While deforestation in the tropics is a major source of atmospheric CO2 so the regeneration of these forests is a major sink of CO2. To determine the magnitude of this sink and assess the carbon balance of tropical forests at regional to global scales an accounting model is required that multiplies carbon flux by the area of each forest regeneration stage. The only feasible tool with which to estimate the area of forest regenerative stages at regional to global scales is remote sensing from spaceborne sensors. The role of remote sensing in studies of tropical deforestation and forest regeneration is reviewed and results from a study focused on tropical forests in Ghana, West Africa, are presented. Emphasis is placed on tropical forest regeneration, which may be characterized typically by increases in leaf biomass, wood biomass and canopy roughness. Remotely sensed red and near infrared radiation can be used to estimate leaf biomass; microwave backscatter can be used to estimate leaf biomass, wood biomass and canopy roughness and multi-angle measurements of radiation can be used to estimate canopy roughness. This paper focused attention on three limitations with the most widely used of these approaches, the use of reflected red and near infrared radiation: (i) the availability of satellite sensor data for tropical forests; (ii) the relationships between remotely sensed data in red and near infrared wavelengths and forest measures that relate to leaf biomass; and (iii) the feasibility of estimating the coverage of forest within a pixel for accurate estimation of forest area. It was found that the probability of obtaining a cloud-free image of Ghana was small, particularly in the wetter months. From the available data, however, some significant relationships between remotely sensed radiance and tree density, a measure related to leaf biomass, were observed. The correlation observed between AVHRR NDVI and tree density suggests that potentially these data could be used to estimate three broad classes of tree density at the regional scale. It was also found that the accuracy of areal estimates of forest coverage could be increased through the use of a spectral mixture model.
