Machine learning approaches are increasingly used across numerous applications in order to learn from data and generate new knowledge discoveries, advance scientific studies and support automated decision making. In this knowledge entry, the fundamentals of Machine Learning (ML) are introduced, focusing on how feature spaces, models and algorithms are being developed and applied in geospatial studies. An example of a ML workflow for supervised/unsupervised learning is also introduced. The main challenges in ML approaches and our vision for future work are discussed at the end.
Spatial interpolation methods use the measured values at given locations to estimate the values at unsampled locations, for example, in computing raster digital elevation models from scattered measured elevations. Since this problem does not have a unique solution, many approaches have been developed to accomplish this task. Methods based on linear superposition of radial basis functions (RBF) centered at the data points include multivariate splines that simultaneously minimize the sum of the deviations from the measured points and the smoothness seminorm referred to also as a roughness penalty. The thin plate spline minimizes the 2D surface curvature and mimics a thin steel plate forced to pass through the data points: its equilibrium shape minimizes the bending energy which is closely related to the surface curvature. There are many generalizations such as spline with tension that controls the plate stiffness, while regularized spline enables direct calculations of surface gradients and curvatures making it effective for terrain modeling with simultaneous topographic analysis. Trivariate splines are used to interpolate meteorological variables with influence of topography. The RBF splines are related to universal kriging with the choice of the covariance function determined by the smoothness seminorm. Multiquadric RBF methods are similar in formulation and performance to splines.