Accuracy is increasingly recognized as an important dimension in geospatial information and analyses. A strategy well suited for map users who usually have limited information about map lineages is proposed for location-specific characterization of accuracy in land cover change maps. Logistic regression is used to predict the probabilities of correct change categorization based on local patterns of map classes in the focal three by three pixel neighborhood centered at individual pixels being analyzed, while kriging is performed to make corrections to regression predictions based on regression residuals at sample locations. To promote uncertainty-informed accuracy characterization and to facilitate adaptive sampling of validation data, standard errors in both regression predictions and kriging interpolation are quantified to derive error margins in the aforementioned accuracy predictions. It was found that the integration of logistic regression and kriging leads to more accurate predictions of local accuracies through proper handling of spatially-correlated binary data representing pixel-specific (in)correct classifications than kriging or logistic regression alone. Secondly, it was confirmed that pixel-specific class labels, focal dominances and focal class occurrences are significant covariates for regression predictions at individual pixels. Lastly, error measures computed of accuracy predictions can be used for adaptively and progressively locating samples to enhance sampling efficiency and to improve predictions. The proposed methods may be applied for characterizing the local accuracy of categorical maps concerned in spatial applications, either input or output.
In China, ground-level ozone has shown an increasing trend and has become a serious ambient pollutant. An accurate spatiotemporal distribution of ground-level ozone concentrations (GOCs) is urgently needed. Generalized linear models (GLMs) and Bayesian maximum entropy (BME) models are practical for predicting GOCs. However, GLMs have limited capacity to capture temporal variations and can miss some short-term and regional patterns, while the performance of BME models may degrade in cases of sparse or imperfect monitoring networks. Thus, to predict nationwide 1 km monthly average GOCs for China, we designed a novel hybrid model containing three modules. (1) A GLM was established to accurately describe the variability in GOCs in the space domain. (2) A BME model incorporating GLM residuals was employed to capture the temporal variability of GOCs in detail. (3) A combination of GLM and BME models was developed based on the specific broad range of each submodel. According to the cross-validation results, the hybrid model exhibited superior performance, with coefficient of determination (R2) values of 0.67. The predictive performance of the large-scale and high-resolution hybrid model is superior to that in previous studies. The nationwide spatiotemporal variability of the GOCs derived from the hybrid model shows that they are valuable indicators for ground-level ozone pollution control and prevention in China.
As in conventional error matrix-based accuracy assessments, collocated reference sample data are often used for characterizing per-pixel (local) accuracies in land-cover change maps so that local accuracy predictions can be made using direct methods. In that way, correctness in “from-to” change categorization at sample pixels is assessed and modeled directly. To circumvent the issue of reference sample data being non-collocated, as is often the case for sample data collected independently for mono-temporal reference land-cover labeling or those added necessarily to reflect landscape changes, the PXCOV (Product rule with adjustment for cross-COVariance between single-date classification correctness) method was developed previously. However, the use of PXCOV becomes complicated when few or no collocated sample data are available and cross-validation cokriging, a procedure involving non-trivial geostatistical modeling, has to be incurred for estimation of cross-correlation. To overcome PXCOV’s lack of practicality when using mostly non-collocated sample data, this paper presents a simple alternative. It is furnished through stratified approximation of cross-correlation and features combined use of minimum and multiplication operators. Specifically, in this composite method (named Fuzzy+Product), minimum operator (resembling fuzzy set “min” operator and thus named Fuzzy) is applied over no-change pixels stratum where maximum correlation is assumed, while multiplication operator (i.e., product rule named Product) is applied for change pixels stratum where cross-correlation is assumed negligible (i.e., minimum correlation), without having to run cross-validation cokriging as in PXCOV. Studies were undertaken to test the proposed method based on datasets collected previously concerning GlobeLand30 2000 and 2010 land-cover at five sites in China. For each site, five model-training samples (being mostly non-collocated) of equal sizes and one independent model-testing sample (collocated) were used. Logistic regression models fitted with relevant sample data were applied to predict local accuracies in single-date classifications using selected map class occurrence pattern indices quantified in optimized moving windows. The area under the curve (AUC) of the receiver operating characteristic was used for evaluating alternative methods. Empirical results confirmed that method Fuzzy+Product is more accurate than both Fuzzy and Product in general and there are no statistically significant differences between it and PXCOV. This indicates Fuzzy+Product being a method of relative simplicity but reasonable accuracy when reference data are non-collocated or mostly so. Its value is likely best manifested when local and global accuracy characterization in multi-temporal change information (discrete and fractional) is concerned.
Ground-level ozone has become the primary air pollutant in many urban areas of China. Oil vapor pollution from gasoline stations accelerates the generation of ground-level ozone, especially in densely populated urban areas with high demands for transportation. An accurate spatiotemporal distribution of ground-level ozone concentrations (GOCs) around gasoline stations is urgently needed. However, urban GOCs vary sharply over short distances, increasing the need for GOCs at a high-spatial resolution. Thus, a high-spatial resolution (i.e., 1 km) concentration retrieval model based on the GLM and BME method was developed to obtain the daily spatiotemporal characteristics of GOCs. The hourly ozone records provided by the national air quality monitoring stations and multiple geospatial datasets were used as input data. The model exhibited satisfactory performance (R2 = 0.75, RMSE = 10.86 µg/m3). The derived GOCs show that the ozone levels at gasoline stations and their adjacent areas (1~3 km away from the gasoline stations) were significantly higher than the citywide average level, and this phenomenon gradually eased with the increasing distance from the gasoline stations. The findings indicate that special attention should be given to the prevention and control of ground-level ozone exposure risks in human settlements and activity areas near gasoline stations.