Abstract:
Wheat consumption has become more widespread and is increasing in South Africa especially in the urban areas. The wheat industry contributes four billion rands to the gross value of agriculture and is a source of employment to approximately 28 000 people. Wheat yield forecasting is crucial in planning for imports and exports depending on the expected yields and wheat health monitoring is important in minimizing crop losses. However, current crop surveying techniques used in South Africa rely on manual field surveys and aerial surveys, which are costly and not timely (after harvest). This research focuses on wheat health monitoring and wheat yield prediction using remote sensing, which is a cost effective, reliable and time saving alternative to manual surveys. Hence, the research objectives were: (i) to identify remotely sensed spectral indices that comprehensively describe wheat health status. (ii) Develop an Normalized Difference Vegetation Index (NDVI) based wheat yield forecasting model and (iii) to evaluate the impact of selected agrometeorological parameters on the NDVI based forecasting model. Landsat 8 images were used for determining spectral indices suitable for wheat health monitoring by relating the spectral indices to the land surface temperature. Results show that the Normalized Difference Water Index (R2 between 0.65 and 0.89) and NDVI (R2 between 0.36 and 0.62) were the most suitable indices for wheat health status monitoring. Whereas, the Normalized Difference Moisture Index (R2 between 0.53 and 0.79) and the Green Normalized Difference Vegetation Index (R2 between 0.28 and 0.41) were found to be less suitable for wheat health monitoring. Moderate Resolution Spectroradiometer (MODIS) derived NDVI for fourteen years was used to build and test a wheat yield forecasting model. The model was significant with an R2 value of 0.73, a p-value of 0.00161 and an RMSE of 0.41 tons ha-1. The study established that the period 30 days before harvest during the anthesis growth stage, is the best period to use the linear regression model for wheat yield forecasting. Satellite derived agrometeorological parameters such as: soil moisture, evapotranspiration and land surface temperature were added to the NDVI based model to form a multi-linear regression model. The addition of these parameters to the NDVI model improved it from an R2 of 0.73 to an R2 of 0.82. Through the use of a correlation matrix, the NDVI (r=0.88) and evapotranspiration (r=0.58) were highly correlated to wheat yield as compared to soil moisture (r=0.27) and land surface temperature (r=-0.02). This research provided evidence that remote sensing can be used at acceptable levels of accuracy for wheat monitoring and wheat yield predictions compared to manual field surveys which are costly and time consuming.
