The Spatiotemporal Changing Dynamics of Miombo Deforestation and Illegal Human Activities for Forest Fire in Kundelungu National Park, Democratic Republic of the Congo

Abstract

In the Kundelungu National Park (KNP), southeast of the Democratic Republic of Congo, illicit human activities including recurrent bushfires contribute to constant regression of forest cover. This study quantifies the landscape dynamics and analyses the spatio-temporal distribution of bushfire occurrence within KNP. Based on classified Landsat images from 2001, 2008, 2015 and 2022, the evolutionary trend of land cover was mapped and quantified through landscape metrics. The spatial transformation processes underlying the observed landscape dynamics were identified based on a decision tree. Finally, the spatio-temporal fire risk assessment was carried out after defining the burnt areas for each year between 2001 and 2022. The obtained results, expressed by the process of dissection and attrition of patches, show that the forest cover has regressed from 2339 km2 to 1596 km2 within the PNK, with an annual deforestation rate varying from 0.8% to 3.4% between 2001 and 2022. Over the same period, the average distance between forest patches has increased significantly, indicating fragmentation and spatial isolation. On the other hand, savannahs as well as field and fallow mosaics have expanded within KNP through the creation of new patches. In addition, several active fires affected more savannahs between 2001 (70 km2 in Integral Zone, 239 km2 in Annex Zone and 309 km2 in KNP) and 2022 (76 km2 in Integral Zone, 744 km2 in Annex Zone and 819 km2 in KNP), limiting their capacity to evolve into forests. Overall, anthropogenic pressure is higher in the Annex Zone of the KNP. Illegal agricultural development and vegetation fires have thus doubled the level of landscape disturbance in 21 years. Our observations justify the need to strengthen protection measures for KNP by limiting repeated human intrusions.

Mapping and Quantification of Miombo Deforestation in the Lubumbashi Charcoal Production Basin (DR Congo): Spatial Extent and Changes between 1990 and 2022

“…The negative spatial dynamics of miombo in the LCBP were accompanied by low densification and regeneration. This may have been caused by recurring vegetation fires in the region and the conversion of areas into agricultural land that were exploited for charcoal production [19,39]. These justify the decrease in the stability of miombo land cover, which is supported by the three-fold decrease in its stability index value, a trend also confirmed previously in Ref.…”

“…Currently, analysing changing landscape trends is facilitated by readily available satellite imagery sources and various machine learning techniques for image classification [31]. Several studies have examined the landscape dynamics in the Katanga region, particularly in the Lubumbashi area, using satellite imagery [22][23][24][32][33][34][35][36][37][38][39]. However, none have reported the dynamics of landscape anthropisation in the Lubumbashi charcoal production basin (LCPB).…”

“…Structural dynamics of landscape Landscape dynamics were quantified using the number of patches and the class area, which are elementary metrics [63] that can be used to easily evaluate the landscape fragmentation phenomenon. Spatial transformation processes were identified using a decision tree (Figure 2) based on the comparison of class area, perimeter and number of patches before and after landscape transformation [39,61,64]. Among these processes, anthropogenic land covers are characterised by aggregation, enlargement and shift.…”

Quantifying Forest Cover Loss during the COVID-19 Pandemic in the Charcoal Production Basin of Lubumbashi (DR Congo) through Remote Sensing and Landscape Analysis

“…For precise analysis, distinct land cover units were identified and coded across scenes, and training areas known as Regions of Interest (ROIs) were delineated in each study year, strategically selected during the dry season for temporal consistency [40]. These ROIs, consisting of polygons with 2 and 4 pixels per area, were tailored for each land cover type, totaling 200 ROIs per type, aimed at reducing transitional effects known as the ʹmixelʹ effect [47,48]. The ROIs were utilized as training data to construct a model based on the random forest algorithm, which combines predictions from multiple decision trees to enhance prediction accuracy and minimize classification errors [49,50].…”

Development of a Data-Based Machine Learning Model for Classifying and Predicting Property Damage Caused by Fire

“…The depth can be considered a representative hyper-parameter; however, overfitting is highly likely to occur in the training dataset with increased depth [22]. Examples that use this technique can be found in [23][24][25][26].…”