Land Cover Dataset Research Articles

Wetland changes in the Amur River Basin: Differing trends and proximate causes on the Chinese and Russian sides

According to the United Nations Sustainable Development Goals (SDGs), understanding the extent of wetlands, their change trends and the proximate causes is important for the conservation of wetlands and endangered waterfowls. Here we studied the world's ninth largest river basin, the Amur River Basin (ARB), with a land area of 2.08 million km2. Our objectives were to address the information deficiencies of spatially explicit wetland distributions and their changes and to quantify the proximate causes of these changes in various periods in the ARB. A hybrid approach combining object-based and hierarchical decision-trees classification (HOHC) was applied to Landsat series images to obtain multitemporal land cover datasets from 1980 to 2016. Further quantitative analysis revealed that the ARB held 184,561 km2 of wetlands in 2016, accounting for 9% of the whole basin area. Among these, 59% of the wetlands were identified on the Russian side, while 40% were on the Chinese side, and 1% were on the Mongolian side. The ARB lost 22% of its wetland (52,246 km2) from 1980 to 2016, with a consistent net loss from 1980 to 2010 but an area gain from 2010 to 2016. Human activities dominated the consistent wetland losses on the Chinese side of the ARB, of which cropland expansion was the primary proximate cause of wetland loss (69%). Conversely, the wetlands on the Russian side had consistent losses from 1980 to 2010 followed by a gain from 2010 to 2016, which could be attributed to climate change. These quantified data will inform decision-making on wetland conservation and benefit scientific studies depending on spatially explicit wetland information.

Effects of using different urban parametrization schemes and land-cover datasets on the accuracy of WRF model over the City of Ottawa

In the face of rapid urbanization and global warming, it is important to acquire a better understanding of urban climate and land-atmosphere interactions operating therein. The Weather Research and Forecasting (WRF) model is a limited area model that has been used to study urban microclimate in many cities across the globe. However, such a study is lacking for the Canada's capital city: Ottawa. In this article, the WRF model is set-up at 1 km spatial resolution over the Ottawa region and its sensitivity towards the use of different urban parametrization schemes and land-cover datasets is investigated from 01 June to 31 August 2018 which includes an extreme heat weather event spanning from June 30 to July 6. WRF-simulations are performed using the Noah land surface model with two urban parametrization schemes of different complexity, i.e., a simple bulk urban parametrization and a more advanced multilayer urban canopy model (UCM). Both WRF-simulations used the default land use-land cover data from the Moderate Resolution Imaging Spectroradiometer (MODIS) available in WRF. Between these two WRF-simulations, the simulation with the multilayer UCM is found to be more accurate than the bulk scheme simulation in terms of modeled near-surface wind-speed, relative humidity, and total precipitation compared to observations recorded at one urban weather gauging station located within the city limits. Finally, a third WRF-simulation is performed with the multilayer UCM but using a higher resolution 30 m land use-land cover data product for the urban domain. This third WRF-experiment improved further the near-surface wind speed, relative humidity, and total precipitation correspondence to observations within the city. It is worthy to mention that modeled total precipitation is found to be sensitive to both urban parametrization schemes and urban land use-land cover data sets.

Feature selection to increase the random forest method performance on high dimensional data

Random Forest is a supervised classification method based on bagging (Bootstrap aggregating) Breiman and random selection of features. The choice of features randomly assigned to the Random Forest makes it possible that the selected feature is not necessarily informative. So it is necessary to select features in the Random Forest. The purpose of choosing this feature is to select an optimal subset of features that contain valuable information in the hope of accelerating the performance of the Random Forest method. Mainly for the execution of high-dimensional datasets such as the Parkinson, CNAE-9, and Urban Land Cover dataset. The feature selection is done using the Correlation-Based Feature Selection method, using the BestFirst method. Tests were carried out 30 times using the K-Cross Fold Validation value of 10 and dividing the dataset into 70% training and 30% testing. The experiments using the Parkinson dataset obtained a time difference of 0.27 and 0.28 seconds faster than using the Random Forest method without feature selection. Likewise, the trials in the Urban Land Cover dataset had 0.04 and 0.03 seconds, while for the CNAE-9 dataset, the difference time was 2.23 and 2.81 faster than using the Random Forest method without feature selection. These experiments showed that the Random Forest processes are faster when using the first feature selection. Likewise, the accuracy value increased in the two previous experiments, while only the CNAE-9 dataset experiment gets a lower accuracy. This research’s benefits is by first performing feature selection steps using the Correlation-Base Feature Selection method can increase the speed of performance and accuracy of the Random Forest method on high-dimensional data.

COMPARISON BETWEEN DIFFERENT VERSIONS OF INPE'S FIRE RISK MODEL FOR THE BRAZILIAN BIOMES

Abstract. Fires are intrinsic disturbances in ecosystems functioning and structure in fire-prone biomes. In recent decades there has been an increase in the number of fire events in Brazilian biomes, especially due to misuse of fire in the land use and deforestation. The spatial and temporal pattern fire risk is a important way to understanding the seasonality and intensity of fire in different climate and fuel conditions. However, consistent long-term assessment at biome level is only possible with the support of remote sensing and modeling information. Thus, the objective of this work was to evaluate the fire risk patterns for the Brazilian biomes in the last years (2015–2018), using the new version of INPE’s fire risk (FR, v2). Regarding the temporal and spatial FR patterns by this new version from FR model, we evaluated that elevation and latitude correction factors, as well as the meteorological and land cover datasets with finer spatial scales can be contributed to adjust better the fire season vulnerability, notably in the less prone-biomes, such as Mata Atlantica, Pampa and Pantanal. However, there is still a need for adjustment to match the spatial active fire distribution, considering a biomass (fuel) map and the vegetation water status indicators. These improvements help to inform with more accuracy the most fire prone areas to define the strategies and decisions for fire combat and management.

Modelling geographical accessibility to support disaster response and rehabilitation of a healthcare system: an impact analysis of Cyclones Idai and Kenneth in Mozambique

ObjectivesModelling and assessing the loss of geographical accessibility is key to support disaster response and rehabilitation of the healthcare system. The aim of this study was therefore to estimate postdisaster...

Residential Urbanicity is Associated with glycosylated hemoglobin levels in Older Adults

Introduction:Environmental pollutants have been previously associated with increased risk of diabetes mellitus (DM) and perturbations in glucose homeostasis. Little is known however about how, and to what extent, neighborhood urbanicity may impact biomarkers of glucose control.Methods:We used mixed effects multiple regression models to evaluate neighborhood urbanicity and glycosylated hemoglobin (HbA1c) levels among 4,121 individuals >57 years of age enrolled in the National Social Life, Health, and Aging Project. We also examined for effect modification by race/ethnicity, socioeconomic status, health status, BMI, and age. Secondary analyses examined whether associations were mediated by PM2.5, NO2, and physical activity. Urbanicity measurements were calculated as the percentage of low- and high-intensity residential, and industrial/ commercial/ transportation land use within 1 km of each participants residence, using data obtained from the US Geological Survey National Land Cover Dataset. HbA1c levels were measured from dried blood spots and were log transformed prior to inclusion in models. All models were adjusted for demographic, socioeconomic, health behavior, medical history, and current medications which have been previously shown to affect HbA1c and DM risk.Results:Each IQR increase in surrounding urban landcover was associated with a 1.0% (95% CI 0.1, 1.8) increase in HbA1c levels. In effect modification models, the associations were found to be over four times stronger among Black participants (4.7%, 95% CI: 0.7 -8.9) relative to White participants (0.8%, 95% CI: 0.0, 1.7; Pinteract 0.04). No significant mediation was observed by any investigated variables.Conclusions:In a cohort of older US men and women, neighborhood urbanicity was significantly associated with HbA1c, with significantly larger affects in Black participants. These associations suggest that urbanicity may be an important contributor to perturbed glucose metabolism in older adults.

Large Uncertainty on Forest Area Change in the Early 21st Century among Widely Used Global Land Cover Datasets

Forests play an important role in the Earth’s system. Understanding the states and changes in global forests is vital for ecological assessments and forest policy guidance. However, there is no consensus on how global forests have changed based on current datasets. In this study, five global land cover datasets and Global Forest Resources Assessments (FRA) were assessed to reveal uncertainties in the global forest changes in the early 21st century. These datasets displayed substantial divergences in total area, spatial distribution, latitudinal profile, and annual area change from 2001 to 2012. These datasets also display completely divergent conclusions on forest area changes for different countries. Among the datasets, total forest area changes range from an increase of 1.7 × 106 km2 to a decrease of 1.6 × 106 km2. All the datasets show deforestation in the tropics. The accuracies of the datasets in detecting forest cover changes were evaluated by a global land cover validation dataset. The spatial patterns of accuracies are inconsistent among the datasets. This study calls for the development of a more accurate database to support forest policies and to contribute to global actions against climate change.

Accounting for land in the United States: Integrating physical land cover, land use, and monetary valuation

Land plays a critical role in both economic and environmental accounting. As an asset, it occupies a unique position at the intersection of the System of National Accounts (SNA), the System of Environmental-Economic Accounting Central Framework (SEEA-CF), and (as a spatial unit) SEEA Experimental Ecosystem Accounting (SEEA-EEA), making land a natural starting point for developing natural capital accounts more generally. We develop a pilot set of national and subnational land accounts for the United States that are consistent with the SEEA-CF and SNA principles, quantified in both physical and monetary terms. The physical accounts utilize detailed land use (National Land Use Database) and land cover (National Land Cover Database) datasets, which provide insights into how land cover in the U.S. is changing over time. To provide aggregate estimates of land values, we use a hedonic approach that exploits fine-grain microdata (“big data” from Zillow) that contains detailed information from hundreds of millions of property transactions and their corresponding physical characteristics covering much of the U.S. Methodologically, we show that it is feasible to produce monetary accounts for land that can be directly linked to and integrated with physical land cover/use. Overall, U.S. land cover has shown declines in forests, cropland, and pasture with increases in barren, scrub/shrub, and developed classes, which are particularly concentrated in the U.S. Southeast. Nominal land values in the U.S. fell about 28% ($7 trillion) from the boom to bust periods in the prior decade, albeit with substantial regional variation, and have subsequently experienced a nearly full recovery in recent years. We estimate private land in the contiguous 48 states to be worth approximately $25.1 trillion in 2016.

Accuracy Assessment of the FROM-GLC30 Land Cover Dataset Based on Watershed Sampling Units: A Continental-Scale Study

Land cover information plays an essential role in the study of global surface change. Multiple land cover datasets have been produced to meet various application needs. The FROM-GLC30 (Finer Resolution Observation and Monitoring of Global Land Cover) dataset is one of the latest land cover products with a resolution of 30 m, which is a relatively high resolution among global public datasets, and the accuracy of this dataset is of great concern in many related researches. The objective of this study was to calculate the accuracy of the FROM-GLC30 2017 dataset at the continental scale and to explore the spatial variation differences of each land type accuracy in different regions. In this study, the visual interpretation land cover results at 20,936 small watershed sampling units based on high-resolution remote sensing images were used as the reference data covering 65 countries in Asia, Europe, and Africa. The reference data were verified by field survey in typical watersheds. Based on that, the accuracy assessment of the FROM-GLC30 2017 dataset was carried out. The results showed (1) the area proportion of different land cover types in the FROM-GLC30 2017 dataset was generally consistent with that of the reference data. (2) The overall accuracy of the FROM-GLC30 2017 dataset was 72.78%, and was highest in West Asia–Northeast Africa, and lowest in South Asia. (3) Among all the seven land cover types, the accuracy of bareland and forest was relatively higher than that of others, and the accuracy of shrubland was the lowest. The accuracy for each land cover type differed among regions. The results of this work can provide useful information for land cover accuracy assessment researches at a large scale and promote the further practical applications of the open-source land cover datasets.

Identifying Agricultural Frontiers for Modeling Global Cropland Expansion

SummaryThe increasing expansion of cropland is major driver of global carbon emissions and biodiversity loss. However, predicting plausible future global distributions of croplands remains challenging. Here, we show that, in general, existing global data aligned with classical economic theories of expansion explain the current (1992) global extent of cropland reasonably well, but not recent expansion (1992–2015). Deviations from models of cropland extent in 1992 (“frontierness”) can be used to improve global models of recent expansion, most likely as these deviations are a proxy for cropland expansion under frontier conditions where classical economic theories of expansion are less applicable. Frontierness is insensitive to the land cover dataset used and is particularly effective in improving models that include mosaic land cover classes and the largely smallholder-driven frontier expansion occurring in such areas. Our findings have important implications as the frontierness approach offers a straightforward way to improve global land use change models.

Integrating C- and L-Band SAR Imagery for Detailed Flood Monitoring of Remote Vegetated Areas

Flood detection and monitoring is increasingly important, especially on remote areas such as African tropical river basins, where ground investigations are difficult. We present an experiment aimed at integrating multi-temporal and multi-source data from the Sentinel-1 and ALOS 2 synthetic aperture radar (SAR) sensors, operating in C band, VV polarization, and L band, HH and HV polarizations, respectively. Information from the globally available CORINE land cover dataset, derived over Africa from the Proba V satellite, and available publicly at the resolution of 100 m, is also exploited. Integrated multi-frequency, multi-temporal, and multi-polarizations analysis allows highlighting different drying dynamics for floodwater over various land cover classes, such as herbaceous vegetation, wetlands, and forests. They also enable detection of different scattering mechanisms, such as double bounce interaction of vegetation stems and trunks with underlying floodwater, giving precious information about the distribution of flooded areas among the different ground cover types present on the site. The approach is validated through visual analysis from Google EarthTM imagery. This kind of integrated analysis, exploiting multi-source remote sensing to partially make up for the unavailability of reliable ground truth, is expected to assume increasing importance as constellations of satellites, observing the Earth in different electromagnetic radiation bands, will be available.

Urban Form as a Technological Driver of Carbon Dioxide Emission: A Structural Human Ecology Analysis of Onroad and Residential Sectors in the Conterminous U.S.

This study investigates the role of urban form as a technological driver of U.S. CO2 emissions for the onroad and residential sectors. The STIRPAT (Stochastic Impacts by Region on Population, Affluence, and Technology) human structural ecology framework is extended by drawing from science and technology studies (STS) to theorize urban form as a sociotechnical system involving practices and knowledge that contribute to urban land use as a material artifact on the landscape influencing emissions. Questions addressed are: (1) “What is the influence of urban form on total sector CO2 emissions?” and (2) “How does the influence of urban form on CO2 emissions differ for metropolitan versus non-metropolitan status?” Spatial error regression models were estimated using county-level CO2 emissions data from Project Vulcan. The National Land Cover Dataset (NLCD) was used to quantify measures of urban form. Other independent variables were derived from U.S. Census data. Results demonstrate carbon reduction benefits achievable through a developed land use mix containing a greater proportion of high intensity relative to low intensity use. Urban form matters, but it matters differently in terms of sign, significance, and interpretation depending on emission sector and metro versus non-metro status. A focus on urban form provides policymakers potential leverage for carbon mitigation efforts that focus on total emissions as opposed to per capita emission. A feature of the research is its integration of concepts and theory from structural human ecology, STS, land change science, and GIScience.

LAI-indicated vegetation dynamic in ecologically fragile region: A case study in the Three-North Shelter Forest program region of China

The Three-North Shelter Forest Program (TNSFP) region of China is an important ecological region covering more than 42.4% of China’s land area. Several ecological restoration projects have been implemented in this region, but evaluation work is relatively limited partly due to the selection of suitable remote sensing products. In this study, leaf area index is proposed as a suitable indicator to evaluate the ecological restoration situation as it is an indicator reflects the growth of green vegetations. The Global Land Surface Satellite (GLASS) LAI dataset and land cover dataset from 2000 to 2015 were adopted and analyzed in this study. The forest cover area slightly increased, while the grassland area decreased. Although large-scale forestation movements were carried out across the entire TNSFP region, the forest growth conditions mainly improved in regions with an annual precipitation greater than 400 mm. Furthermore, LAI-precipitation correlations were very high (r greater than 0.6) in the regions across 400 mm isohyets, but lower (r < 0.4) far away from 400 mm isohyets. In relatively arid regions, forest growth did not show an obvious increase trend and sometimes a decrease trend. The increase in the LAI of these regions was mainly due to the restoration of grassland. Therefore, although LAI in most of the TNSFP region showed increasing trends, it is still not sufficient to state that these trends were caused by forestation projects in China.

Comparison of ESA climate change initiative land cover to CORINE land cover over Eastern Europe and the Baltic States from a regional climate modeling perspective

High-quality land use and land cover (LULC) information is of crucial importance for the performance of regional climate models (RCMs), in particular at high spatial resolutions down to convection permitting scales below 4 km. Several satellite-based high-resolution products are currently available for implementation into RCMs. One of the most recent products is the European Space Agency Climate Change Initiative Land Cover (ESA CCI LC) dataset. While the ESA CCI LC has been assessed globally, an evaluation against regional, independent LULC datasets is necessary to identify LULC inaccuracies in the respective region of interest and to give regional climate modelers estimates for the uncertainty in the land use forcing. In the present work the ESA CCI LC dataset is compared to the COoRdination and INformation on the Environment (CORINE) Land Cover (CLC). Agreement between the datasets is assessed by proportional area comparison (PAC). The resulting agreement measures are compared to the results of a majority approach (MA) to explore possible differences between the methods. Three timesteps of ESA CCI LC matching the timesteps of CLC are assessed to take a change in agreement over time into account. In addition to the quantification of agreement, spatial patterns of possible issues with ESA CCI LC are identified through utilization of geospatial information systems (GIS). Using the PAC, the agreement of ESA CCI LC with CLC is found to be ∼76 % for the research area (RA). Although the agreement decreases slightly using the PAC, no substantial differences in agreement measures were found compared to the results of the MA. Dominant LULC categories agriculture and forest show an agreement of over 80 % with CLC. A few major issues were found for grassland, wetlands, settlements, and water bodies in the RA of which some might influence RCM performance if the dataset is implemented without adjustment. We highly recommend to apply the PAC to other regions in Europe and further globally to investigate if the found issues are also found elsewhere. The use of more independent regional and specified datasets for validation but also for possible improvement of the ESA CCI LC dataset is suggested.

Understanding Completeness and Diversity Patterns of OSM-Based Land-Use and Land-Cover Dataset in China

OpenStreetMap (OSM) data are considered essential for land-use and land-cover (LULC) mapping despite their lack of quality. Most relevant studies have employed an LULC reference dataset for quality assessment, but such a reference dataset is not freely available for most countries and regions. Thus, this study conducts an intrinsic quality assessment of the OSM-based LULC dataset (i.e., without using a reference LULC dataset) by examining the patterns of both its completeness and diversity. With China chosen as the study area, an OSM-based LULC dataset of the country was first generated and validated by using various accuracy measures. Both its completeness and diversity patterns were then mapped and analyzed in terms of each prefecture-level division of the country. The results showed the following: (1) While the overall accuracy was as high as 82.2%, most complete regions of China were not mapped well owing to a lack of diverse LULC classes. (2) In terms of socioeconomic factors and the number of contributors, higher correlations were noted for diversity patterns than completeness patterns; thus, the diversity pattern is a better reflection of socioeconomic factors and the spatial patterns of contributors. (3) Both the completeness and the diversity patterns can be combined to better understand an OSM-based LULC dataset. These results indicate that it is useful to consider diversity as a supplement for intrinsically assessing the quality of an OSM-based LULC dataset. This analytical method can also be applied to other countries and regions.

Role of vegetation in representing land surface temperature in the CHTESSEL (CY45R1) and SURFEX-ISBA (v8.1) land surface models: a case study over Iberia

Abstract. Earth observations were used to evaluate the representation of land surface temperature (LST) and vegetation coverage over Iberia in two state-of-the-art land surface models (LSMs) – the European Centre for Medium-Range Weather Forecasts (ECMWF) Carbon-Hydrology Tiled ECMWF Scheme for Surface Exchanges over Land (CHTESSEL) and the Météo-France Interaction between Soil Biosphere and Atmosphere model (ISBA) within the SURface EXternalisée modeling platform (SURFEX-ISBA) for the 2004–2015 period. The results showed that the daily maximum LST simulated by CHTESSEL over Iberia was affected by a large cold bias during summer months when compared against the Satellite Application Facility on Land Surface Analysis (LSA-SAF), reaching magnitudes larger than 10 ∘C over wide portions of central and southwestern Iberia. This error was shown to be tightly linked to a misrepresentation of the vegetation cover. In contrast, SURFEX simulations did not display such a cold bias. We show that this was due to the better representation of vegetation cover in SURFEX, which uses an updated land cover dataset (ECOCLIMAP-II) and an interactive vegetation evolution, representing seasonality. The representation of vegetation over Iberia in CHTESSEL was improved by combining information from the European Space Agency Climate Change Initiative (ESA-CCI) land cover dataset with the Copernicus Global Land Service (CGLS) leaf area index (LAI) and fraction of vegetation coverage (FCOVER). The proposed improvement in vegetation also included a clumping approach that introduces seasonality to the vegetation cover. The results showed significant added value, removing the daily maximum LST summer cold bias completely, without reducing the accuracy of the simulated LST, regardless of season or time of the day. The striking performance differences between SURFEX and CHTESSEL were fundamental to guiding the developments in CHTESSEL highlighting the importance of using different models. This work has important implications: first, it takes advantage of LST, a key variable in surface–atmosphere energy and water exchanges, which is closely related to satellite top-of-atmosphere observations, to improve the model's representation of land surface processes. Second, CHTESSEL is the land surface model employed by ECMWF in the production of their weather forecasts and reanalysis; hence systematic errors in land surface variables and fluxes are then propagated into those products. Indeed, we showed that the summer daily maximum LST cold bias over Iberia in CHTESSEL is present in the widely used ECMWF fifth-generation reanalysis (ERA5). Finally, our results provided hints about the interaction between vegetation land–atmosphere exchanges, highlighting the relevance of the vegetation cover and respective seasonality in representing land surface temperature in both CHTESSEL and SURFEX. As a whole, this work demonstrated the added value of using multiple earth observation products for constraining and improving weather and climate simulations.

Evaluating the Impacts of Land Cover and Soil Texture Changes on Simulated Surface Wind and Temperature

AbstractIn this study, Chinese Academy of Sciences land cover dataset (CAS_LC) and soil texture dataset (CAS_ST) as well as Tsinghua University land cover dataset (TU_LC) were incorporated into the Weather Research and Forecasting (WRF) model to evaluate the impacts of land cover and soil texture changes on the surface wind and air temperature as compared with outdated default datasets. Six modeling scenarios including single updating for the three new datasets, combined updating of new datasets (CAS_LC + CAS_ST, TU_LC + CAS_ST), and default datasets are designed. WRF simulations were conducted with three resolutions (9, 3, and 1 km) for the period of 14 July to 31 September 2017 over Harbin. Results of the 10‐m wind speed, 10‐m wind direction, and 2‐m air temperature (T2) were evaluated against WRF default datasets (WLS). Results show that TU_LC‐related simulations with lower mean absolute error (MAE) demonstrated the usefulness of this dataset in the wind speed simulation, while the changing soil texture exerted a limited effect on modeled wind. CAS_LC improved simulated wind direction and T2. Significant changes in the three variables were frequently induced by urban and built‐up land or water bodies related transformations. The 3‐km grid resolution was sufficient to reasonably simulate the surface winds and air temperature. The roughness length was a main factor disturbing the wind speed, while the land cover transformation was responsible for the changes in surface temperature and heat fluxes, thereby affecting T2 simulation.

How an urban parameterization affects a high‐resolution global climate simulation

The impact of urban areas on the global and regional climate has been assessed using the global Conformal Cubic Atmospheric Model (CCAM) including an urban canyon parameterization at a global resolution of 50 km. Simulations were produced with and without urban areas to assess urban impacts for the historical period 1980–2000. Two different land cover and urban datasets (one based on IGBP‐DIS, the other on MODIS) were tested. In addition, simulations were performed for the end of the 21st century with the RCP8.5 scenario. Evaluation of the historical climate simulations indicates realistic local urban effects, such as higher daily minimum air temperatures (tasmin), higher sensible heat flux and lower latent heat flux at urban grid cells. In regions with large fractions of urban areas, some regional changes are also noted. In addition, there are significant regional effects far away from the main urban areas, which are similar in magnitude to the effects of the different non‐urban land cover input datasets. Under the projected warming at the end of the 21st century (with no land cover change), there is a decrease in anthropogenic heating, primarily during wintertime. There is a slightly smaller increase in daily maximum temperature and a slightly larger increase in tasmin in urban areas compared to rural areas. This leads to a smaller increase in the diurnal temperature range within urban areas. The tasmin changes also imply an increase in the urban heat island effect for larger cities. The results of this sensitivity study show that there is a detectable impact of urban areas on high‐resolution global climate simulations. Consequently, there is a need to include urban areas in global simulations, as well as in studies of land‐use change.

THE LAND COVER CLASSIFICATION USING A FEATURE PYRAMID NETWORKS ARCHITECTURE FROM SATELLITE IMAGERY

Abstract. Extracting land cover information from satellite imagery is of great importance for the task of automated monitoring in various remote sensing applications. Deep convolutional neural networks make this task more feasible, but they are limited by the small dataset of annotated images. In this paper, we present a fully convolutional networks architecture, FPN-VGG, that combines Feature Pyramid Networks and VGG. In order to accomplish the task of land cover classification, we create a land cover dataset of pixel-wise annotated images, and employ a transfer learning step and the variant dice loss function to promote the performance of FPN-VGG. The results indicate that FPN-VGG shows more competence for land cover classification comparing with other state-of-the-art fully convolutional networks. The transfer learning and dice loss function are beneficial to improve the performance of on the small and unbalanced dataset. Our best model on the dataset gets an overall accuracy of 82.9%, an average F1 score of 66.0% and an average IoU of 52.7%.

Data Analysis of Land Use Change and Urban and Rural Impacts in Lagos State, Nigeria

This study examines land use change and impacts on urban and rural activity in Lagos State, Nigeria. To achieve this, multi-temporal land use and land cover (LULC) datasets derived from the GlobeLand30 product of years 2000 and 2010 for urban and rural areas of Lagos State were imported into ArcMap 10.6 and converted to raster files (raster thematic maps) for spatial analysis in the FRAGSTATS situated in the Patch Analyst. Thus, different landscape metrics were computed to generate statistical results. The results have shown that fragmentation of cultivated lands increased in the rural areas but decreased in the urban areas. Also, the findings display that land-use change resulted in incremental fragmentation of forest in the urban areas, and reduction in the rural areas. The fragmentation measure of diversity increased in the urban areas, while it decreased in the rural areas during the period of study. These results suggest that cultivated land fragmentation is a complex process connected with socio-economic trends at regional and local levels. In addition, this study has shown that landscape metrics can be used to understand the spatial pattern of LULC change in an urban-rural context. Finally, the outcomes of this study will help the policymakers at the three levels of governments in Nigeria to make crucial informed decisions about sustainable land use.