Integrating field and optical RapidEye data for above-ground biomass estimation: A study in the tropical peat-swamp forest of Sebangau, Central Kalimantan, Indonesia

he global impacts of greenhouse gas emissions from deforestation and forest degradation on climate change have increased international concern. This concern led to an agreement in the forestry sector to reduce carbon emissions from deforestation and forest degradation, and to conserve, enhance, and sustainably manage forest carbon stocks, referred to as the REDD+ scheme. The implementation of the REDD+ scheme requires an outline of a system for measuring, reporting and verifying progress and changes. Paramount to this system is the establishment of “business-as-usual” baselines, against which the succession of carbon emission reductions of a country can be measured and compared. Hence, information on the amount of forest biomass and forest carbon stock is essential. Combining remote sensing and field data has been recognized to increase the effectiveness in gathering this information, compared to that from the field data alone. The integration of remote sensing and field data also allows making wall-to-wall above-ground biomass (AGB) mapping over large areas. However, to achieve sound results, the integration requires the compatibility of the two datasets. This study took place in the lowland tropical peatland of Sebangau, Central Kalimantan Province, Indonesia. The forestlands, including the underlying thick peat deposits, play an important role in terrestrial carbon storage. Even so, numerous pressures to the area for decades have caused a large part of these peatlands to be devastated and vulnerable to fire. This study aims to contribute to the methodological basis of the integration of field inventory and optical RapidEye data for forest AGB estimation using the case of Sebangau tropical peat-swamp forest. Three sub-studies of related topics were conducted: (i) the effect of forest restricted visibility in the basal area estimates from angle count plot method; (ii) the integration between field inventory and optical RapidEye data for AGB estimation model; and (iii) the effect of different field plot sizes in the AGB estimation models derived from field and RapidEye data. Angle Count Plot is known to be an efficient method for basal area estimation due to rapid application in the field. However, the estimation is rather sensitive to miscounted trees. In this study, the impact of the visibility condition in the Sebangau forest on the basal area estimation from Angle Count Plots was analyzed. Based on field measurements and simulation studies on a 1 ha plot of complete tree measurements, this study determined the maximum distance of visibility of the forest and the suitable basal area factor necessary to employ under this visibility condition. This study found a maximum distance of visibility of 6.6 m and recommends using a basal area factor of 5 for implementing the angle count plot method in the Sebangau forest to reduce the visibility effects on the estimation. For comparison, a dataset from an open savanna forest of Ncaute, Namibia, with almost no visibility restriction, was also used. The results showed that there was a visibility issue for estimating basal area using angle count plot in the Sebangau forest, while it was not found in the Ncaute forest. For the second sub-study, the field plots of the Sebangau peat-swamp forest were combined with the predictor variables derived from RapidEye data to build an AGB estimation model through stepwise multiple linear regression. For better understanding, this modelling approach was applied to two different inventory datasets from a temperate forest of Hainich, Germany. The relationships between field-observed AGB and predictor variables derived from RapidEye data were analyzed across three different inventory datasets representing two different forest types. Results in the study showed that in general, the Pearson correlation coefficients r of field-observed AGB estimates and RapidEye predictors were weak for each dataset. The resulting AGB models exhibited weak performance, given by some common indicators such as Adj R2 and RMSEr. The AGB model for the Hainich forest, which was derived from a dataset with larger plot size, showed a better performance than other AGB models with an Adj R2 of 0.65 and a RMSEr of 10.26%. The AGB estimations with integration of RapidEye data showed a higher relative efficiency, in terms of their variances, when compared to the one derived solely from the field-data. In the third sub-study, the effects of different plot sizes in the resulting AGB model were analyzed using three different scenarios with the Hainich forest dataset of larger plot size. In the first scenario, the same predictors were continuously used to estimate the AGB of different field plot sizes. In the second and third scenario, different predictors were allowed to be selected in the AGB model. Results from the first scenario clearly showed a decreasing performance of the AGB model with decreasing plot size. Additionally, in the second and third scenarios a decreasing pattern in the model’s predictive power (RMSEr) was found by decreasing the plot size. This pattern was not found in the model Adj R2. Instead, the results showed that the relationships between the variability of the field-based AGB estimates and the variability of the RapidEye predictor variables are more complex. This study also showed that the plot perimeter lengths were significantly correlated with the model RMSEr, as well as the model relative maximum residuals. These correlations were slightly higher than those between the plot sizes and the model RMSEr and the model relative maximum residuals.