Remote sensing scientists are increasingly adopting machine learning classifiers for land cover and land use (LCLU) mapping, but model selection, a critical step of the machine learning classification, has usually been ignored in the past research. In this paper, step-by-step guidance (for classifier training, model selection, and map production) with supervised learning model selection is first provided. Then, model selection is exhaustively applied to different machine learning (e.g. Artificial Neural Network (ANN), Decision Tree (DT), Support Vector Machine (SVM), and Random Forest (RF)) classifiers to identify optimal polynomial degree of input features (d) and hyperparameters with Landsat imagery of a study region in China and Ghana. We evaluated the map accuracy and computing time associated with different versions of machine learning classification software (i.e. ArcMap, ENVI, TerrSet, and R). The optimal classifiers and their associated polynomial degree of input features and hyperparameters vary for the two image datasets that were tested. The optimum combination of d and hyperparameters for each type of classifier was used across software packages, but some classifiers (i.e. ENVI and TerrSet ANN) were customized due to the constraints of software packages. The LCLU map derived from ENVI SVM has the highest overall accuracy (72.6%) for the Ghana dataset, while the LCLU map derived from R DT has the highest overall accuracy (48.0%) for the FNNR dataset. All LCLU maps for the Ghana dataset are more accurate compared to those from the China dataset, likely due to more limited and uncertain training data for the China (FNNR) dataset. For the Ghana dataset, LCLU maps derived from tree-based classifiers (ArcMap RF, TerrSet DT, and R RF) routines are accurate, but these maps have artefacts resulting from model overfitting problems.

The extensive record of Landsat imagery is commonly used to map urban land-cover and land-use change. Random forest (RF) classification was applied for mapping more detailed urban land-use and change categories than is typically attempted with Landsat data. Two dates of Landsat imagery (1990 and 2015) were utilized with surface reflectance, Vegetation-Impervious-Soil (V-I-S) fractions, grey-level cooccurrence matrix (GLCM) of V-I-S, and temporal variation of V-I-S inputs. GLCM V-I-S and temporal variation of Vegetation as input features of RF classifiers slightly improved accuracies of land use maps. A change map derived from an overlay analysis between the 2015 map and a Landsat-derived urban expansion map was more accurate than one from post-classification comparison of 1990 and 2015 maps. For the Taiwan study area, Transportation Corridor land use tended to lead conversion to Residential and Employment types in relatively undeveloped districts, and extensive urban land-use change occurred in peri-urban areas.

Spatial autocorrelation in model residuals can have a significant impact on the results of spatial or space-time models. This can result in misleading estimates of the influence of different factors, potentially exaggerating or even reversing the perceived effects of these factors. This study also considers the potential implications of the Modifiable Areal Unit Problem (MAUP) in the context of spatial-temporal models. In this case study for southeastern Ghana, we examined whether and how spatial autocorrelation in model residuals might generate bias in regression coefficients when explaining women’s body mass index (BMI) across urban and rural areas. Eigenvector spatial filtering, with various settings of influential zones, was systematically tested in a latent trajectory model to detect the impacts of spatial autocorrelation.

Urban expansion is a form of land cover and land use change (LCLUC) that occurs globally, and population growth can be a driver of and be driven by LCLUC. Determining the cause–effect relationship is challenging because the temporal resolution of population data is limited by decadal censuses for most countries. The purpose of this study is to explore the relationship and relative timing between population change and land use change based on a case study of northern Taiwan from 1990 to 2015. A unique dataset on population was acquired from annually-updated governmental-based population registers maintained at the district level, and land-use expansion data (Residential, Employment, and Transportation Corridor categories) were derived from dense time series of Landsat imagery. Linear regression was applied to understand the general relationship between population and land use and their changes. The strongest relationships were found between population and areal extent of Residential land use, and between population change and Residential areal change. Lagged correlation analysis was implemented for identifying the time lag between population growth and land use change. Most districts exhibited Residential and Employment expansion prior to population growth, especially for districts in the periphery of metropolitan areas. Conversely, the core of metropolitan areas exhibited population growth prior to Residential and Employment expansion. Residential and Employment expansion were deemed to be drivers of population change, so population change was modeled with ordinary least square and geographically weighted regression with Residential and Employment expansion in both synchronized and time lag manners. Estimated population growth was found to be the most accurate when geographic differences and time lags from urban land use expansion were both incorporated.