GIS and Remote Sensing for Environmental Monitoring and Natural Resource Management: Concepts, Methods and Applications

Geographic data rarely satisfy the independence and identical-distribution assumptions on which conventional statistical inference rests. Observations that are close in space tend to resemble one another, relationships between variables often change from place to place, and the results of an analysis can depend on the size and configuration of the spatial units used to collect or aggregate data. These three properties, namely spatial autocorrelation, spatial heterogeneity and scale dependence, have generated separate methodological traditions in quantitative geography, spatial statistics, spatial econometrics, ecology and epidemiology. Existing reviews have tended to examine one property within one discipline, leaving their interdependence and the consequences for inference insufficiently synthesised. This critical narrative review evaluates the theoretical foundations, measurement approaches, modelling strategies and applied evidence associated with the three properties and assesses how they interact. Literature was identified through structured searches of multidisciplinary, mathematical, economic and biomedical scholarly indexes, supplemented by citation tracking, and was appraised for conceptual contribution, methodological rigour and the strength of supporting evidence. The synthesis shows that the case for accounting for spatial dependence in inference is well established, but disagreement persists over whether neglecting it biases regression coefficients, how spatial weights should be specified, and whether spatial random effects confound covariate effects. Local models have made spatial heterogeneity tractable, yet apparent spatial variation in coefficients can also arise from non-linearity, collinearity, small local samples or unmodelled dependence. Evidence on the modifiable areal unit problem confirms that aggregation can alter correlation and regression results, although the magnitude of the effect depends strongly on the underlying process, the level of autocorrelation and the model specification. In machine learning applications, the same properties govern the credibility of accuracy assessment and the transferability of models across places. The review argues that dependence, heterogeneity and scale should be treated as jointly determined features of a data-generating process rather than as separate statistical nuisances. Priorities include estimand-explicit spatial causal analysis, shared simulation benchmarks, routine reporting of scale and zoning sensitivity, and validation designs that reflect the intended prediction domain.