Land-surface parameters for spatial predictive mapping and modeling

Land-surface parameters derived from digital land surface models (DLSMs) (for example, slope, curvature, topographic position, roughness, aspect, heat load index, and moisture index) can serve as key predictor variables in a wide variety of mapping modeling tasks relating to geomorphic processes, landform delineation, ecological habitat characterization, geohazard, soil, wetland, general thematic modeling. However, selecting features the large number potential derivatives that may be predictive for specific feature or process complicated, existing literature offer contradictory incomplete guidance. The availability multiple data sources need define moving window shapes, sizes, cell weightings further complicate optimizing space. This review focuses on calculation use DLSM empirical spatial applications, which rely training explanatory make predictions landscape processes over defined geographic extent. target audience this is researchers analysts undertaking most widely used terrain variables. To outline best practices highlight future research needs, we range land-surface steepness, local relief, rugosity, slope orientation, solar insolation, characterize their relationship processes. We then discuss important considerations when such assist answering two critical questions: What conditions does given measure characterize? How might particular metric relate phenomenon being mapped, modeled, studied? recommend landscape- problem-specific pilot studies answer, extent possible, these questions interest task. describe techniques reduce size space using selection reduction methods, assess importance contribution metrics, parameterize windows at varying scales alternative methods while highlighting strengths, drawbacks, knowledge gaps techniques. Recent developments, explainable machine learning convolutional neural network (CNN)-based deep learning, guide and/or minimize engineering ease DLSMs tasks.