Abstract:Waldo Tobler frequently reminded us that the law named after him was nothing more than calling for exceptions. This article discusses one of these exceptions. Spatial relations between points are frequently modeled as vectors in which both distance and direction are of equal prominence. However, in Tobler's first law of geography, such a relation is described only from the perspective of distance by relating the decreasing similarity of observations in some attribute space to their increasing distance in geogr… Show more
“…In fact, in normal spatial analysis, isotropicity is the "default" assumption in most of the time. Although anisotropic versions of many geospatial analysis techniques have been developed such as directional kriging (Te Stroet and Snepvangers 2005), anisotropic clustering (Mai et al 2018), direction remains on the level of an afterthought (Zhu et al 2019b). A similar situation can be seen in the current location encoding research, or GeoAI research in general.…”
Section: Definitionsmentioning
confidence: 91%
“…Property 2.2 is a reflection of the Generalized First Law of Geography (Zhu et al 2019b) which includes direction into the consideration of similarities. In this paper, we call a location encoder an isotropic location encoder if it only preserves the spatial proximity but ignores the variance of location embeddings when direction changes.…”
A common need for artificial intelligence models in the broader geoscience is to represent and encode various types of spatial data, such as points (e.g., points of interest), polylines (e.g., trajectories), polygons (e.g., administrative regions), graphs (e.g., transportation networks), or rasters (e.g., remote sensing images), in a hidden embedding space so that they can be readily incorporated into deep learning models. One fundamental step is to encode a single point location into an embedding space, such that this embedding is learning-friendly for downstream machine learning models such as support vector machines and neural networks. We call this process location encoding. However, there lacks a systematic review on the concept of location encoding, its potential applications, and key challenges that need to be addressed. This paper aims to fill this gap. We first provide a formal definition of location encoding, and discuss the necessity of location encoding for GeoAI research from a machine learning perspective. Next, we provide a comprehensive survey and discussion about the current landscape of location encoding research. We classify location encoding models into different categories based on their inputs and encoding methods, and compare them based on whether they are parametric, multi-scale, distance preserving, and direction aware. We demonstrate that existing location encoding models can be unified under a shared formulation framework. We also discuss the application of location encoding for different types of spatial data. Finally, we point out several challenges in location encoding research that need to be solved in the future.
“…In fact, in normal spatial analysis, isotropicity is the "default" assumption in most of the time. Although anisotropic versions of many geospatial analysis techniques have been developed such as directional kriging (Te Stroet and Snepvangers 2005), anisotropic clustering (Mai et al 2018), direction remains on the level of an afterthought (Zhu et al 2019b). A similar situation can be seen in the current location encoding research, or GeoAI research in general.…”
Section: Definitionsmentioning
confidence: 91%
“…Property 2.2 is a reflection of the Generalized First Law of Geography (Zhu et al 2019b) which includes direction into the consideration of similarities. In this paper, we call a location encoder an isotropic location encoder if it only preserves the spatial proximity but ignores the variance of location embeddings when direction changes.…”
A common need for artificial intelligence models in the broader geoscience is to represent and encode various types of spatial data, such as points (e.g., points of interest), polylines (e.g., trajectories), polygons (e.g., administrative regions), graphs (e.g., transportation networks), or rasters (e.g., remote sensing images), in a hidden embedding space so that they can be readily incorporated into deep learning models. One fundamental step is to encode a single point location into an embedding space, such that this embedding is learning-friendly for downstream machine learning models such as support vector machines and neural networks. We call this process location encoding. However, there lacks a systematic review on the concept of location encoding, its potential applications, and key challenges that need to be addressed. This paper aims to fill this gap. We first provide a formal definition of location encoding, and discuss the necessity of location encoding for GeoAI research from a machine learning perspective. Next, we provide a comprehensive survey and discussion about the current landscape of location encoding research. We classify location encoding models into different categories based on their inputs and encoding methods, and compare them based on whether they are parametric, multi-scale, distance preserving, and direction aware. We demonstrate that existing location encoding models can be unified under a shared formulation framework. We also discuss the application of location encoding for different types of spatial data. Finally, we point out several challenges in location encoding research that need to be solved in the future.
“…Spatial interpolation (SI) or spatial prediction is a crucial topic in geosciences and related fields such as geology 1 , 2 , geography 3 – 5 , hydrology 6 , 7 , environment 8 – 11 , and agriculture 12 . To address various concerns in these disciplines, a series of SI methods are developed, which differ in interpolation objectives and basics 13 , 14 .…”
Section: Introductionmentioning
confidence: 99%
“…Nevertheless, no matter what kinds of contexts are being faced, enhancing the estimation accuracy and reliability is a common goal that most SI methods pursue, and so does the typical SI method—inverse distance weighting (IDW) 1 , 5 , 15 – 21 . In general, the interpolation accuracy of the conventional IDW or its variants could be improved by choosing a set of appropriate parameters such as the search model of local samples or observed data 3 , 22 – 24 , the type of distance metric 19 , 25 , 26 , and the exponent imposed on the distance 7 , 22 , 23 , 27 , 28 .…”
Many geoscience problems involve predicting attributes of interest at un-sampled locations. Inverse distance weighting (IDW) is a standard solution to such problems. However, IDW is generally not able to produce favorable results in the presence of clustered data, which is commonly used in the geospatial data process. To address this concern, this paper presents a novel interpolation approach (DIDW) that integrates data-to-data correlation with the conventional IDW and reformulates it within the geostatistical framework considering locally varying exponents. Traditional IDW, DIDW, and ordinary kriging are employed to evaluate the interpolation performance of the proposed method. This evaluation is based on a case study using the public Walker Lake dataset, and the associated interpolations are performed in various contexts, such as different sample data sizes and variogram parameters. The results demonstrate that DIDW with locally varying exponents stably produces more accurate and reliable estimates than the conventional IDW and DIDW. Besides, it yields more robust estimates than ordinary kriging in the face of varying variogram parameters. Thus, the proposed method can be applied as a preferred spatial interpolation method for most applications regarding its stability and accuracy.
“…This law provides the foundation of the fundamental concepts in spatial dependence and spatial autocorrelation, and is utilized specifically in spatial interpolation techniques. Spatial autocorrelation (Zhu et al, 2019) is a key concept that is used to analyse the degree of dependency among observations (samples) in a given geographic space. Distance between neighbours, lengths of shared borders, and orientation are just some of the measurements used in conjunction, when modelling a given field, to estimate the unknowns.…”
Defining spatial distribution of airborne volcanic ash in the neighbourhood of an erupting volcano is a synoptic scale problem, severely impacting lives and livelihoods. Robust algorithms are needed to model such complex phenomenon from sparse field data. This study investigated optimal modelling of the spatial dispersion of ash using Empirical Bayesian Kriging (EBK): a geostatistical, probabilistic algorithm. Both distance and ash temperature values of samples from the 2010 Icelandic eruption were spatially correlated using semivariograms to generate prediction and error surfaces. Results showed that block averages were 90% accurate as validated against NCEP NWP model data. The work supports the utility of EBK in datasets where spatial autocorrelation is not significant. Furthermore, the results could help generate risk maps to delineate safety zones for aircrafts.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.