Integrating remote sensing and spatial statistics to model herbaceous biomass distribution in a tropical savanna

Modelling herbaceous biomass is critical for an improved understanding of wildlife feeding patterns and distribution as well as for the development of early warning systems for fire management. Most savannas in South Africa are characterized by complex stand structure and abundant vegetation species. This has prohibited accurate estimation of biomass in such environments. We investigated the possibility of improving biomass predictions in tropical savannas using cokriging. Individual bands and ratios computed from Moderate Resolution Imaging Spectroradiometer ( MODIS) imagery were correlated with field measurements of biomass covering the Kruger National Park, South Africa. The band that yielded the highest correlation with biomass was then used for further analysis using cokriging. Three variogram models were developed: one for the herbaceous biomass, one for the best MODIS band and a cross variogram between all pairs of variables involved in the estimation. The variogram models were then used in cokriging to predict biomass distribution over the whole study area. Results indicate that a combination of remotely sensed data with field biomass measurements through cokriging improves the estimation accuracy compared to ordinary kriging and stepwise linear regression. Given the high temporal resolution of the freely available MODIS imagery, the result is critical for the improved monitoring and management of wildlife habitats.