Using Random Forest Classification to generate Better Wet Area Maps from High Resolution Digital Elevation Models: a Study in a Boreal Forest Landscape

Forested wetlands and wet soils near streams and lakes are often missing from current maps which makes wet areas difficult to manage. Wet soils also have lower bearing capacity making them more susceptible to soil disturbance from heavy forestry machines resulting in sediment and mercury transport to nearby surface waters. Topographical modelling of hydrological features such as topographical wetness index and depth to water has been suggested as a solution to this problem. These methods are easy to implement on large scales but are also static and do not take differences in spatial runoff patterns or soil textures into account. Therefore the aim of this study was to evaluate if machine learning can be used to create more accurate maps of wet soils. We used a machine learning approach where the National Forest Inventory of Sweden was used to train two random forest classifiers. The first model used a number of topographically derived variables along with soil types and runoff data while the second model only used topographically derived variables. The predicted maps were evaluated using Cohen’s kappa index of agreement, out of bag error and visual inspection of the Krycklan catchment. Both the predicted maps had a substantial agreement with the field plots and agreed well with the authors’ first-hand knowledge of the Krycklan catchment. Additionally there was sufficient information in the digital elevation model and including additional factors such as quaternary deposits and runoff only improved the overall accuracy by 1 % and increased kappa by 0.02.

当前地图往往遗漏了森林湿地和溪流湖泊附近的湿土区域，这给湿地区域的管理带来了困难。湿土的承重能力较低，使其更容易受到重型林业机械的土壤扰动，进而导致泥沙和汞等污染物进入邻近的地表水体。针对这一问题的解决方案之一，是提出对地形水文特征如地形湿度指数和水深进行地形建模。这些方法易于在大规模上实施，但同时也具有静态性，未能充分考虑空间径流模式和土壤纹理的差异。因此，本研究的目的是评估机器学习是否能够被用于创建更精确的湿土地图。本研究采用机器学习方法，利用瑞典国家森林资源清查数据对两种随机森林分类器进行训练。第一个模型采用了地形派生变量、土壤类型和径流数据等多种因素，而第二个模型仅使用了地形派生变量。通过Cohen's kappa一致性指数、袋外误差以及Krycklan流域的视觉检查对预测地图进行了评估。两个预测地图与实地测量数据具有显著的一致性，并且与作者对Krycklan流域的亲身了解相符。此外，数字高程模型中包含了足够的信息，而包括第四纪沉积物和径流等因素的额外因素，整体准确率提高了1%，kappa指数增加了0.02。