfactorily (r > 0.99) by both FS and SM, but variability less accurately (r > 0.80 for the standard deviations of T, r > 0.97 for R and ET). The mean values of several components of the nitrogen cycle simulated with SUNDIAL were generally reproduced well for both methods of weather generation, but less accurately for mean weather. For leaching, the root mean square errors were 2.6, 2.3 and 11.4 kg/ha for FS, SM and M, respectively. The sectioning method generally gave a poor estimate of variation, which was significantly underestimated for the majority of variables, in the case of leaching by a factor of three. Where variance is important, FS is preferred; weather data generated by this method may be used with confidence for risk assessments of denitrication and crop N uptake.

KW - Decision Support System

KW - nitrogen fertiliser

KW - stochastic model

KW - weather generator

KW - COMPUTER-SIMULATION

KW - DAILY PRECIPITATION

KW - SOIL

KW - CROP

KW - VARIABILITY

KW - WINTER

U2 - 10.1007/s10705-005-3031-3

DO - 10.1007/s10705-005-3031-3

M3 - Article

VL - 73

SP - 257

EP - 266

JO - Nutrient Cycling in Agroecosystems

JF - Nutrient Cycling in Agroecosystems

SN - 1385-1314

ER -