ABSTRACT
TanDEM-X Digital Surface Models (DSMs) are sensitive to the forest vertical structure function (see Chapter 6), and therefore offer considerable potential for mapping forest dynamics. In this chapter, it is proposed to combine two InSAR DSMs obtained at different dates. In principle, barring the instrument’s configuration changes and seasonality effects, the height differences measured by the two DSMs will be related to vegetation volume changes. Positive changes may be connected to forest regrowth (volume gain), and negative changes to forest degradation or deforestation (volume loss).
The potential of the technique based on this principle is proven by a study that aims at estimating proxies of the forest volume gain and loss in connection with tropical forest degradation in the Republic of the Congo. The experiment was carried out within the framework of the DLR Announcement of Opportunity (AO) TanDEM-X AO (VEGE6702).
Novel analysis methods are adopted. In contrast to classical detection problems, where the data sample statistic is used to estimate probabilities of detection and false alarm, an object-oriented approach is considered. In our case, the signal is composed of objects (spatial random fields), these being defined by the spatial relationship of their constituent samples, i.e. by their morphological properties. Objects hold information on the forest height changes (∆h) and area. Once an object is identified by a clustering algorithm, the hypothesis test of change/no change is provided by within-object mean values and their standard errors. In this way, the impact of coherence noise is greatly reduced.
One more advantage of the object-oriented approach is that it provides leeway in measuring the shape complexity of the objects (patterns of vertical structure change). Shape analysis can make an important connection with the type of process that caused the changes. For example, sharp regular edges and simple geometric objects, such as rectangles, can be ascribed to large-scale deforestation, while more fragmented shapes with irregular borders may be associated with selective logging processes or natural regeneration.
Results of the analysis are reported in terms of:
Probability distributions of height changes (negative, positive) and area for all detected objects.
Standard error of the objects’ mean height change.
Significance of the detected objects’ change with respect to control no-change objects provided by a mean shift measure (effect size) and by statistical decision theory.
Statistic of negative to positive objects spatial location.
Statistic of objects’ proximity to roads.
Stratification of objects by land management.
Shape analysis based on morphological measures (fractal exponent and rectangularity).
The forest height change map by InSAR DSM differencing was compared with a change map generated using Very High Resolution (VHR) optical imagery (0.5 m resolution). Comparison was based on a hit or miss correspondence (spatial overlap) between objects in the two datasets. The comparison relies on observations governed by different physics and therefore reinforces evidence of true events in case of a hit, and suggests important clues about the discrepancies in case of a miss.One important outcome of the study was to have laid the path for the detection of forest regrowth, an important driver in the sequestration of anthropogenic CO2 emissions.
