Quantifying Efficiency of Sliding-Window Based Aggregation Technique by Using Predictive Modeling on Landform Attributes Derived from DEM and NDVI

Topographic features impact biomass and other agriculturally relevant observables. However, conventional tools for processing digital elevation model (DEM) data in geographic information systems have severe limitations. Typically, 3-by-3 window sizes are used for evaluating the slope, aspect and curvature. As a consequence, high resolution DEMs have to be resampled to match the size of typical topographic features, resulting in low accuracy and limiting the predictive ability of any model using such features. In this paper, we examined the usefulness of DEM-derived topographic features within Random Forest models that predict biomass. Our model utilized the derived topographic features and achieved 95.31% accuracy in predicting Normalized Difference Vegetation Index (NDVI) compared to a 51.89% accuracy obtained for window size 3-by-3 in the traditional resampling model. The efficacy of partial dependency plots (PDP) in terms of interpretability was also assessed