Faculty of Biology, Ludwig Maximilian University of Munich, 80539 München, GERMANY.
Though the forests of the world are vital for all humans, their area is decreasing. While boreal forests even grow in size, tropical forests are severely threatened by deforestation. The elevated deforestation rates of the tropics have severe impact on the global climate as approximately 20% of the total human-induced greenhouse gas emissions stem from deforestation processes. Furthermore, the humid tropics are among the most species-rich ecosystems of the world and the proceeding deforestation severely threatens their high biodiversity. To observe and analyze tropical deforestation satellite based monitoring is mandatory and the only possibility due to the vast and often inaccessible study areas. In comparison to radar systems, optical sensors provide more detailed information to distinguish different types of vegetation. However, the principal limitation of optical imaging systems is that they cannot penetrate clouds, which are quite frequent in the humid tropics. Multitemporal compositing can be used to derive cloud-free mosaics but depending on the revisit time of the satellite not all sensors can be used for the purpose of monitoring larger areas in the humid tropics. Possible solution is the use of low or medium resolution systems, which have shorter orbit repeat cycles, thus acquiring imageries at a higher frequency. This PhD thesis deals with the applicability of different techniques to monitor the fast deforestation activities in the humid tropics, as well as a detailed analysis of the deforestation processes and their underlying causes. Main focus was put on the tropical forest ecosystems of Borneo, as Southeast Asia shows the highest deforestation rate of all humid tropics and Borneo, still having the largest remaining area of tropical rainforest in that realm, is representative for many processes leading to deforestation in the region.
In a first step the ability of multispectral change detection analysis was tested to quickly detect land cover changes. This study was carried out both in Central Africa and Borneo to investigate various satellite sensor systems such as Landsat, MODIS, MERIS and SPOT VEGETATION in different tropical regions with their prevailing land use patterns. Under ideal conditions such as no atmospheric disturbances the use of multispectral change detection analysis based on optical data is a promising approach to quickly monitor changes in the forest cover. Most important advantage of this technique is the omission of the classification step, thus providing much faster results. However, applying this system in the humid tropics implicated pivotal limitations. The frequent cloud coverage of the humid tropics prevents from monitoring larger areas as optical satellite data interfere with clouds or haze. Thus, cloud-free composites had to be used. Depending on the algorithm, these composites either show artifacts due to the selection of single pixels and thus cannot be used as input for multispectral change detection, or when applying compositing methods, which reduce the amount of artifacts considerably, the advantage of easily detecting changes are lost due to the higher number of single images and the longer period of data acquisition. Thus, owing to these limitations inherent to the humid tropics the concept of quickly detecting changes by applying multispectral change detection analysis in these areas is not feasible in an operational way – irrespectively in Central Africa or Borneo.
Therefore another approach was analyzed. The next studies showed that a post classification methodology based on cloud-free composites derived from an average composition method can be successfully used for land cover mapping in the humid tropics. This approach provides more detailed information about the land cover than the multispectral change detection analysis. Based on medium resolution MODIS surface reflectance data it was possible to classify the land cover of Borneo, discriminating 11 land cover classes, also including 6 different forest types. In 2002 about 57% of the land was covered with forests and between 2002 and 2005 the annual deforestation rate was 1.7%. The forest areas of the island are dominated by 74% of dipterocarp and more than 23% of the carbon rich peat swamp forests - the latter one showing an increased deforestation rate of 2.2%. The analysis of the spatial pattern of deforestation showed, that deforestation usually does not start in the middle of closed forests. Almost 98% of all deforestation occurred within a range of 5 km to the forest edges. Causes of deforestation are manifold, such as conversion to farmland, legal and illegal logging for timber supply and fire, which is often linked to deforestation because it is the cheapest means in land preparation. In fact 98% of all forest fires were detected in the same 5 km buffer zone, showing that fire is the major driver of forest degradation and deforestation. Fires inside closed forests, which were observed by active fire detecting sensors, indicate the existence of human access such as small-scale infrastructure, which cannot be detected using medium or low resolution surface reflectance data alone. 94% of all fires detected in a peat swamp forest in Central Kalimantan were located closer than 1 km to existing small-scale logging infrastructures. As fire is one of the major causes of deforestation, the spatial and temporal pattern of all fires detected between 1997 and 2006 was investigated. No single sensor is completely covering the investigation period, therefore several sensors for active fire detection such as AVHRR, ATSR and MODIS had to be combined to analyze the fires retrospectively. About 16.2 Mha, corresponding to 21% of the land surface of Borneo, have been affected by fire at least once and 6% more than one time. During El Niño years with their prolonged droughts the impact of the forest fires increased considerably, affecting about 1 Mha, which is more than 3 times larger than in non-El Niño years. Although ecosystem and land use are similar, there is a pronounced difference in fire occurrence across the different countries of Borneo. In normal years the percentage of the fire affected area in Kalimantan was almost 5 times larger than in the non-Indonesian part of Borneo. During El Niño driven droughts this percentage almost doubled in Indonesia, while the El Niño phenomenon seemed to have no influence on fire occurrence in Brunei and the Malaysian part of the island. This suggests, that El Niño related droughts are not the only cause of increased fire occurrence and do not necessarily lead to a higher number of fires.
The results of this PhD thesis both provide crucial information concerning monitoring techniques based on optical satellite data in the humid tropics and improve the understanding of the drivers of deforestation – information decisive to better protect the remaining rainforests and an essential contribution regarding global climate change.