Land Cover Classes Of Interest Research Articles

Using image segmentation models to analyse high-resolution earth observation data: new tools to monitor disease risks in changing environments.

In the near future, the incidence of mosquito-borne diseases may expand to new sites due to changes in temperature and rainfall patterns caused by climate change. Therefore, there is a need to use recent technological advances to improve vector surveillance methodologies. Unoccupied Aerial Vehicles (UAVs), often called drones, have been used to collect high-resolution imagery to map detailed information on mosquito habitats and direct control measures to specific areas. Supervised classification approaches have been largely used to automatically detect vector habitats. However, manual data labelling for model training limits their use for rapid responses. Open-source foundation models such as the Meta AI Segment Anything Model (SAM) can facilitate the manual digitalization of high-resolution images. This pre-trained model can assist in extracting features of interest in a diverse range of images. Here, we evaluated the performance of SAM through the Samgeo package, a Python-based wrapper for geospatial data, as it has not been applied to analyse remote sensing images for epidemiological studies. We tested the identification of two land cover classes of interest: water bodies and human settlements, using different UAV acquired imagery across five malaria-endemic areas in Africa, South America, and Southeast Asia. We employed manually placed point prompts and text prompts associated with specific classes of interest to guide the image segmentation and assessed the performance in the different geographic contexts. An average Dice coefficient value of 0.67 was obtained for buildings segmentation and 0.73 for water bodies using point prompts. Regarding the use of text prompts, the highest Dice coefficient value reached 0.72 for buildings and 0.70 for water bodies. Nevertheless, the performance was closely dependent on each object, landscape characteristics and selected words, resulting in varying performance. Recent models such as SAM can potentially assist manual digitalization of imagery by vector control programs, quickly identifying key features when surveying an area of interest. However, accurate segmentation still requires user-provided manual prompts and corrections to obtain precise segmentation. Further evaluations are necessary, especially for applications in rural areas.

Developing and applying a multi-purpose land cover validation dataset for Africa

The production of global land cover products has accelerated significantly over the past decade thanks to the availability of higher spatial and temporal resolution satellite data and increased computation capabilities. The quality of these products should be assessed according to internationally promoted requirements e.g., by the Committee on Earth Observation Systems-Working Group on Calibration and Validation (CEOS-WGCV) and updated accuracy should be provided with new releases (Stage-4 validation). Providing updated accuracies for the yearly maps would require considerable effort for collecting validation datasets. To save time and effort on data collection, validation datasets should be designed to suit multiple map assessments and should be easily adjustable for a timely validation of new releases of land cover products. This study introduces a validation dataset aimed to facilitate multi-purpose assessments and its applicability is demonstrated in three different assessments focusing on validating discrete and fractional land cover maps, map comparison and user-oriented map assessments. The validation dataset is generated primarily to validate the newly released 100 m spatial resolution land cover product from the Copernicus Global Land Service (CGLS-LC100). The validation dataset includes 3617 sample sites in Africa based on stratified sampling. Each site corresponds to an area of 100 m × 100 m. Within site, reference land cover information was collected at 100 subpixels of 10 m × 10 m allowing the land cover information to be suitable for different resolution and legends. Firstly, using this dataset, we validated both the discrete and fractional land cover layers of the CGLS-LC100 product. The CGLS-LC100 discrete map was found to have an overall accuracy of 74.6 ± 2.1% (at 95% confidence level) for the African continent. Fraction cover products were found to have mean absolute errors of 9.3, 8.8, 16.2, and 6.5% for trees, shrubs, herbaceous vegetation and bare ground, respectively. Secondly, for user-oriented map assessment, we assessed the accuracy of the CGLS-LC100 map from four user groups' perspectives (forest monitoring, crop monitoring, biodiversity and climate modelling). Overall accuracies for these perspectives vary between 73.7% ± 2.1% and 93.5% ± 0.9%, depending on the land cover classes of interest. Thirdly, for map comparison, we assessed the accuracy of the Globeland30-2010 map at 30 m spatial resolution. Using the subpixel level validation data, we derived 15,252 sample pixels at 30 m spatial resolution. Based on these sample pixels, the overall accuracy of the Globeland30-2010 map was found to be 66.6 ± 2.4% for Africa. The three assessments exemplify the applicability of multi-purpose validation datasets which are recommended to increase map validation efficiency and consistency. Assessments of subsequent yearly maps can be conducted by augmenting or updating the dataset with sample sites in identified change areas.

Multi-Scale Analysis of Very High Resolution Satellite Images Using Unsupervised Techniques

This article is concerned with the use of unsupervised methods to process very high resolution satellite images with minimal or little human intervention. In a context where more and more complex and very high resolution satellite images are available, it has become increasingly difficult to propose learning sets for supervised algorithms to process such data and even more complicated to process them manually. Within this context, in this article we propose a fully unsupervised step by step method to process very high resolution images, making it possible to link clusters to the land cover classes of interest. For each step, we discuss the various challenges and state of the art algorithms to make the full process as efficient as possible. In particular, one of the main contributions of this article comes in the form of a multi-scale analysis clustering algorithm that we use during the processing of the image segments. Our proposed methods are tested on a very high resolution image (Pléiades) of the urban area around the French city of Strasbourg and show relevant results at each step of the process.

Wetland and landscape indices for assessing the condition of semiarid Mediterranean saline wetlands under agricultural hydrological pressures

During last decades semiarid Mediterranean saline wetlands have undergone several hydrological and biological changes as a consequence of increased water inputs from agricultural areas. Specific indices are needed in order to assess the condition of these unique ecosystems in relation to major hydrological disturbances at watershed level. Through the long-term study of selected plant taxa in a set of representative wetlands in Murcia province (SE Spain), together with the characterization of their watershed agricultural land uses, plant indicators of wetland condition were sought and then combined into a wetland condition index. Percentages of land cover classes of interest were weighted taking into account land cover arrangement and receiving wetland size. Characteristic perennial plant taxa were sampled in 1989 and 2008 and significant taxa frequency changes at each wetland site were determined. Regression analysis was used to relate wetland plant taxa frequency and watershed condition during the study period. Limonium spp., Arthrocnemum glaucum, Phragmites australis, Tamarix canariensis and Atriplex halimus showed significant relationships with watershed condition. Indicator taxa were thus identified and their frequencies were combined into an integrated index of wetland condition, which showed a robust relationship with watershed hydrological condition.

Image segmentation scale parameter optimization and land cover classification using the Random Forest algorithm

This paper describes an approach to using the Random Forest classification algorithm to quantitatively evaluate a range of potential image segmentation scale alternatives in order to identify the segmentation scale(s) that best predict land cover classes of interest. The image segmentation scale selection process was used to identify three critical image object scales that when combined produced an optimal level of land cover classification accuracy. Following segmentation scale optimization, the Random Forest classifier was then used to assign land cover classes to 11 scenes of SPOT satellite imagery in North and South Dakota with an average overall accuracy of 85.2 percent.

Partial unmixing as a tool for single surface class detection and time series analysis

In this paper we present the results of time series analysis for a coal mining region based on partial unmixing. We test the method also known as mixture tuned matched filtering on an eight image Landsat 5 TM and Landsat 7 ETM+ time series covering the period from 1987 to 2003. Common change detection methods often include the comparison of two interactively generated classification results, such as derived from Maximum Likelihood classification. These approaches often yield highly accurate results. However, disadvantages include a strong analyst bias and hardly repeatable results. For a quantitative monitoring of a single surface class' development over time they are often not recommendable. Our goal was to test an unbiased quantitative way to assess the development of coal surfaces, such as outcropping coal seams, coal storage piles, coal waste piles, and coal washery discard, within multiple date satellite imagery. Partial unmixing approaches were developed to detect one or few target materials surrounded by—or mixed with—an unknown background material. The main advantage is that only the spectral characteristics of the material of interest must be known, and the desired material can furthermore occur with subpixel coverage. Crisp pixel classificators like maximum likelihood on the other hand require knowledge of all classes. They can only map materials which dominate a pixel. Linear unmixing procedures such as partial unmixing require a thorough radiometric pre‐processing of data. Furthermore, the accuracy and representativity of selected input spectra must be granted. In this paper we demonstrate that partial unmixing is a powerful method to detect and extract single landcover classes of interest relatively fast and unbiased. The subpixel fraction percentages should be interpreted in a relative way only. We furthermore show that partial unmixing represents a standardized method for time series analyses and allows for a quantitative assessment of the temporal development of an area. Challenges lie in the validation of partial unmixing results, which we realized through thresholding of unmixing results and accuracy assessment with ground truth polygons mapped in situ. Furthermore, we performed an indirect comparison with results of a multi‐endmember unmixing.

The use of decision tree and multiscale texture for classification of JERS-1 SAR data over tropical forest

The objective of this paper is to study the use of a decision tree classifier and multiscale texture measures to extract thematic information on the tropical vegetation cover from the Global Rain Forest Mapping (GRFM) JERS-1 SAR mosaics. The authors focus their study on a coastal region of Gabon, which has a variety of land cover types common to most tropical regions. A decision tree classifier does not assume a particular probability density distribution of the input data, and is thus well adapted for SAR image classification. A total of seven features, including wavelet-based multiscale texture measures (at scales of 200, 400, and 800 m) and multiscale multitemporal amplitude data (two dates at scales 100 and 400 m), are used to discriminate the land cover classes of interest. Among these layers, the best features for separating classes are found by constructing exploratory decision trees from various feature combinations. The decision tree structure stability is then investigated by interchanging the role of the training samples for decision tree growth and testing. They show that the construction of exploratory decision trees can improve the classification results. The analysis also proves that the radar backscatter amplitude is important for separating basic land cover categories such as savannas, forests, and flooded vegetation. Texture is found to be useful for refining flooded vegetation classes. Temporal information from SAR images of two different dates is explicitly used in the decision tree structure to identify swamps and temporarily flooded vegetation.