**4. Vegetation description, classification, and mapping**

According to this review, seven vegetation description, classification, and mapping studies have been conducted in Botswana's protected areas in the last two decades. Most of the studies were carried out in Northern Botswana. The results of the review on the phytosociological and remote sensing methods used by researchers to produce vegetation maps in Botswana's protected are summarized in **Table 2**. The table provides information on the study area, satellite imagery used classification method, and the reference of the researchers who conducted the studies. Van Rooyen [25] used Landsat ETM+ to classify and map the entire Kgalagadi Transfrontier Park (KTP). This produced a vegetation map consisting of 13 major plant communities that were found on Botswana side of the KTP (**Figure 2**). The study found that the vegetation varies from open to dense tree savanna.

In Chobe National Park, Herrero et al. [13] mapped vegetation changes in Chobe riverfront using Landsat TM and AVHRR. The study used random forest because it is a good classification method in spatially and temporally complex heterogeneous savanna landscapes [13]. The overall classification accuracy was 79.8% for 1989– 1990 and 78.5% for 2008–2009 Fox et al. [26] used Landsat 5TM, 7ETM+, and 8OLI to study land cover change (LCC) in Northern Botswana which included Chobe National Park and the six forest reserves. The study found that LCC processes in semi-arid savannas in Southern Africa are influenced by environmental and anthropogenic factors. Interactive self-organizing (ISO) clustering was the classification


*Notes: AVHRR, Advanced Very High Resolution Radiometer; ETM, Enhanced Thematic Mapper; OLI, Operational Land Imager; ISO, Interactive Self-Organizing; MODIS, Moderate Resolution Imaging Spectroradiometer; TM, Thematic Mapper.*

#### **Table 2.**

*Satellite imagery and classification methods used to map the vegetation in different protected areas in Botswana.*

**Figure 2.** *Vegetation map of Kgalagadi Transfrontier Park [25].*

method used resulting in 86.7% overall accuracy and a Kappa coefficient of 0.832, with the highest confusion coming from woodland and shrubland [26]. In Northern Botswana, Sianga and Fynn [14] conducted a study in Savuti-Mababe-Linyanti ecosystem, which also covers Chobe National Park and the forest reserves. The authors used RapidEye &and Landsat to classify and map 15 plant communities in this ecosystem. The study used maximum-likelihood supervised classification and concluded that vegetation map will provide an important database for research in wildlife habitat selection and monitoring of plant communities [14]. Basalumi et al. [27] classified four carbon classes with Landsat 5TM and produced above ground carbon stock map of Kasane Forest Reserve (**Figure 3**). The supervised classification method used was Support Vector Machine and it yielded 97.8% overall classification accuracy. The study suggested that in miombo woodlands, the use of Landsat was ideal for monitoring biomass and carbon stock [27].

Mishra et al. [28] used MODIS to broadly and physiognomically map six vegetation morphology types in Central Kalahari Game Reserve and Khutse Game Reserve. The random forest classification method was used for this study and overall accuracy was 91.9% and Kappa coefficient was 0.88. Lori et al. [29] classified and described nine plant communities in Khutse Game Reserve. Lori [30] has the details of this study which include the mapping of these plant communities using Sentinel-2A imagery (**Figures 4** and **5**). **Figure 6** shows one of these nine plant communities, that is, *Heliotropium lineare*-*Enneapogon desvauxii* community. Maximum-likelihood supervised classification method resulted in overall classification accuracy of 61.67% and overall Kappa coefficient of 58.18%. The heterogeneous savanna vegetation in the study area might have contributed to the optimal

*Use of Phytosociology and Remote Sensing to Classify and Map the Vegetation in Protected Areas… DOI: http://dx.doi.org/10.5772/intechopen.100178*

**Figure 3.** *Above ground carbon stock map of Kasane Forest reserve [27].*

#### **Figure 4.**

*Sentinel-2A natural color RGB (red, green, and blue) imagery with red squares representing the sampling plots in Khutse Game Reserve [30].*

overall accuracy and medium Kappa value [30]. This study differs from the one by Mishra et al. [28] because it used Sentinel-2A imagery that has a high spatial scale (i.e., 10 m) to indicate fine-scale spatial heterogeneity of the area, as compared to MODIS with a low spatial resolution (i.e., 232 m) [28].

In this review, Landsat satellite imagery was the most commonly used. This might be due to the fact that Landsat is the most advanced, free, and easy to access online. The results indicate that maximum-likelihood supervised classification and random forest were the most common classification methods used to classify and map the vegetation and each of the seven studies used different satellite imagery. The results show that there is still a lot that needs to be done in terms of mapping

**Figure 5.**

*Plant community map of Khutse game reserve [30].*

#### **Figure 6.**

*A pan habitat consisting of Heliotropium lineare-Enneapogon desvauxii plant community in Khutse Game Reserve. Photo credit: Tsholofelo Lori.*

and monitoring vegetation in Botswana's protected areas using remote sensing. Even though different researchers use different satellites with different spatial resolutions, there is a general agreement in methods used between different studies in remote sensing of protected areas in Botswana.

*Use of Phytosociology and Remote Sensing to Classify and Map the Vegetation in Protected Areas… DOI: http://dx.doi.org/10.5772/intechopen.100178*
