Search

The conversion of natural land cover into human-dominated land use systems has significant impacts on the environment. Global mapping and monitoring of human-dominated land use extent via satellites provides an empirical basis for assessing land use pressures. Here, we present a novel 2019 global land cover, land use, and ecozone map derived from Landsat satellite imagery and topographical data using derived image feature spaces and algorithms suited per theme. From the map, we estimate the spatial extent and dispersion of land use disaggregated by climate domain and ecozone, where dispersion is the mean distance of land use to all land within a subregion. We find that percent of area under land use and distance to land use follow a power law that depicts an increasingly random spatial distribution of land use as it extends across lands of comparable development potential. For highly developed climate/ecozones, such as temperate and sub-tropical terra firma vegetation on low slopes, area under land use is contiguous and remnant natural land cover have low areal extent and high fragmentation. The tropics generally have the greatest potential for land use expansion, particularly in South America. An exception is Asian humid tropical terra firma vegetated lowland, which has land use intensities comparable to that of temperate breadbaskets such as the United States’ corn belt. Wetland extent is inversely proportional to land use extent within climate domains, indicating historical wetland loss for temperate, sub-tropical, and dry tropical biomes. Results highlight the need for planning efforts to preserve natural systems and associated ecosystem services. The demonstrated methods will be implemented operationally in quantifying global land change, enabling a monitoring framework for systematic assessments of the appropriation and restoration of natural land cover.

The lack of sufficient ground truth data has always constrained supervised learning, thereby hindering the generation of up-to-date satellite-derived thematic maps. This is all the more true for those applications requiring frequent updates over large areas such as cropland mapping. Therefore, we present a method enabling the automated production of spatially consistent cropland maps at the national scale, based on spectral-temporal features and outdated land cover information. Following an unsupervised approach, this method extracts reliable calibration pixels based on their labels in the outdated map and their spectral signatures. To ensure spatial consistency and coherence in the map, we first propose to generate seamless input images by normalizing the time series and deriving spectral-temporal features that target salient cropland characteristics. Second, we reduce the spatial variability of the class signatures by stratifying the country and by classifying each stratum independently. Finally, we remove speckle with a weighted majority filter accounting for per-pixel classification confidence. Capitalizing on a wall-to-wall validation data set, the method was tested in South Africa using a 16-year old land cover map and multi-sensor Landsat time series. The overall accuracy of the resulting cropland map reached 92%. A spatially explicit validation revealed large variations across the country and suggests that intensive grain-growing areas were better characterized than smallholder farming systems. Informative features in the classification process vary from one stratum to another but features targeting the minimum of vegetation as well as short-wave infrared features were consistently important throughout the country. Overall, the approach showed potential for routinely delivering consistent cropland maps over large areas as required for operational crop monitoring.

Humid tropical forests provide numerous global ecosystem services, but are under continuing threat of clearing from economic drivers. Here, we report primary humid tropical forest extent for the year 2001, and primary forest loss and distance to loss from 2002–2014 for the largest rainforest countries of Brazil, Democratic Republic of the Congo (DRC), and Indonesia. Brazil’s total area of primary forest loss is more than twice that of Indonesia and five times that of DRC. Despite unprecedented success in slowing deforestation along its forest frontier, Brazil’s most remote forests are increasingly nearer to loss, as extractive activities such as logging and mining intrude upon previously intact forests. In absolute terms, DRC has the lowest area of primary forest loss; however, its forests are increasingly encroached upon as smallholder agriculturalists move into remaining forests, often to escape conflict and insecurity. The decrease in DRC forests’ distance to loss as a function of area of forest loss was five times that of Brazil or Indonesia. In 2014, Indonesia had the least area of remaining primary forest. Despite an announced moratorium on concession licenses in 2011, Indonesia exhibited a rate of primary forest loss twice that of DRC and triple that of Brazil by the end of the study period. Forest loss dynamics in Indonesia range from industrial-scale clearing of coastal peatlands to logging of interior montane rainforests. While results illustrate considerable variation in forest loss dynamics between the three countries, the dominant narrative is of ongoing exploitation of primary humid tropical forests.

An error in the unit conversion from pixels to hectares lead to all the areal quantities in the text being smaller than they should have been. Only the number of hectares were changed; none of the text nor tables were changed. The changes do not affect the overall results or conclusions.

A Landsat-based humid tropical forest disturbance alert was implemented for Peru, the Republic of Congo and Kalimantan, Indonesia. Alerts were mapped on a weekly basis as new terrain-corrected Landsat 7 and 8 images were made available; results are presented for all of 2014 and through September 2015. The three study areas represent different stages of the forest land use transition, with all featuring a variety of disturbance dynamics including logging, smallholder agriculture, and agroindustrial development. Results for Peru were formally validated and alerts found to have very high user’s accuracies and moderately high producer’s accuracies, indicating an appropriately conservative product suitable for supporting land management and enforcement activities. Complete pan-tropical coverage will be implemented during 2016 in support of the Global Forest Watch initiative. To date, Global Forest Watch produces annual global forest loss area estimates using a comparatively richer set of Landsat inputs. The alert product is presented as an interim update of forest disturbance events between comprehensive annual updates. Results from this study are available for viewing and download at http://glad.geog.umd.edu/forest-alerts and http://www.globalforestwatch.org .

This study examines the suitability of 250 m MODIS (MODerate Resolution Imaging Spectroradiometer) data for mapping global cropland extent. A set of 39 multi-year MODIS metrics incorporating four MODIS land bands, NDVI (Normalized Difference Vegetation Index) and thermal data was employed to depict cropland phenology over the study period. Sub-pixel training datasets were used to generate a set of global classification tree models using a bagging methodology, resulting in a global per-pixel cropland probability layer. This product was subsequently thresholded to create a discrete cropland/non-cropland indicator map using data from the USDA-FAS (Foreign Agricultural Service) Production, Supply and Distribution (PSD) database describing per-country acreage of production field crops. Five global land cover products, four of which attempted to map croplands in the context of multiclass land cover classifications, were subsequently used to perform regional evaluations of the global MODIS cropland extent map. The global probability layer was further examined with reference to four principle global food crops: corn, soybeans, wheat and rice. Overall results indicate that the MODIS layer best depicts regions of intensive broadleaf crop production (corn and soybean), both in correspondence with existing maps and in associated high probability matching thresholds. Probability thresholds for wheat-growing regions were lower, while areas of rice production had the lowest associated confidence. Regions absent of agricultural intensification, such as Africa, are poorly characterized regardless of crop type. The results reflect the value of MODIS as a generic global cropland indicator for intensive agriculture production regions, but with little sensitivity in areas of low agricultural intensification. Variability in mapping accuracies between areas dominated by different crop types also points to the desirability of a crop-specific approach rather than attempting to map croplands in aggregate.

While publicly available, cost-free coarse and medium spatial resolution satellite data such as MODIS and Landsat perform well in characterizing industrial cropping systems, commercial high spatial resolution satellite data are often preferred alternative for fine scale land tenure agricultural systems such as found in Pakistan. In this article, we integrated commercial 5 m spatial resolution RapidEye and free 30 m Landsat imagery in characterizing winter wheat in Punjab province, Pakistan. Specifically, we used 5 m spatial resolution RapidEye imagery from peak of the winter wheat growing season to derive training data for the characterization of time-series Landsat data. After co-registration, each RapidEye image was classified into wheat/no wheat labels at the 5 m resolution and then aggregated as percent cover to 30 m Landsat grid cells. We produced four maps, two using RapidEye derived continuous training data (of percent wheat cover) as input to a regression tree model, and two using RapidEye derived categorical training data as input to a classification tree model. From the RapidEye-derived 30 m continuous training data, we derived Map 1 as percent wheat per pixel, and Map 2 as binary wheat/no wheat classification derived using a 50% threshold applied to Map 1. To create the categorical wheat/no wheat training data, we first converted the continuous training data to a wheat/no wheat classification, and then used these categorical RapidEye training data to produce a categorical wheat map from the Landsat data. Two methods for categorizing the training data were used. The first method used a 50% wheat/no wheat threshold to produce Map 3, and the second method used only pure wheat (≥75% cover) and no wheat (≤25% cover) training pixels to produce Map 4. The approach of Map 4 is analogous to a standard method in which whole, pure, high-confidence training pixels are delineated. We validated the wheat maps with field data collected using a stratified, two-stage cluster design. Accuracy of the maps produced from the percent cover training data (Map 1 and Map 2) was not substantially better than the accuracy of the maps produced from the categorical training data as all methods yielded similar overall accuracies (±standard error): 88% (±4%) for Map 1, 90% (±4%) for Map 2, 90% (±4%) for Map 3, and 87% (±4%) for Map 4. Because the percent cover training data did not produce significantly higher accuracies, sub-pixel training data are not required for winter wheat mapping in Punjab. Given sufficient expertise in supervised classification model calibration, freely available Landsat data are sufficient for crop mapping in the fine-scale land tenure system of Punjab. For winter wheat mapping in Punjab and other like landscapes, training data for supervised classification may be collected directly from Landsat images without the need for high resolution reference imagery.

Southeast Asia has some of the highest deforestation rates globally, with Malaysia being identified as a deforestation hotspot. The Malayan tiger, a critically endangered subspecies of the tiger endemic to Peninsular Malaysia, is threatened by habitat loss and fragmentation. In this study, we estimate the natural forest loss and conversion to plantations in Peninsular Malaysia and specifically in its tiger habitat between 1988 and 2012 using the Landsat data archive. We estimate a total loss of 1.35 Mha of natural forest area within Peninsular Malaysia over the entire study period, with 0.83 Mha lost within the tiger habitat. Nearly half (48%) of the natural forest loss area represents conversion to tree plantations. The annual area of new plantation establishment from natural forest conversion increased from 20 thousand ha year−1 during 1988–2000 to 34 thousand ha year−1 during 2001–2012. Large-scale industrial plantations, primarily those of oil palm, as well as recently cleared land, constitute 80% of forest converted to plantations since 1988. We conclude that industrial plantation expansion has been a persistent threat to natural forests within the Malayan tiger habitat. Expanding oil palm plantations dominate forest conversions while those for rubber are an emerging threat.

Land cover maps play an integral role in environmental management. However, countries and institutes encounter many challenges with producing timely, efficient, and temporally harmonized updates to their land cover maps. To address these issues we present a modular Regional Land Cover Monitoring System (RLCMS) architecture that is easily customized to create land cover products using primitive map layers. Primitive map layers are a suite of biophysical and end member maps, with land cover primitives representing the raw information needed to make decisions in a dichotomous key for land cover classification. We present best practices to create and assemble primitives from optical satellite using computing technologies, decision tree logic and Monte Carlo simulations to integrate their uncertainties. The concept is presented in the context of a regional land cover map based on a shared regional typology with 18 land cover classes agreed on by stakeholders from Cambodia, Laos PDR, Myanmar, Thailand, and Vietnam. We created annual map and uncertainty layers for the period 2000–2017. We found an overall accuracy of 94% when taking uncertainties into account. RLCMS produces consistent time series products using free long term historical Landsat and MODIS data. The customizable architecture can include a variety of sensors and machine learning algorithms to create primitives and the best suited smoothing can be applied on a primitive level. The system is transferable to all regions around the globe because of its use of publicly available global data (Landsat and MODIS) and easily adaptable architecture that allows for the incorporation of a customizable assembly logic to map different land cover typologies based on the user's landscape monitoring objectives