ABSTRACT In this study, we analyse past spatio-temporal land use and land cover (LULC) change dynamics in the Upper-Mzingwane sub-catchment (UMS) located in the semi-arid Matabeleland south region of Zimbabwe using Landsat 5 Thematic Mapper (L5-TM) and Landsat 8 Operational Land Imager (L8-OLI) imagery for the periods of 1989, 2004, 2013, and 2018. We also model plausible future LULC scenarios for UMS. We distinguished five LULC classes, i.e. water, bareland, dense woodland, shrubland, and grassland using a hybrid approach entailing image classification in R-software using the Random Forest (RF) and Support Vector Machine (SVM) algorithms. Accuracy assessment and Kappa statistics revealed better performance of the SVM hence its outputs were used in the change analysis (i.e. to quantify LULC transitions and trends statistically). We then utilized the TerrSet Land Change Modeller (LCM) (using the Markov Chain algorithm with Multi-Layer Perceptron) to model future LULC scenarios up to 2038 at 5-year intervals. Results revealed that the grass, shrub, and woody vegetation are predominant land covers covering 48.5%, 31.5%, and 18.8% in 1989 and 54.4%, 28.8%, and 15.8%, respectively, in 2018. Dense woodland cover was projected to experience the greatest net loss of 43.57% while shrubland, grassland, water, and bareland increase by 10.73%, 4.5%, 26.85%, and 15.09%, respectively, between 2023 and 2038. We concluded that the UMS has since 1989, been losing and will continue to lose dense woodland cover into the future possibly due to increased human activities such as small scale and illegal gold mining in the area. As such, immediate remedial action needs to be taken to reverse the observed and possible future negative LULC change trends especially for woodland cover so as to avert likely adverse socio-economic, hydrological, and ecological consequences within and beyond the UMS.

中文翻译：

---

津巴布韦南部Matabeleland省半干旱Upper-Mzingwane子流域过去和未来多时相土地利用和土地覆盖变化分析

摘要 在本研究中，我们使用 Landsat 5 专题地图绘制器（L5 -TM) 和 Landsat 8 Operational Land Imager (L8-OLI) 图像在 1989、2004、2013 和 2018 年期间。我们还为 UMS 模拟了可能的未来 LULC 场景。我们使用混合方法在 R 软件中使用随机森林 (RF) 和支持向量机 (SVM) 算法进行图像分类，从而区分了五个 LULC 类，即水、裸地、茂密的林地、灌木地和草地。准确度评估和 Kappa 统计表明 SVM 的性能更好，因此其输出用于变化分析（即统计量化 LULC 转换和趋势）。然后，我们利用 TerrSet 土地变化建模器 (LCM)（使用带有多层感知器的马尔可夫链算法）以 5 年为间隔对 2038 年之前的未来 LULC 情景进行建模。结果显示，草、灌木和木本植被是主要的土地覆盖，1989 年的覆盖率分别为 48.5%、31.5% 和 18.8%，2018 年的覆盖率分别为 54.4%、28.8% 和 15.8%。预计将出现茂密的林地覆盖2023 年至 2038 年间，最大的净损失为 43.57%，而灌木地、草地、水域和裸地分别增加了 10.73%、4.5%、26.85% 和 15.09%。我们得出结论，UMS 自 1989 年以来一直在损失和未来将继续失去茂密的林地覆盖，可能是由于该地区的小规模和非法金矿开采等人类活动的增加。因此，