State-of-the-art cloud computing platforms such as Google Earth Engine (GEE) enable regional-to-global land cover and land cover change mapping with machine learning algorithms. However, collection of high-quality training data, which is necessary for accurate land cover mapping, remains costly and labor-intensive. To address this need, we created a global database of nearly 2 million training units spanning the period from 1984 to 2020 for seven primary and nine secondary land cover classes. Our training data collection approach leveraged GEE and machine learning algorithms to ensure data quality and biogeographic representation. We sampled the spectral-temporal feature space from Landsat imagery to efficiently allocate training data across global ecoregions and incorporated publicly available and collaborator-provided datasets to our database. To reflect the underlying regional class distribution and post-disturbance landscapes, we strategically augmented the database. We used a machine learning-based cross-validation procedure to remove potentially mis-labeled training units. Our training database is relevant for a wide array of studies such as land cover change, agriculture, forestry, hydrology, urban development, among many others.

Sample-based estimates augmented by complete coverage land-cover maps were used to estimate area and describe patterns of annual land-cover change across the conterminous United States (CONUS) between 1985 and 2016. Most of the CONUS land cover remained stable in terms of net class change over this time, but a substantial gross change dynamic was captured by the annual and cumulative time intervals. The dominant types of changes can be grouped into natural resource cycles, increases in urbanization, and surface-water dynamics. The annual estimates over the 30-year time series showed a reduction in the rate of urban expansion after 2006, new growth in cropland after 2007, but a net overall decline in cropland since 1985, and two eras of net tree cover loss, the first one early in the time series and the second starting in 2012. Our study provides a holistic assessment of the CONUS land-cover conversion (class) change and can serve as a new benchmark for future research.

Land cover maps are essential for characterizing the biophysical properties of the Earth’s land areas. Because land cover information synthesizes a rich array of information related to both the ecological condition of land areas and their exploitation by humans, they are widely used for basic and applied research that requires information related to land surface properties (e.g., terrestrial carbon models, water balance models, weather, and climate models) and are core inputs to models and analyses used by natural resource scientists and land managers. As the Earth’s global population has grown over the last several decades rates of land cover change have increased dramatically, with enormous impacts on ecosystem services (e.g., biodiversity, water supply, carbon sequestration, etc.). Hence, accurate information related to land cover is essential for both managing natural resources and for understanding society’s ecological, biophysical, and resource management footprint. To address the need for high-quality land cover information we are using the global record of Landsat observations to compile annual maps of global land cover from 2001 to 2020 at 30 m spatial resolution. To create these maps we use features derived from time series of Landsat imagery in combination with ancillary geospatial data and a large database of training sites to classify land cover at annual time step. The algorithm that we apply uses temporal segmentation to identify periods with stable land cover that are separated by breakpoints in the time series. Here we provide an overview of the methods and data sets we are using to create global maps of land cover. We describe the algorithms used to create these maps and the core land cover data sets that we are creating through this effort, and we summarize our approach to accuracy assessment. We also present a synthesis of early results and discuss the strengths and weaknesses of our early map products and the challenges that we have encountered in creating global land cover data sets from Landsat. Initial accuracy assessment for North America shows good overall accuracy (77.0 ± 2.0% correctly classified) and 79.8% agreement with the European Space Agency (ESA) WorldCover product. The land cover mapping results we report provide the foundation for robust, repeatable, and accurate mapping of global land cover and land cover change across multiple decades at 30 m spatial resolution from Landsat.

The U.S. Geological Survey’s Land Change Monitoring, Assessment, and Projection (LCMAP) initiative involves detecting changes in land cover, use, and condition with the goal of producing land change information to improve the understanding of the Earth system and provide insights on the impacts of land surface change on society. The change detection method ingests all available high-quality data from the Landsat archive in a time series approach to identify the timing and location of land surface change. Annual thematic land cover maps are then produced by classifying time series models. In this paper, we describe the optimization of the classification method used to derive the thematic land cover product. We investigated the influences of auxiliary data, sample size, and training from different sources such as the U.S. Geological Survey’s Land Cover Trends project and National Land Cover Database (NLCD 2001 and NLCD 2011). The results were evaluated and validated based on independent data from the training dataset. We found that refining the auxiliary data effectively reduced artifacts in the thematic land cover map that are related to data availability. We improved the classification accuracy and stability considerably by using a total of 20 million training pixels with a minimum of 600,000 and a maximum of 8 million training pixels per class within geographic windows consisting of nine Analysis Ready Data tiles (450 by 450 km2). Comparisons revealed that the NLCD 2001 training data delivered the best classification accuracy. Compared to the original LCMAP classification strategy used for early evaluation (e.g., Trends training data, 20,000 samples), the optimized classification strategy improved the annual land cover map accuracy by an average of 10%.

Providing rapid access to land surface change data and information is a goal of the U.S. Geological Survey. Through the Land Change Monitoring, Assessment, and Projection (LCMAP) initiative, we have initiated a monitoring capability that involves generating a suite of 10 annual land cover and land surface change datasets across the United States at a 30-m spatial resolution. During the LCMAP automated production, on a tile-by-tile basis, erroneous data can occasionally be generated due to hardware or software failure. While crucial to assure the quality of the data, rapid evaluation of results at the pixel level during production is a substantial challenge because of the massive data volumes. Traditionally, product quality relies on the validation after production, which is inefficient to reproduce the whole product when an error occurs. This paper presents a method for automatically evaluating LCMAP results during the production phase based on 14 indices to quickly find and flag erroneous tiles in the LCMAP products. The methods involved two types of comparisons: comparing LCMAP values across the temporal record to measure internal consistency and calculating the agreement with multiple intervals of the National Land Cover Database (NLCD) data to measure the consistency with existing products. We developed indices on a tile-by-tile basis in order to quickly find and flag potential erroneous tiles by comparing with surrounding tiles using local outlier factor analysis. The analysis integrates all indices into a local outlier score (LOS) to detect erroneous tiles that are distinct from neighboring tiles. Our analysis showed that the methods were sensitive to partially erroneous tiles in the simulated data with a LOS higher than 2. The rapid quality assessment methods also successfully identified erroneous tiles during the LCMAP production, in which land surface change results were not properly saved to the products. The LOS map and indices for rapid quality assessment also point to directions for further investigations. A map of all LOS values by tile for the published LCMAP shows all LOS values are below 2. We also investigated tiles with high LOS to ensure the distinction with neighboring tiles was reasonable. An index in this study shows the overall agreement between LCMAP and NLCD on a tile basis is above 71.5% and has an average at 89.1% across the 422 tiles in the conterminous United States. The workflow is suitable for other studies with a large volume of image products.

Growing demands for temporally specific information on land surface change are fueling a new generation of maps and statistics that can contribute to understanding geographic and temporal patterns of change across large regions, provide input into a wide range of environmental modeling studies, clarify the drivers of change, and provide more timely information for land managers. To meet these needs, the U.S. Geological Survey has implemented a capability to monitor land surface change called the Land Change Monitoring, Assessment, and Projection (LCMAP) initiative. This paper describes the methodological foundations and lessons learned during development and testing of the LCMAP approach. Testing and evaluation of a suite of 10 annual land cover and land surface change data sets over six diverse study areas across the United States revealed good agreement with other published maps (overall agreement ranged from 73% to 87%) as well as several challenges that needed to be addressed to meet the goals of robust, repeatable, and geographically consistent monitoring results from the Continuous Change Detection and Classification (CCDC) algorithm. First, the high spatial and temporal variability of observational frequency led to differences in the number of changes identified, so CCDC was modified such that change detection is dependent on observational frequency. Second, the CCDC classification methodology was modified to improve its ability to characterize gradual land surface changes. Third, modifications were made to the classification element of CCDC to improve the representativeness of training data, which necessitated replacing the random forest algorithm with a boosted decision tree. Following these modifications, assessment of prototype Version 1 LCMAP results showed improvements in overall agreement (ranging from 85% to 90%).

•∼95% of deforestation was within 5.5km of a road or 1.0km of a navigable river.•35.2% of the Brazilian Amazon was highly accessible by river or road.•Accessible unprotected land was 43.6% deforested.•Accessible protected land was 10.9% deforested.•Protected areas avoided or displaced 34–39 thousandkm2 of deforestation.

Forest degradation in the Brazilian Amazon due to selective logging and forest fires may greatly increase the human footprint beyond outright deforestation. We demonstrate a method to quantify annual deforestation and degradation simultaneously across the entire region for the years 2000–2010 using high-resolution Landsat satellite imagery. Combining spectral mixture analysis, normalized difference fraction index, and knowledge-based decision tree classification, we mapped and assessed the accuracy to quantify forest (0.97), deforestation (0.85) and forest degradation (0.82) with an overall accuracy of 0.92. We show that 169,074 km2 of Amazonian forest was converted to human-dominated land uses, such as agriculture, from 2000 to 2010. In that same time frame, an additional 50,815 km2 of forest was directly altered by timber harvesting and/or fire, equivalent to 30% of the area converted by deforestation. While average annual outright deforestation declined by 46% between the first and second halves of the study period, annual forest degradation increased by 20%. Existing operational monitoring systems (PRODES: Monitoramento da Florestal Amazônica Brasileira por Satélite) report deforestation area to within 2% of our results, but do not account for the extensive forest degradation occurring throughout the region due to selective logging and forest fire. Annual monitoring of forest degradation across tropical forests is critical for developing land management policies as well as the monitoring of carbon stocks/emissions and protected areas.

investment has been organized into four-year plans: Brazil in Action (1996-1999), Advance Brazil (2000-2003), PPA [Multi-Annual Plan] (2004-2007), PAC [Program for the Acceleration of Growth] (2008-2011), and PAC-2 (2012-2015). Each plan has included a long list of roads, dams and other large infrastructure projects in the Amazon. Several of these projects have been included in a number of plans because economic constraints have prevented the completion of the projects at the pace initially imagined. This is the case with projects such as the Belo Monte Hydroelectric Dam and the BR-163 (Cuiaba - Santarem) and BR319 (Manaus-Porto Velho) Highways. These delayed projects are now either under construction or about to start. Models have been developed by different groups to predict the future consequences if projects such as these are undertaken. Different models capture different aspects of the problem, and many of them indicate large increases in deforestation and degradation with serious environmental and social implications. One of the models takes as a point of departure the assumption that roads would have negligible or even beneficial effects on total deforestation, but this contradicts what is observed in the real world.

Desde 1996 o planejamento de investimentos do governo brasileiro tem sido organizado em planos quadrianuais: Brasil em Ação (1996-1999), Avança Brasil (2000-2003), PPA [Plano Plurianual] (2004-2007), PAC [Programa de Aceleração do Crescimento] (2008-2011), e PAC-2 (2012-2015). Cada plano tem incluído uma longa lista de rodovias, barragens e outros grandes projetos de infraestrutura na Amazônia. Vários desses projetos têm sido incluídos em uma série de planos, pois restrições econômicas não permitiram a realização das obras no ritmo inicialmente imaginado. Este é o caso de obras como a hidrelétrica de Belo Monte e as rodovias BR-163 (Santarém-Cuiabá) e BR-319 (Manaus-Porto Velho). Estes projetos adiados estão hoje sendorealizados ou próximos à realização. Uma série de modelos tem sido elaborada por diferentes grupos para prever as consequências futuras, caso essas obras sejam realizadas. Diferentes modelos captam diferentes aspectos da problemática, e vários deles indicam grandes aumentos de desmatamento e degradação, com graves implicações ambientais e sociais. Um dos modelos parte da premissa de que as estradas teriam efeitos nulos ou até benéficos sobre total de desmatamento, mas essa suposição contradiz o que é observado no mundo real.

Recognizing the importance of preserving biodiversity and ecosystem services, human society has established extensive protected area networks to conserve these resources in recent decades. Are protected areas working as expected? Empirical coarse-scale assessments of this question across large regions, or even globally, tend to answer "yes", while fine-scale studies of individual protected areas often and repeatedly answer "no". We conducted a first fine-scale analysis of Brazil's extensive Amazonian protected area network (1.8 million km(2)) and have quantitatively estimated conservation effectiveness in light of changing human development pressures in the surrounding landscape. The overall network maintained intact forest cover for 98.6% of protected forest lands, largely agreeing with previous coarse-scale studies. However, detailed examination of 474 individual protected areas unveils a broad range of efficacy. Many protected areas (544,800 km(2)) experience default protection simply due to their remoteness. Many others (396,100 km(2)) have provided highly effective protection in the face of substantial human development pressure. Conversely, 12% (38) of protected areas have failed to protect the 27,300 km(2) that they encompass, and another 7% (23) provide only marginal protection of 37,500 km(2). Comprehensive landscape assessments of protected area networks, with frequent monitoring at scales matching the patterns of human-caused disturbances, are necessary to ensure the conservation effectiveness and long term survival of protected areas in rapidly changing landscapes. The methods presented here are globally adaptable to all forested protected areas. (c) 2012 Elsevier Ltd. All rights reserved.

There is increasing consensus that the global climate will continue to warm over the next century. The biodiversity-rich Amazon forest is a region of growing concern because many global climate model (GCM) scenarios of climate change forecast reduced precipitation and, in some cases, coupled vegetation models predict dieback of the forest. To date, fires have generally been spatially co-located with road networks and associated human land use because almost all fires in this region are anthropogenic in origin. Climate change, if severe enough, could alter this situation, potentially changing the fire regime to one of increased fire frequency and severity for vast portions of the Amazon forest. High moisture contents and dense canopies have historically made Amazonian forests extremely resistant to fire spread. Climate will affect the fire situation in the Amazon directly, through changes in temperature and precipitation, and indirectly, through climate-forced changes in vegetation composition and structure. The frequency of drought will be a prime determinant of both how often forest fires occur and how extensive they become. Fire risk management needs to take into account landscape configuration, land cover types and forest disturbance history as well as climate and weather. Maintaining large blocks of unsettled forest is critical for managing landscape level fire in the Amazon. The Amazon has resisted previous climate changes and should adapt to future climates as well if landscapes can be managed to maintain natural fire regimes in the majority of forest remnants.

As valuation of ecosystem goods and services derived from ecological processes becomes increasingly important in environmental decision-making, the need to quantify how restoration activities influence ecosystem function has grown more urgent, particularly within income-generating or subsistence-providing landscapes where economic needs and biodiversity goals must be balanced. However, quantification of restoration effects is often hindered by logistical issues, which include (1) the difficulty of systematically monitoring responses over large areas and (2) lack or loss of comparison sites necessary for assessing treatment effect. We explored the use of remote sensing to quantify the effects of native grass seeding and prescribed burns on ecosystem forage provisioning services within a California (U.S.A.) rangeland landscape. We used Landsat time series to monitor forage (green biomass) dynamics within 296 ha of treatment areas-distributed throughout a 36-km(2) watershed-for 6 years and to identify post hoc comparison areas when a priori comparisons were lacking. Remote sensing analysis documented gains and losses in forage provisioning services due to restoration efforts and provided critical information for adaptive management. Our results demonstrate the degree to which invaded grasslands can be resistant to change and suggest that increasing the functional complexity of restoration mixes might help increase forage availability and reduce opportunities for weed reinvasion.

The Brazilian Amazon is currently experiencing the world9s highest absolute rate of forest destruction and is likely to suffer even greater degradation in the future because of government plans to invest $40 billion from 2000 to 2007 in dozens of major new highways and infrastructure projects. We developed two computer models that integrate spatial data on deforestation, logging, mining, highways and roads, navigable rivers, vulnerability to wildfires, protected areas, and existing and planned infrastructure projects, in an effort to predict the condition of Brazilian Amazonian forests by the year 2020. Both models suggest that the region9s forests will be drastically altered by current development schemes and land-use trends over the next 20 years.