Francisco Alonso-Sarría

Optimal Combination of Classification Algorithms and Feature Ranking Methods for Object-Based Classification of Submeter Resolution Z/I-Imaging DMC Imagery

Object-based image analysis allows several different features to be calculated for the resulting objects. However, a large number of features means longer computing times and might even result in a loss of classification accuracy. In this study, we use four feature ranking methods (maximum correlation, average correlation, Jeffries–Matusita distance and mean decrease in the Gini index) and five classification algorithms (linear discriminant analysis, naive Bayes, weighted k-nearest neighbors, support vector machines and random forest). The objective is to discover the optimal algorithm and feature subset to maximize accuracy when classifying a set of 1,076,937 objects, produced by the prior segmentation of a 0.45-m resolution multispectral image, with 356 features calculated on each object. The study area is both large (9070 ha) and diverse, which increases the possibility to generalize the results. The mean decrease in the Gini index was found to be the feature ranking method that provided highest accuracy for all of the classification algorithms. In addition, support vector machines and random forest obtained the highest accuracy in the classification, both using their default parameters. This is a useful result that could be taken into account in the processing of high-resolution images in large and diverse areas to obtain a land cover classification.

Modification of the random forest algorithm to avoid statistical dependence problems when classifying remote sensing imagery

Abstract Random forest is a classification technique widely used in remote sensing. One of its advantages is that it produces an estimation of classification accuracy based on the so called out-of-bag cross-validation method. It is usually assumed that such estimation is not biased and may be used instead of validation based on an external data-set or a cross-validation external to the algorithm. In this paper we show that this is not necessarily the case when classifying remote sensing imagery using training areas with several pixels or objects. According to our results, out-of-bag cross-validation clearly overestimates accuracy, both overall and per class. The reason is that, in a training patch, pixels or objects are not independent (from a statistical point of view) of each other; however, they are split by bootstrapping into in-bag and out-of-bag as if they were really independent. We believe that putting whole patch, rather than pixels/objects, in one or the other set would produce a less biased out-of-bag cross-validation. To deal with the problem, we propose a modification of the random forest algorithm to split training patches instead of the pixels (or objects) that compose them. This modified algorithm does not overestimate accuracy and has no lower predictive capability than the original. When its results are validated with an external data-set, the accuracy is not different from that obtained with the original algorithm. We analysed three remote sensing images with different classification approaches (pixel and object based); in the three cases reported, the modification we propose produces a less biased accuracy estimation.

Improving Classification Accuracy of Multi-Temporal Landsat Images by Assessing the Use of Different Algorithms, Textural and Ancillary Information for a Mediterranean Semiarid Area from 2000 to 2015

The aim of this study was to evaluate three different strategies to improve classification accuracy in a highly fragmented semiarid area using, (i) different classification algorithms with parameter optimization in some cases; (ii) different feature sets including spectral, textural and terrain features; and (iii) different seasonal combinations of images. A three-way ANOVA was used to discern which of these approaches and their interactions significantly increases accuracy. Tukey–Kramer contrast using a heteroscedasticity-consistent estimation of the kappa covariances matrix was used to check for significant differences in accuracy. The experiment was carried out with Landsat TM, ETM and OLI images corresponding to the period 2000–2015. A combination of four images using random forest and the three feature sets was the best way to improve accuracy. Maximum likelihood, random forest and support vector machines do not significantly increase accuracy when textural information was added, but do so when terrain features were taken into account. On the other hand, sequential maximum a posteriori increased accuracy when textural features were used, but reduced accuracy substantially when terrain features were included. Random forest using the three feature subsets and sequential maximum a posteriori with spectral and textural features had the largest kappa values, around 0.9.

An open-source web mapping tool to estimate wind energy in the Iberian Peninsula

Abstract The main disadvantage of wind energy is its high spatial and temporal variability. This paper presents a web mapping tool to communicate to private users both the available wind resource and information to evaluate the suitability of several types of turbine for any point on the Iberian Peninsula. This tool performs on-the-fly three-dimensional interpolation of wind data from a 10 km horizontal grid database previously obtained using a regional climate model and generates a PDF report. It integrates several open-source GIS applications to build a coherent platform that performs advanced calculations and provides graphics and reports of high quality.

Calibration and spatial modelling of daily ET0 in semiarid areas using Hargreaves equation

Evapotranspiration is difficult to measure and, when measured, its spatial variability is not usually taken into account. The recommended method to estimate evapotranspiration, Penman-Monteith FAO, requires variables not available in most weather stations. Simplified but less accurate methods, as Hargreaves equation, are normally used. Several approaches have been proposed to improve Hargreaves equation accuracy. In this work, 14 calibrations of the Hargreaves equation are compared. Three goodness of fit statistics were used to select the optimal, in terms of simplicity and accuracy. The best option was an annual linear regression. Its parameters were interpolated using regression-kriging combining Random Forest and Ordinary Kriging. Twelve easy to obtain ancillary variables were used as predictors. The same approach was used to interpolate Hargreaves and Penman-Monteith-FAO ET0 on a daily basis; the Hargreaves ET0 layers and the parameter layers were used to obtain calibrated ET0 estimations. To compare the spatial patterns of the three estimations the daily layers were integrated into annual layers. The results of the proposed calibration are much more similar to Penman-Monteith FAO results than those obtained with Hargreaves equation. The research was conducted in south-east Spain with 79 weather stations with data from 01/01/2003 to 31/12/2014.

A new approach to the openness index for landform characterisation

Abstract Openness is a multi-scale geomorphometric feature that has not been widely used despite its potential. The original approach, which averages zenith and nadir angles in the eight main compass directions, is modified to take into account openness in all available directions; in addition, openness is calculated in different directions and different scales. A statistical analysis and Random Forest classification are carried out to check whether the modifications introduced provide significantly different results from those of the original approach. In addition, it was tested whether multi-scale and multi-direction openness provide relevant and complementary information to total openness. The results show that the original algorithm produces biased, systematically higher, openness estimations. In addition, multi-scale and multi-direction openness produce more accurate Random Forest classifications. Accuracy increases from 0.62 when using total openness to 0.66 when using the multi-scale approach, 0.73 when using the multi-direction approach and 0.75 when both are used.

Implementing and validating of pan-sharpening algorithms in open-source software

Several approaches have been used in remote sensing to integrate images with different spectral and spatial resolutions in order to obtain fused enhanced images. The objective of this research is three-fold. To implement in R three image fusion techniques (High Pass Filter, Principal Component Analysis and Gram-Schmidt); to apply these techniques to merging multispectral and panchromatic images from five different images with different spatial resolutions; finally, to evaluate the results using the universal image quality index (Q index) and the ERGAS index. As regards qualitative analysis, Landsat-7 and Landsat-8 show greater colour distortion with the three pansharpening methods, although the results for the other images were better. Q index revealed that HPF fusion performs better for the QuickBird, IKONOS and Landsat-7 images, followed by GS fusion; whereas in the case of Landsat-8 and Natmur-08 images, the results were more even. Regarding the ERGAS spatial index, the ACP algorithm performed better for the QuickBird, IKONOS, Landsat-7 and Natmur-08 images, followed closely by the GS algorithm. Only for the Landsat-8 image did, the GS fusion present the best result. In the evaluation of spectral components, HPF results tended to be better and ACP results worse, the opposite was the case with the spatial components. Better quantitative results are obtained in Landsat-7 and Landsat-8 images with the three fusion methods than with the QuickBird, IKONOS and Natmur-08 images. This contrasts with the qualitative evaluation reflecting the importance of splitting the two evaluation approaches (qualitative and quantitative). Significant disagreement may arise when different methodologies are used to asses the quality of an image fusion. Moreover, it is not possible to designate, a priori, a given algorithm as the best, not only because of the different characteristics of the sensors, but also because of the different atmospherics conditions or peculiarities of the different study areas, among other reasons.

Land-use and Flood Risk Changes in Coastal Areas of South-eastern Spain

Flash floods are a major natural hazard in the southeast of Spain. A semi-arid climate along with severe droughts with extreme rainfall events are the cause for a scarce vegetation cover. For geological reasons, the steep slopes near the sea have generated this situation (Romero Diaz/Maurandi Guirado 2000; Camarasa Belmonte 2002). Human occupation in this area was scarce and the land use limited to some dry land cultivations. But during the last 25 years, two new trends have increased the human risk from natural flood hazards.