External-interface computer code has been written for the cotton growth model, Gossym, such that it can be operated from the ArcView GIS interface. Remote-sensing data have been incorporated as an estimator of plant height for feedback to the model. Management zones are delineated automatically based on available spatially variable data, and Gossym subsequently calculates outputs for each management zone, and uses current remote-sensing data in the calculations. This advanced Gossym model system also gathers local weather data automatically over the internet. Development and use of the model system are also described. Site-specific field data and remotely sensed images have been collected extensively on two agricultural fields in Mississippi from 1998 through 2003. Evaluation of yield prediction based on the Gossym model system indicated that current remote-sensing data can enhance accuracy. Details of experimentation and data analysis are presented.
If farmers could predict yield on a spatially variable basis, they could better understand risks and returns in applying costly inputs such as fertilizers, etc. To this end, several remotely sensed images of a cotton field were collected during the 2002 growing season, along with daily high and low temperatures. Image data were converted to normalized-difference vegetation index (NDVI), and temperature data were used to normalize NDVI changes over periods between image collections. Remote-sensing and weather data were overlaid in a geographic information system (GIS) with data from the field: topography, soil texture, and historical cotton yield. All these data were used to develop relationships with yield data collected at the end of the 2002 season. Stepwise regression was conducted at grid-cell sizes from 10 m square (100 m2) to 100 m square (10,000 m2) in 10-m increments. Relationships at each cell size were calculated with data available at the beginning of the season, at the first image date, at the second image date, and so on. Stepwise linear regression was used to select variables at each date that would constitute an appropriate model to predict yield. Results indicated that, at most dates, model accuracy was highest at the 100-m cell size. Remotely sensed data combined with weather data contributed much information to the models, particularly with data collected within 2.5 months of planting. The most appropriate model had an R2 value of 0.63, and its average prediction error was about 0.5 bale/ha (0.2 bale/ac, or roughly 100 lb/ac).