Skip to content Skip to sidebar Skip to footer

Analyzing 23 Years of Woody Vegetation Change in Namibia

Hard to believe it has already been several years since we started to support Dr. Glynis Humphrey and her colleagues at the University of Cape Town with the provision of Trimble eCognition software. Initially, we assisted them during the COVID-19 pandemic as they could not be on campus to use the software and then the horrible 2021 Table Mountain wildfire ravaged the UCT campus and destroyed the facilities.

So, it is with great pleasure that I share the work that Humphrey et al. recently published in the African Journal of Range & Forage Science. Their article is entitledWoody cover change in relation to fire history and land-use in the savanna-woodlands of north-east Namibia (1996–2019)” and it examines mapping woody cover over a 23 year period from a variety of different image sources (aerial and satellite) with the Object-based Image Analysis (OBIA) tools within the eCognition software. Understanding the dynamics of these savanna ecosystems is crucial for proper management. “For the past century woody plants have increased in many grasslands and savannas worldwide” and according to the authors, this encroachment “reduces the potential for the coexistence between grass and trees causing a shift in ecosystem structure and composition” that ultimately impacts the biodiversity and people living and working there.

For this purpose, the authors chose the Bwabwata National Park in north-east Namibia as their study area (6,247 km2). The park is located within the Kavango Zambezi-Transfrontier Conservation Area (KAZA-TFCA) stretching between the Kwando River in the East and the Okavango River in the West. In order to fully cover the study area from both a spatial and temporal perspective, three different sets of image data were collected: 1) fixed-point repeat photographs were available for 37 sites from 1999 and 2019, 2) historical aerial imagery was available for 1996 (2-2.5 m) and 2007 (1 m), 3) Sentinel-2A satellite imagery (10 m) was acquired for 2019.

The aerial images and satellite imagery were then fed into eCognition for analysis. Three vegetation classes were defined by the authors: 1) trees, 2) shrub-grass mosaic and 3) grassland and bare ground. A random forest (RF) supervised classification approach was chosen by Humphrey et al. and performed on image objects resulting from a multiresolution segmentation. Using an error matrix, the authors conducted accuracy assessment of the automated classification results for the different datasets. The lowest overall accuracy was found in the results from the 1996 aerial data, 79%. Significantly higher overall accuracies were recorded for the 2007 (84%) and 2019 (91%) classifications and “overall, there was good congruence in the qualitative assessment between the ground-truthed repeat photographs and the vegetation classifications from the 1996 aerial and 2019 satellite imagery, which supports the authenticity of the woody cover changes observed”.

The authors go on to provide an extensive exploration of the woody cover change results. In general, a decline of 10.6% was noted for the tree class and an increase of 8.1% was recorded for the shrub-grass mosaic class across the entire park during the 23 year period.  But a more detailed look showed that the decrease in tree cover was more prevalent in the western sections of the park. In addition, habitat fragmentation over time was analyzed and an increase was recorded for all classes the authors explored.

Change table for the different vegetation classes within Bwabwata National Park.

In their analysis of woody cover change in the various land use areas within the Park, the authors concluded that clear differences can be seen between inhabited and conservation areas – particularly a growth in the shrub-grass mosaic class and decrease in the size of trees. The differences in vegetation cover can be associated with the difference in fire histories in the Park – “Shrub-grass mosaic increased and trees declined in the west of the park, where late season burning prevailed. In the east of the park, large trees survived, and we attributed this to the practice of early season burning”.

Examples of the repeat photographs & woody cover classification results (black pixels: trees; dark grey pixels: shrub-grass mosaic; white pixels: grassland and bare ground) showing change between 1999 & 2019 (a: 1999; b: 2019) and east (c: 1999; d: 2019) in Bwabwata National park (1996–2019).

Overall, an interesting study and for me it is always interesting to not only see the versatility of remote sensing data in answering so many different questions, but how the Trimble eCognition software can be integrated to automate the analytical process and transform geospatial data into valuable information to improve our understanding of our fields of study, now and in the future.