| Abstract: |
| Assessments of change in C-stock change have a uncertainty in the plot level data used, as well as in the upscaling process. In the upscaling, satellite image analysis can lead to uncertainties through classification error and inappropriate resolution, among others. Classification error and resolution are often not independent from each other. Three different resolutions in upscaling are: information, spatial and temporal. Information resolution refers to the level of detail of information to be extracted, interpreted and separated by image classification. A land cover map of forest and non-forest only has a low information resolution. Spatial resolution is determined by the size of earth surface represented by one pixel and temporal resolution is the frequency of assessment in one particular period of time.
This study looks at how information and temporal resolutions lead to uncertainties in the upscaling process across Indonesian provinces of different forest transition stages. Due to different drivers of land use/cover changes, spatio-temporal patterns and diversity of land use types are different across forest transition stages. At the earliest stage, when undisturbed forest is the dominant land use/cover, both spatial and temporal variation are low, and land uses are not diverse. At latter stages, spatial and temporal variation increases and land use types are more diverse, until relatively stable stage is reached, when temporal variation is low, spatial variation is highest and land uses are most diverse.
We did a wall-to-wall analysis of East Kalimantan, Jambi and Lampung provinces that represent gradient in forest transition (FT) stages in Indonesia, from the earliest to the most advanced during the period of 1990-2005. In East Kalimantan, land use/cover changes are dominated by logging (area of undisturbed forest declined from 73.8% to 51.7% and area of logged over forest increases from 13.7% to 27.7 %) while in Jambi most marked land use/cover changes are forest conversion, mostly to oil palm and rubber (forest area declined from 54% to 34%; rubber area is now larger than forest area; and oil palm increased from 3% to 11%). Lampung is in the most advanced forest transition (FT) stage among the three, is almost stabilized in terms of land use/cover (forest area declined from 14% to 8% under national park and coffee plantation is extensive).
Three levels of information resolution are explored: (I) forest and non-forest, (II) forest is classified into forest on mineral soil, swamp and mangrove, and (III) forest classes of level II are further classified based on tree densities due to logging. Within each level, we introduce sub-level, in which we further separate non-forest classes: (A) tree-based, non-tree based and non-vegetation systems, (B) types of tree-based, non-tree based and non-vegetation systems.
The changes in carbon stock are estimated under different information resolution levels. In early FT (East Kalimantan), there is a big jump in emission estimates from level II to level III. Failing to take degradation into account results in a huge underestimation of carbon emission (estimation from level I and II are about a third of that from level III). For intermediate stage of FT (Jambi) and advanced stage of FT (Lampung), carbon emission estimates are more sensitive to sub-level diff
erentiation, i.e., finer classification in non-forest classes.
Another factor to be considered is the classification error which often correlates with information resolution; the higher information resolution to be separated, the higher the classification error is. With the hierarchical, object-based classification we use in producing the land cover maps, the decrease in the classification accuracy is much less marked compared to those in carbon emissions (90% classification accuracy in average under level I to 80% in average under level III with the finest sub-level).
We further explore the sensitivity of carbon emission estimates to temporal resolutions. We compare the emission estimates resulted from two time series (1990 and 2005) and those from three time series (period I: 1990 to 2000 and period II: 2000 to 2005). In the most advanced FT (Lampung), annual emissions from two time series, period I and period II are very similar. Within this stabilized systems, a frequent monitoring is not necessary. In early FT (East Kalimantan), more recent annual emission is higher than the total during the study period, while in intermediate FT (Jambi), more recent annual emission is lower.
This trend difference is important in terms of deciding reference period and baseline in REDD mechanism. As an illustration, we set a certain level of annual emission in the next 5 years in the two provinces. Within this illustration, if we consider the changes in the rate of increase in annual emissions, East Kalimantan shows positive reduction in annual emission, while Jambi shows negative reduction in annual emission, whilst annual emission still increases in East Kalimantan and decreases in Jambi. Splitting the study period into two therefore can pick up trends of emissions under Business As Usual (BAU) which otherwise is missed and results in overestimation of reduced emission from a REDD scheme for Jambi and underestimation for East Kalimantan. For intermediate and early FT, higher temporal resolution matters.
Selecting resolution for carbon stock change assessment should consider spatio-temporal variation and the diversity of land use types. FT stages explain characterize these patterns through driver-specific processes. Therefore FT framework is useful in terms of developing nested REDD mechanism from sub-national to national level in addressing technical and non-technical issues such as baseline and reference period negotiations. |
|
