Table 1

Descriptive list of reviewed literature regarding land price estimation/mapping grouped by estimation approach: (1) hedonic models, (2) geostatistical methods, (3) machine learning algorithms, and (4) comparison of various approaches"

Figure 1

Fukushima prefecture and its administrative boundaries, topographic features, transportation lines, and evacuation zones after the Fukushima Daiichi Nuclear Plant disaster (as of September 2015)"

Figure 2

Changes in land prices averaged by land type in Fukushima prefecture (2005-2018)"

Table 2

The three mathematical models used for kriging and their abbreviations"

Table 3

Summary of spatial prediction models used in this study: Linear, nonlinear, and regression trees models are grouped as proposed by Kuhn and Johnson (2013). Abbreviations are used to refer to each method in the manuscript"

Table 4

List of explanatory variables selected in this study with their data sources and the related abbreviations"

Table 5

Overview of datasets used in the study, their sources, and the year of release"

Figure 3

Methodological framework of the study"

Figure 4

The distribution of land price samples in the study area"

Table 6

Regression results with detailed explanatory variables and their estimated coefficients"

Figure 5

Fitted semi-variograms for the kriging models for the year 2015: (a) Exp: Exponential (b) Gau: Gaussian (c) Sph: Spherical. The nugget, range, and sill values and the mathematical models are shown in the bottom right corner"

Figure 6

The results of the regression kriging for the year 2015 using the exponential model (upper), Gaussian model (middle), and spherical model (lower). On the left are the estimated log-transformed land prices using regression kriging. On the right are the validation errors in the training samples. Capital letters denote major cities within Fukushima prefecture, which are A: Fukushima, B: Koriyama, C: Iwaki, D: Aizuwakamtsu, and E: Shirakawa"

Table 7

Prediction errors of validation and cross-validation tests for the three kriging models"

Figure 7

Land price maps for the year 2015 predicted from officially published land price observations using regression kriging based on three mathematical models (ordered from left to right): (1) Krig.EXP: Exponential model, (2) Krig.GAU: Gaussian model, and (3) Krig.SPH: Spherical model"

Figure 8

Boxplots of performance of machine learning methods in terms of the MAE, the RMSE, and R2 for the year 2015"

Table 8

Prediction errors and accuracy of machine learning methods"

Figure 9

Observed land prices vs. predicted land prices for the year 2015 in the testing samples by different machine learning methods (ordered from left to right, up to down): (1) GLM: generalized linear model, (2) GAMS: generalized linear model using splines, (3) SVMLinear: support vector machines with linear kernel, (4) MARS: multivariate adaptive regression spline, (5) kNN: k-nearest neighbors, (6) SVMRadial: support vector machines with radial basis function kernel, (7) Cubist, (8) GBM: stochastic gradient boosting and (9) RF: random forest"

Figure 10

Land price maps for the year 2015 predicted from officially published land price observations using machine learning algorithms (ordered from left to right, up to down): (1) GLM: generalized linear model, (2) GAMS: generalized linear model using splines, (3) SVMLinear: support vector machines with linear kernel, (4) MARS: multivariate adaptive regression spline, (5) kNN: k-nearest neighbors, (6) SVMRadial: support vector machines with radial basis function kernel, (7) Cubist, (8) GBM: stochastic gradient boosting and (9) RF: random forest"

Figure 11

Maps of differences in the 2015 land prices between the best-performing machine learning algorithms: (1) RF: Random Forest, (2) Cubist, (3) MARS: Multivariate Adaptive Regression Spline and (4) GAMS: Generalized Linear Model using Splines and kriging exponential model. A1, A2, A3, and A4 show zoomed-in maps of Koriyama city and its outskirts"

Figure 12

Area percentage of RF- and krig.EXP-based estimated land price for the year 2015 distributed by predefined ranges in Fukushima prefecture and its subregions"