Skip to main content
ARS Home » Southeast Area » Florence, South Carolina » Coastal Plain Soil, Water and Plant Conservation Research » Research » Publications at this Location » Publication #73241


item Sadler, Edward

Submitted to: Agronomy Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/28/1997
Publication Date: N/A
Citation: N/A

Interpretive Summary: Knowing final crop yield is valuable to farmers, commodity interests, and the food industry, and prediction methods have therefore been the subject of much research. Computer simulation models use descriptions of the soils, the crop, and the weather to calculate amounts of sunlight, water, nutrients, and carbon dioxide that move between the environment and the crop. The ultimate result of the simulation is usually crop yield, but along the way, the model must know the temperature during the day. Usually, only daily maximums and minimums are recorded, but many models need input data every hour. Several methods exist that take daily extremes and calculate what happened in between, but most have simple mathematical shapes, which only suggest the daily pattern (i.e., straight lines between the extremes). A new method lets the weather data describe its own pattern, which is possible after adjusting for day-to-day changes in average temperature and season-to-season changes in daylength. With it, daily patterns can be calculated with at least as much accuracy as existing methods in almost all cases, and in many, with substantially more accuracy than by any other method. Building this method into existing crop models would require some changes, but the new procedure is simpler than most of the older ones. Models using the new procedure will no longer be subject to systematic errors during parts of the day.

Technical Abstract: Air temperature is a key driving factor in many crop growth models. Several such models require input of hourly temperature data, which must often be estimated from daily extremes. The objective of this work was to develop an empirical model that reconstructs the diurnal temperature curve from measured or generated daily extremes. Temperature was normalized to range from 0 at the minimum to 1 at the maximum for the day. Time was also normalized to reduce seasonal variation in time of the minimum temperature. Calibration of the model included developing the empirical cumulative distribution function of normalized temperature as a function of normalized time for a year's data, fitting a Beta distribution to the data, and evaluating the 50th percentile for each time. The resulting vectors of normalized time and temperature were used to generate diurnal patterns from daily extremes. The model was calibrated with one year's data for 11 sites, and tested for additional years at each site. For these 26 site-years, annual mean r^2 ranged from 0.67 to 0.87, with values highest in Arizona, intermediate in South Carolina, and lowest in Oregon. The performance of the TFIT model was better than or equal to that of the next-best model in 16 of 26 site-years. Normalizing both time and temperature produced patterns applicable at widely separate sites with minimal loss of predictive accuracy.