|dc.contributor.advisor||Klug, John R.||
|dc.contributor.advisor||Greenfield, Peter L.||
|dc.contributor.advisor||Dicks, Harvey M.||
|dc.creator||Brüggemann, Edgar Alfred.||
|dc.description||Thesis (M.Sc.Agric.)-University of Natal, Pietermaritzburg, 2000.||en
|dc.description.abstract||Commercial sugarcane records for 19 seasons from 146 fields were obtained from selected
estates in the KwaZulu-Natal Midlands. The estates, located at Kranskop, Umvoti and
Richmond, are representative of the higher-potential rainfed sugarcane production region of the
Midlands sugarcane belt. Extensive editing and cleaning of the agronomic records was required.
Regression models were developed to determine which parameters of the field records were
consistently associated with sugarcane yield (TCH) and could be used for yield predictions.
Depending on the predictor variables selected, the best models based on 535 crop cycles
accounted for 55% and 43% ofthe observed yield variation respectively. Linear regression was
an appropriate analytical technique since the assumptions of normality and homoscedasticity
were upheld and multicollinearity was not a problem in the models. The models were validated
using an independent data set of 47 observations and satisfactory performances were confirmed.
The 95% confidence limits of yield predictions for the population mean lie within 10% of long-term
mean yields. These predictions could be useful for estate resource allocation and harvest
Key physical field attributes associated with sugarcane yield were locality, aspect, altitude, soil
type and effective rooting depth. Season and rainfall were important climatic variables. Ofthe
factors influenced by management, sugarcane variety, plant / ratoon status, crop cycle, N and K
nutrition and the topsoil Ca:Mg ratio were important yield predictors, depending on the equation
used. The relative importance of individual predictors varies with the specific combination of
resources for a particular observation.
The models were linked to a geographic information system to demonstrate an application ofthe
models for yield prediction in response to spatial variables. These predictions showed that the
models could be used at a general scale within estates to identify areas of differing production
potential. Reliable yield predictions could not be made for individual fields and within-field
resource variations could not be adequately accounted for.||en
|dc.subject||Crops and soils--Computer simulation.||en
|dc.title||Modelling the production potential of land for sugarcane in the KwaZulu-Natal Midlands sugarcane belt.||en