The effects of natural processes on deposited mineral material of postindustrial sites is underestimated. Natural vegetation development on mineral material substratum is an unappreciated way of site management. Due to the classification‐based approach to assembly of plant community diversity, remote sensing methods have limited application. We aimed to assess whether remotely sensed data allow for building predictive models, able to recognise vegetation variability along the main gradients of species composition. We assessed vegetation in 321 study plots on four coal‐mine spoil heaps in Silesia (S Poland). We determined the main gradients of species composition using detrended correspondence analysis (DCA), and we identified how DCA scores describe vegetation variability. DCA axes explained 38.5%, 35.4%, 31.4%, and 20.1% of species composition variability. We built machine learning models of DCA scores using multispectral satellite images and airborne laser scanning data as predictors. We obtained good predictive power of models for the first two DCA axes (R² = 0.393 and 0.443, root mean square errors, RMSE = 0.571 and 0.526) and low power for the third and fourth DCA axes (R² = 0.216 and 0.064, RMSE = 0.513 and 0.361). These scores allowed us to prepare a vegetation map based on DCA scores, and distinguish meadow‐like from forest‐edge‐like vegetation, and to identify thermophilous and highly productive vegetation patches. Our approach allowed us to account for species composition gradients, which improved remote sensing‐based vegetation surveys. This method may be used for planning future management.

中文翻译：

---

利用遥感跟踪工业后植被的发展

低估了自然过程对后工业场所沉积矿物质的影响。矿物材料基质上的天然植被发育是一种不受重视的场地管理方式。由于基于分类的植物群落多样性组装方法，遥感方法的应用受到限制。我们旨在评估遥感数据是否可用于建立预测模型，从而能够识别沿物种组成主要梯度的植被变化。我们在西里西亚（S波兰）的四个煤sp堆上的321个研究区中评估了植被。我们使用去趋势对应分析（DCA）确定了物种组成的主要梯度，并且确定了DCA得分如何描述植被变异性。DCA轴解释了38.5％，35.4％，31.4％和20。1％的物种组成变异性。我们使用多光谱卫星图像和机载激光扫描数据作为预测指标，建立了DCA分数的机器学习模型。我们获得了前两个DCA轴模型的良好预测能力（R ² = 0.393和0.443，均方根误差，RMSE = 0.571和0.526），第三和第四DCA轴的低功耗（R ² = 0.216和0.064，RMSE = 0.513和0.361）。这些分数使我们能够根据DCA分数准备植被图，区分草甸状和林缘状的植被，并识别嗜热和高产的植被斑块。我们的方法使我们能够考虑物种组成梯度，从而改善了基于遥感的植被调查。此方法可用于计划未来的管理。