pone.0338920.t008 -

Urban expansion and Land Use Land Cover (LULC) change pose critical challenges for sustainable urban planning and risks to food security. This study analyzes multi-temporal Landsat imagery from 1990 to 2020 for five major cities, Islamabad, Karachi, Lahore, Peshawar, and Quetta in Pakistan using the Smile Random Forest (SRF) algorithm within the Google Earth Engine (GEE) platform. Classification accuracies ranged from 86–90%, with Cohen’s Kappa coefficients between 0.86 and 0.92, demonstrating substantial to almost perfect agreement. The results reveal significant increases in urban areas: Karachi expanded from 12.4% in 1990 to 41.3% in 2020, Lahore from 15.2% to 39.8%, and Islamabad from 9.1% to 28.6%, primarily at the expense of vegetation and barren land. Elevation also influenced LULC dynamics, with higher-altitude cities like Quetta exhibiting slower but more resource-constrained urban development. A change matrix quantified class transitions, showing that urban land predominantly expanded into agricultural and vegetative land areas, raising concerns about long-term food security. Future projections using the MOLUSCE–ANN model indicate continued urban expansion by 2030, particularly in Karachi and Lahore, where built-up areas are projected to exceed 45% of total land cover. Compared with previous studies that employed Classification and Regression Trees (CART), Support Vector Machine (SVM), and CA–Markov models in single-city or short-term contexts, this study provides a multi-decadal, multi-city analysis with predictive capacity and robust validation, offering novel insights into Pakistan’s urbanization trajectory. By linking LULC change to the implications for natural resources and food security, the study contributes actionable evidence to support actions against disaster risk reduction, sustainable development and SDGs-aligned with urban policies.

城市扩张与土地利用/土地覆盖（Land Use Land Cover, LULC）变化给可持续城市规划带来严峻挑战，同时对粮食安全构成威胁。本研究依托谷歌地球引擎（Google Earth Engine, GEE）平台，采用Smile随机森林（Smile Random Forest, SRF）算法，对巴基斯坦伊斯兰堡、卡拉奇、拉合尔、白沙瓦与奎达5座主要城市1990至2020年的多时相Landsat影像开展分析。本研究的分类精度介于86%至90%之间，科恩kappa系数处于0.86至0.92区间，表明分类结果具有高度一致性至近乎完美的一致性。研究结果显示城市用地规模显著增长：卡拉奇城市用地占比从1990年的12.4%升至2020年的41.3%，拉合尔从15.2%升至39.8%，伊斯兰堡则从9.1%升至28.6%，城市扩张主要以植被与裸地为代价。海拔高度同样对LULC动态变化存在影响，奎达等高海拔城市的城市扩张速度相对平缓，但受资源约束更为显著。通过变化矩阵量化了地类转移情况，结果显示城市用地主要向农业与植被用地扩张，这引发了学界对长期粮食安全的担忧。采用MOLUSCE–ANN模型开展的未来情景预测显示，至2030年城市用地仍将持续扩张，尤其在卡拉奇与拉合尔两市，其建成区占比预计将超过总土地覆盖的45%。相较于此前采用分类与回归树（Classification and Regression Trees, CART）、支持向量机（Support Vector Machine, SVM）以及CA–Markov模型开展的单城市或短期相关研究，本研究开展了跨数十年的多城市分析，兼具预测能力与严谨的验证环节，为巴基斯坦城市化轨迹研究提供了全新视角。本研究将LULC变化与自然资源及粮食安全的影响相关联，为减灾防灾、可持续发展以及契合城市政策的可持续发展目标（Sustainable Development Goals, SDGs）相关行动提供了可落地的实证依据。