Fernando J. Aguilar

Assessing geometric accuracy of the orthorectification process from GeoEye-1 and WorldView-2 panchromatic images

Abstract GeoEye-1 and WorldView-2 are the commercial very high resolution (VHR) satellites more innovative, unexplored and presenting the highest available resolutions nowadays. The attainable geopositioning accuracies from GeoEye-1 and WorldView-2 single panchromatic images, both along the sensor orientation and orthorectification phases, are analyzed at the same study area and by using exactly the same ancillary data. The accuracy assessment was carried out depending on the following factors: (i) type of input satellite image (GeoEye-1 Geo, WorldView-2 Ortho Ready Standard and WorldView-2 Basic), (ii) sensor orientation model used (rigorous and based on rational function), (iii) number of well-distributed ground control points (GCPs) used in the triangulation process, (iv) off-nadir viewing angle, and finally (v) vertical accuracy of the DEM employed to conduct the orthorectification process. Regardless of satellite or product, the best horizontal geopositioning accuracies were always attained by using third order 3D rational functions with vendors rational polynomial coefficients data refined by a zero order polynomial adjustment (RPC0). Focusing on WorldView-2 products, worse accuracies were yielded from Basic images than from Ortho Ready Standard level ones. As a general rule, and for attaining sub-pixel planimetric accuracies for the orthorectified GeoEye-1 Geo and WorldView-2 Ortho Ready Standard images and using RPC0 model with 7 GCPs, users should avoid off-nadir angles higher than 20° and use a very accurate DEM.

The accuracy of grid digital elevation models linearly constructed from scattered sample data

In this paper, a theoretical‐empirical model is developed for modelling the accuracy of a grid digital elevation model (DEM) linearly constructed from scattered sample data. The theoretical component integrates sample data accuracy in the model by means of the error‐propagation theory. The empirical component seeks to model what is known as information loss, i.e. the sum of the error due purely to sampling the continuous terrain surface with a finite grid interval and the interpolation error. For this purpose, randomly spaced data points, supposed to be free of error, were converted into regularly gridded data points using triangulation with linear interpolation. Original sample data were collected with a 2×2 m sampling interval from eight different morphologies, from flat terrain to highly rugged terrain, applying digital photogrammetric methods to large‐scale aerial stereo imagery (1 : 5000). The DEM root mean square error was calculated by the true validation method over several sets of check points, obtaining the different sampling densities tested in this work. Several empirical models are calibrated and validated with the experimental data set by modelling the DEM accuracy by combining two variables such as sampling density and a descriptive attribute of terrain morphology. These empirical models presented a morphology based on the product of two potential functions, one related to the terrain roughness and another related to the sampling density. The terrain descriptors tested were average terrain slope, standard deviation of terrain slope, standard deviation of unitary vectors perpendicular to the topographic surface (SDUV), standard deviation of the difference in height between adjacent samples in the grid DEM (SDHD), and roughness estimation by first‐, second‐, or third‐degree surface fitting error. The values obtained for those terrain descriptors were reasonably independent from the number and spatial distribution of the sample data. The models based on descriptors SDHD, SDUV, and standard deviation of slope provided a good fitting to the data observed (R 2>0.94) in the calibration phase, model SDHD being the one that yielded the best results in validation. Therefore, it would be possible to establish a priori the optimum grid size required to generate or store a DEM of a particular accuracy, with the saving in computing time and file size that this would mean for the digital flow of the mapping information in GIS.

Generation and Quality Assessment of Stereo-Extracted DSM From GeoEye-1 and WorldView-2 Imagery

Digital surface models (DSMs) extracted from 15 different stereo pairs attained by the combination of GeoEye-1 (GE1) and WorldView-2 (WV2) panchromatic very high resolution (VHR) satellite images are tested. Two of them are pure same-date along-track stereo pairs, one from each VHR satellite, whereas the rest are mixed multidate across-track ones. A quality assessment on the DSMs extracted from the aforementioned stereo pairs, involving both accuracy and completeness, is carried out. Several factors are tested such as sensor model used in the bundle adjustment, number of ground control points (GCPs), radiometric characteristics, satellite imaging geometry, time between acquisition dates, and target land cover. A highly accurate light detection and ranging elevation data is used as ground truth. Overall, 3-D rational functions refined by a zero-order polynomial adjustment by using 7 or 12 GCPs performed slightly better regarding both DSM vertical accuracy and completeness. In relation to the pure stereo pairs, the DSM extracted from the GE1 stereo pair attained better vertical accuracy over the whole study area (90th percentile linear error, LE90, of 2.04 m) but lower completeness (74.50%) than the WV2 one (2.56 m and 83.35%, respectively). The undergoing hypothesis is that the blurrier images from WV2 could have influenced in the improvement of the matching success rate while reducing the vertical accuracy of extracted points. When all the 15 stereo pairs are considered, the vertical accuracy mainly depends on the convergence angle. In addition, the temporal difference between acquisition dates turned out to be the most influential factor regarding completeness values.

Detecting greenhouse changes from QuickBird imagery on the Mediterranean coast

In this study a very high resolution image from the QuickBird satellite was used to detect new greenhouses built since the last update of the information system utilized for the study area. The area, located in the southeast of Spain, has the highest concentration of greenhouses in Europe, which makes it the heart of the economy of this region. The methodology proposed in this paper is based on the comparison of the classification of a current image with the information system corresponding to the last update. Maximum likelihood classification method was employed and different band combinations were used to define the training areas and to carry out the classification process. The optimal band combination for the detection of greenhouses was calculated by means of a variance analysis. The process was completed with the delineation of the new greenhouses with two algorithms programmed in Visual Basic 6.0, one to eliminate the loops shown around the greenhouses detected, and the other one, based on the Hough transformation, to delineate the contour of the polygons corresponding to the new greenhouses. The proposed methodology achieved (1) a value for true greenhouse surface of about 91.45% of the whole surface, (2) a very low value for undetected greenhouses (five greenhouses from a total of 202 that were built, representing 1.49% of the surface of new greenhouses), and (3) a low number of pixels wrongly classified as greenhouses.

Performance evaluation of object based greenhouse detection from Sentinel-2 MSI and Landsat 8 OLI data: A case study from Almería (Spain)

Abstract This paper shows the first comparison between data from Sentinel-2 (S2) Multi Spectral Instrument (MSI) and Landsat 8 (L8) Operational Land Imager (OLI) headed up to greenhouse detection. Two closely related in time scenes, one for each sensor, were classified by using Object Based Image Analysis and Random Forest (RF). The RF input consisted of several object-based features computed from spectral bands and including mean values, spectral indices and textural features. S2 and L8 data comparisons were also extended using a common segmentation dataset extracted form VHR World-View 2 (WV2) imagery to test differences only due to their specific spectral contribution. The best band combinations to perform segmentation were found through a modified version of the Euclidian Distance 2 index. Four different RF classifications schemes were considered achieving 89.1%, 91.3%, 90.9% and 93.4% as the best overall accuracies respectively, evaluated over the whole study area.

Geometric accuracy assessment of the orthorectification process from very high resolution satellite imagery for Common Agricultural Policy purposes

This study has, as its main aim, the assessment of different sensor models to achieve the best geometric accuracy in orthorectified imagery products obtained from IKONOS Geo Ortho Kit and QuickBird basic imagery. The final orthoimages are compared, both geometrically and visually, with the panchromatic orthophotos based on a photogrammetric flight with an approximate scale of 1 : 20 000, which are now used for the European Union Common Agricultural Policy in Andalusia (Spain). Two‐dimensional root mean square (RMS2d) errors in independent check points are used as accuracy indicators. The ancillary data were generated by high accuracy methods: (1) check and ground control points (GCPs) were measured with a differential global positioning system and (2) an accurate digital elevation model was used for image orthorectification. Two sensor models were used to correct the satellite data: (1) a three‐dimensional (3D) rational function refined by the user with zero‐ (RPC0) or first‐(RPC1) order polynomial adjustment and (2) the 3D Toutin physical model (CCRS). For the IKONOS image, the best results in the final orthoimages (RMS2d of about 1.15 m) were obtained when the RPC0 model was used. Neither a large number of GCPs (more than nine), nor a better distribution of them, improved the results obtained with the RPC0. For the QuickBird image, the CCRS model generated the best results (RMS2d of about 1.04 m), although it was sensitive to the number and distribution of the GCPs used in its computation.

Accuracy assessment of digital elevation models using a non-parametric approach

This paper explores three theoretical approaches for estimating the degree of correctness to which the accuracy figures of a gridded Digital Elevation Model (DEM) have been estimated depending on the number of checkpoints involved in the assessment process. The widely used average‐error statistic Mean Square Error (MSE) was selected for measuring the DEM accuracy. The work was focused on DEM uncertainty assessment using approximate confidence intervals. Those confidence intervals were constructed both from classical methods which assume a normal distribution of the error and from a new method based on a non‐parametric approach. The first two approaches studied, called Chi‐squared and Asymptotic Student t, consider a normal distribution of the residuals. That is especially true in the first case. The second case, due to the asymptotic properties of the t distribution, can perform reasonably well with even slightly non‐normal residuals if the sample size is large enough. The third approach developed in this article is a new method based on the theory of estimating functions which could be considered much more general than the previous two cases. It is based on a non‐parametric approach where no particular distribution is assumed. Thus, we can avoid the strong assumption of distribution normality accepted in previous work and in the majority of current standards of positional accuracy. The three approaches were tested using Monte Carlo simulation for several populations of residuals generated from originally sampled data. Those original grid DEMs, considered as ground data, were collected by means of digital photogrammetric methods from seven areas displaying differing morphology employing a 2 by 2 m sampling interval. The original grid DEMs were subsampled to generate new lower‐resolution DEMs. Each of these new DEMs was then interpolated to retrieve its original resolution using two different procedures. Height differences between original and interpolated grid DEMs were calculated to obtain residual populations. One interpolation procedure resulted in slightly non‐normal residual populations, whereas the other produced very non‐normal residuals with frequent outliers. Monte Carlo simulations allow us to report that the estimating function approach was the most robust and general of those tested. In fact, the other two approaches, especially the Chi‐squared method, were clearly affected by the degree of normality of the residual population distribution, producing less reliable results than the estimating functions approach. This last method shows good results when applied to the different datasets, even in the case of more leptokurtic populations. In the worst cases, no more than 64–128 checkpoints were required to construct an estimate of the global error of the DEM with 95% confidence. The approach therefore is an important step towards saving time and money in the evaluation of DEM accuracy using a single average‐error statistic. Nevertheless, we must take into account that MSE is essentially a single global measure of deviations, and thus incapable of characterizing the spatial variations of errors over the interpolated surface.

Geopositioning Accuracy Assessment of GeoEye-1 Panchromatic and Multispectral Imagery

Currently GeoEye-1 is the World’s highest resolution commercial satellite. This paper analyses the attainable geopositioning accuracy from a single GeoEye-1 Geo image, both through the sensor orientation and orthorectification phases, for panchromatic ( PAN ) and multispectral ( MS ) products. Different 3D sensor models as well as the number and distribution of the ground control points ( GCP s) used for the sensor orientation were tested. Planimetric Root Mean Square Errors ( RMSE 2D ) close to 0.7 pixels, both for PAN and MS images, were attained using the third order 3D rational functions with the vendor’s rational polynomial coefficients data afterwards being refined by a zero order polynomial adjustment ( RPC 0). Furthermore, the RPC 0 sensor model proved to be significantly independent regarding the number and distribution of the GCP s. The RPC 0 model yielded RMSE 2D close to 0.46 m and 1.56 m for the PAN and MS orthorectified images, respectively, using a very accurate lidar-derived digital elevation model.

Object-Based Greenhouse Classification from GeoEye-1 and WorldView-2 Stereo Imagery

Remote sensing technologies have been commonly used to perform greenhouse detection and mapping. In this research, stereo pairs acquired by very high-resolution optical satellites GeoEye-1 (GE1) and WorldView-2 (WV2) have been utilized to carry out the land cover classification of an agricultural area through an object-based image analysis approach, paying special attention to greenhouses extraction. The main novelty of this work lies in the joint use of single-source stereo-photogrammetrically derived heights and multispectral information from both panchromatic and pan-sharpened orthoimages. The main features tested in this research can be grouped into different categories, such as basic spectral information, elevation data (normalized digital surface model; nDSM), band indexes and ratios, texture and shape geometry. Furthermore, spectral information was based on both single orthoimages and multiangle orthoimages. The overall accuracy attained by applying nearest neighbor and support vector machine classifiers to the four multispectral bands of GE1 were very similar to those computed from WV2, for either four or eight multispectral bands. Height data, in the form of nDSM, were the most important feature for greenhouse classification. The best overall accuracy values were close to 90%, and they were not improved by using multiangle orthoimages.

Relationship between atmospheric corrections and training-site strategy with respect to accuracy of greenhouse detection process from very high resolution imagery

Frequently, satellite images that are acquired to extract a target surface are atmospherically corrected prior to the detection process. Thus, the unification of measure units is achieved, and atmospheric effects are removed from various imagery sources or taken at different dates. In this paper, four increasing levels of atmospheric corrections are applied (Top-Of-Atmosphere transformation: TOA; Apparent Reflectance Model: ARM; Flat Areas Model: FAM; Non-Flat Areas Model: NFAM). Then, the classification process is carried out using two strategies of training-site definitions (statistically purified and crude training sites) and two satellite imagery sources (QuickBird and Ikonos). Three-way Analysis of Variance (ANOVA) tests and Fishers least-significant difference tests are included in quality classification assessment, based on four accuracy indexes. Two images from both remote sensors are orthorectified, and then it is checked that all selected atmospheric correction levels have significantly different influences on the statistics of both orthoimages. Taking into account the conditions established in this work, it is concluded that a lower atmospheric correction level would be preferred because it does not present significantly worse results than other levels considered. Training sites would not be statistically purified, and QuickBird or Ikonos would be chosen, depending on the aspect of the greenhouse detection accuracy preferred.