Estimating Missing Weather Data for Agricultural Simulations Using Group Method of Data Handling

Abstract

Contacting weather stations via modems to obtain weather data for crop simulations has become a common practice. Users sometimes encounter gaps in these data, and techniques are needed to estimate weather variables for days when the data are absent. The authors hypothesized that such estimations can be made using data from before and after the day with no data. Dependencies of weather variables of a particular day on weather variables from several days before and after could be very complex. To find and to express these dependencies, group method of data handling (GMDH), which is a tool for modeling complex ?input?output? relationships by building hierarchical polynomial regression networks, was used. Data on daily solar radiation, maximum and minimum temperatures, and wind runs collected daily in Stoneville, Mississippi, during May?September of 1982?92 were used. Fourteen-hundred sequential 7-day datasets from the database were extracted. For each dataset, the authors assumed that weather variables on the fourth day were unknown and had to be found from the weather variables of days 1, 2, 3, 5, 6, and 7. Seventy-five percent of these data were used to find the hierarchical polynomial regression, and 25% were used to evaluate it. Correlation coefficients between calculated and actual parameters were similar for training and evaluation datasets. Coefficients of determination (R2) were about 0.88 for minimum temperature, 0.80 for maximum temperature, and 0.80 for wind run. Accuracy of the solar radiation and precipitation estimates was lower, and R2 was about 0.2?0.3 but improved to 0.5?0.6 for the training dataset and 0.3 for the validation dataset for both variables when an additional indicator variable that shows the presence or absence of rain was included. The next day after the day with missing data gave the most essential information. Increasing the number of missing days resulted in gradual deterioration of the accuracy for all variables but wind run. GMDH can be a useful tool for filling gaps in weather data from weather stations installed in the field.