Existing Land Cover Products Research Articles

Landsat-based classification in the cloud: An opportunity for a paradigm shift in land cover monitoring

Satellite-derived land cover maps play an important role in many applications, including monitoring of smallholder-dominated agricultural landscapes. New cloud-based computing platforms and satellite sensors offer opportunities for generating land cover maps designed to meet the spatial and temporal requirements of specific applications. Such maps can be a significant improvement compared to existing products, which tend to be coarser than 300m, are often not representative of areas with fast-paced land use change, and have a fixed set of cover classes. Here, we present two approaches for land cover classification using the Landsat archive within Google Earth Engine. Random forest classification was performed with (1) season-based composites, where median values of individual bands and vegetation indices were generated from four years for each of four seasons, and (2) metric-based composites, where different quantiles were computed for the entire four-year period. These approaches were tested for six land cover types spanning over 18,000 locations in Zambia, with ground “truth” determined by visual inspection of high-resolution imagery from Google Earth. The methods were trained on 30% of these points and tested on the remaining 70%, and results were also compared with existing land cover products. Overall accuracies of about 89% were achieved for the season- and metric-based approaches for individual classes, with 93% and 94% accuracy for distinguishing cropland from non-cropland. For the latter task, the existing Globeland30 dataset based on Landsat had much lower accuracies (around 77% on average), as did existing cover maps at coarser resolutions. Overall, the results support the use of either season or metric-based classification approaches. Both produce better results than those obtained from previous classifiers, which supports a general paradigm shift away from dependence on standard static products and towards custom generation of on-demand cover maps designed to fulfill the needs of each specific application.

Derivation of fractional urban signals in better capturing urbanization process

Rapid urbanization has changed land use and urban structure in China and therefore greatly modified land surface properties and land–atmosphere interactions, causing further local and regional climate change. Climate model simulation and urbanization process analysis are usually limited by poor accuracy of coarse-resolution land use/cover products employed in regional climate models. This study sought to identify better urban representation from Landsat images and monitor urban expansion by change detection of spatial patterns and urban fractions in southeastern coastal region of China. We used the improved normalized indices-based method to classify urban and built-up areas from Landsat images in Jiading District, Shanghai. Classification results were evaluated at both the pixel scale and the model grid scale, with overall accuracy of 88% and k coefficient of 0.76. Moreover, urbanization process over the Guangzhou–Foshan–Dongguan area was examined from 2000 to 2009. We aggregated the original results of urban classification data from Landsat images as fractional cover information in 1-km grids. The total fractional urban change in 2000–2005 (10.65%) was approximately three times greater than in 2005–2009 (3.38%). We also compared the fractional cover of urban expansion with the corresponding period of MODIS land cover products. It showed that existing land cover products in models had deviations and could not capture well the underlying conditions and urbanization process. Different fractional covers of urban scenarios were expected to provide better inputs for accurate modeling of critical environmental feedbacks over expanding urban clusters.

Integrating Crowdsourced Data with a Land Cover Product: A Bayesian Data Fusion Approach

For many environmental applications, an accurate spatial mapping of land cover is a major concern. Currently, land cover products derived from satellite data are expected to offer a fast and inexpensive way of mapping large areas. However, the quality of these products may also largely depend on the area under study. As a result, it is common that various products disagree with each other, and the assessment of their respective quality still relies on ground validation datasets. Recently, crowdsourced data have been suggested as an alternate source of information that might help overcome this problem. However, crowdsourced data still remain largely discarded in scientific studies due to their inherent poor quality assurance. The aim of this paper is to present an efficient methodology that allows the user to code information brought by crowdsourced data even if no prior quality estimation is at hand and possibly to fuse this information with existing land cover products in order to improve their accuracy. It is first suggested that information brought by volunteers can be coded as a set of inequality constraints about the probabilities of the various land use classes at the visited places. This in turn allows estimating optimal probabilities based on a maximum entropy principle and to proceed afterwards with a spatial interpolation of these volunteers’ information. Finally, a Bayesian data fusion approach can be used for fusing multiple volunteers’ contributions with a remotely-sensed land cover product. This methodology is illustrated in this paper by focusing on the mapping of croplands in Ethiopia, where the aim is to improve the mapping of cropland as coming out from a land cover product with mitigated performances. It is shown how crowdsourced information can seriously improve the quality of the final product. The corresponding results also suggest that a prior assessing of remotely-sensed data quality can seriously improve the benefit of crowdsourcing campaigns, so that both sources of information need to be accounted together in order to optimize the sampling efforts.

A MODIS approach to predicting stream water quality in Wisconsin

Efforts to predict water quality often rely heavily on static landuse/land-cover (LULC) classifications derived from remote sensing imagery. However, LULC classifications are infrequently updated, and the development of regular (annual) land cover maps that accurately capture intra- and inter-annual change may be expensive in terms of data and labor costs. In addition, existing land cover products may not include classes relevant to the assessment of water quality. Conversely, data from the Moderate Resolution Imaging Spectroradiometer (MODIS) instruments are freely available, preprocessed, frequently updated, and available in a range of useful data products. Our goal was to predict water quality for streams in Wisconsin, USA using a model constructed exclusively from MODIS products. We used streamwater nitrate and dissolved phosphorus data from 2001 to 2004 from two previous studies (Robertson et al., 2006; Stanley & Maxted, 2008) to evaluate the potential for using MODIS in an empirical model to predict water quality, thereby reducing reliance on possibly dated land cover maps. Using predictors derived exclusively from MODIS data products, we successfully predicted 80% of the variation in measured nitrate concentrations. Remote sensing based models explained 51% of the variation in dissolved phosphorus concentrations. Predictions of water quality were developed on both a per-pixel and a watershed basis. The remote sensing approach offers the potential for spatially continuous estimates of stream nutrient concentrations, and complements the suite of approaches currently used to assess water quality; including periodic stream surveys models, and point-source data.

Exploiting the Classification Performance of Support Vector Machines with Multi-Temporal Moderate-Resolution Imaging Spectroradiometer (MODIS) Data in Areas of Agreement and Disagreement of Existing Land Cover Products

Several studies have focused in the past on global land cover (LC) datasets harmonization and inter-comparison and have found significant inconsistencies. Despite the known discrepancies between existing products derived from medium resolution satellite sensor data, little emphasis has been placed on examining these disagreements to improve the overall classification accuracy of future land cover maps. This work evaluates the classification performance of a least square support vector machine (LS-SVM) algorithm with respect to areas of agreement and disagreement between two existing land cover maps. The approach involves the use of time series of Moderate-resolution Imaging Spectroradiometer (MODIS) 250-m Normalized Difference Vegetation Index (NDVI) (16-day composites) and gridded climatic indicators. LS-SVM is trained on reference samples obtained through visual interpretation of Google Earth (GE) high resolution imagery. The core of the training process is based on repeated random splits of the training dataset to select a small set of suitable support vectors optimizing class separability. A large number of independent validation samples spread over three contrasting regions in Europe (Eastern Austria, Macedonia and Southern France) are used to calculate classification accuracies for the LS-SVM NDVI-derived LC map and for two (globally available) LC products: GLC2000 and GlobCover. The LS-SVM LC map reported an overall accuracy of 70%. Classification accuracies ranged from 71% where GlobCover and GLC2000 agreed to 68% for areas of disagreement. Results indicate that existing LC products are as accurate as the LS-SVM LC map in areas of agreement (with little margin for improvements), while classification accuracy is substantially better for the LS-SVM LC map in areas of disagreement. On average, the LS-SVM LC map was 14% and 18% more accurate compared to GlobCover and GLC2000, respectively.

Cropland for sub-Saharan Africa: A synergistic approach using five land cover data sets

This paper presents a methodology for the creation of a cropland map for Africa through the combination of five existing land cover products: GLC-2000, MODIS Land Cover, GlobCover, MODIS Crop Likelihood and AfriCover. A synergy map is created in which the products are ranked by experts, which reflects the likelihood or probability that a given pixel is cropland. The cropland map is then calibrated with national and sub-national crop statistics using a novel approach. Preliminary validation of the map was undertaken and the results are presented. The resulting cropland map has an accuracy of 83%, which is higher than the accuracy of any of the individual maps. The cropland map is freely available at agriculture.geo-wiki.org.

Use of a dark object concept and support vector machines to automate forest cover change analysis

An automated method was developed for mapping forest cover change using satellite remote sensing data sets. This multi-temporal classification method consists of a training data automation (TDA) procedure and uses the advanced support vector machines (SVM) algorithm. The TDA procedure automatically generates training data using input satellite images and existing land cover products. The derived high quality training data allow the SVM to produce reliable forest cover change products. This approach was tested in 19 study areas selected from major forest biomes across the globe. In each area a forest cover change map was produced using a pair of Landsat images acquired around 1990 and 2000. High resolution IKONOS images and independently developed reference data sets were available for evaluating the derived change products in 7 of those areas. The overall accuracy values were over 90% for 5 areas, and were 89.4% and 89.6% for the remaining two areas. The user's and producer's accuracies of the forest loss class were over 80% for all 7 study areas, demonstrating that this method is especially effective for mapping major disturbances with low commission errors. IKONOS images were also available in the remaining 12 study areas but they were either located in non-forest areas or in forest areas that did not experience forest cover change between 1990 and 2000. For those areas the IKONOS images were used to assist visual interpretation of the Landsat images in assessing the derived change products. This visual assessment revealed that for most of those areas the derived change products likely were as reliable as those in the 7 areas where accuracy assessment was conducted. The results also suggest that images acquired during leaf-off seasons should not be used in forest cover change analysis in areas where deciduous forests exist. Being highly automatic and with demonstrated capability to produce reliable change products, the TDA-SVM method should be especially useful for quantifying forest cover change over large areas.