FIGURE 3 - uploaded by Temuulen Tsagaan Sankey
Content may be subject to copyright.
![Landsat 5 TM data processing workflow used to create a final image stack for the pixel-based classification for nominal year 1990. The final Landsat image stack for 1990 included 35 bands to classify cropland extent. EVI = enhanced vegetation index; NBR = normalized burn ratio; NDVI = normalized difference vegetation index; NDWI = normalized difference water index; VARI = visible atmospheric resistance index [Colour figure can be viewed at wileyonlinelibrary.com]](profile/Temuulen-Sankey/publication/325109871/figure/fig1/AS:648422298243072@1531607331186/Landsat-5-TM-data-processing-workflow-used-to-create-a-final-image-stack-for-the.png)
Landsat 5 TM data processing workflow used to create a final image stack for the pixel-based classification for nominal year 1990. The final Landsat image stack for 1990 included 35 bands to classify cropland extent. EVI = enhanced vegetation index; NBR = normalized burn ratio; NDVI = normalized difference vegetation index; NDWI = normalized difference water index; VARI = visible atmospheric resistance index [Colour figure can be viewed at wileyonlinelibrary.com]
Source publication
Large‐scale cropland changes have significant implications for global and national food supply as well as degradation in land resources. We examined cropland dynamics at the national scale in Mongolia over the last three decades using Google Earth Engine cloud computing and 11,360 Landsat satellite images. Our overarching goal was to develop the fi...
Similar publications
There are four characteristics typical of the distribution of China’s population in poverty: (1) concentration in the mountainous areas; (2) concentration in the western region; (3) concentration in environmentally fragile areas; (4) concentration in ethnic minority areas. In particular, poverty in this large country is very much related to environ...
Citations
... Geographical environmental conditions include soil, climate, and terrain, which usually affect the spatial patterns of cropland abandonment (Alcantara et al., 2013;Kerckhof et al., 2016). Socio-economic factors are regarded as direct causes of cropland abandonment (Sankey et al., 2018;Ustaoglu and Collier, 2018). Previous research typically employed indicators such as rural population, urbanization level, cultivation distance, and road accessibility to characterize socioeconomic changes relevant to agricultural production (Han and Song, 2020). ...
... It is expected that livestock will suffer from heat stress as a result of higher temperatures, reduced water availability, and lower quantity/quality of fodder. Further exacerbating livestock stressors are overgrazing, soil erosion and soil dryness, and land use change contributing to the reduction in pasture and crop land availability or suitability (Li et al., 2021;Han et al., 2021;Sankey et al., 2018;Klinge et al., 2018;Fawzy et al., 2020). ...
Mongolia's projected warming is far above the global average and could exceed 5 • C by the end of the century. The reliance on pastoral livestock and rainfed agriculture along with its fragile ecosystems put Mongolia's economy at risk of adverse climate change impacts, particularly from climate extreme events. Eighty percent of Mongolia's agricultural sector is concentrated in animal husbandry with around one third of the population relying on this livelihood. Beyond livestock, food production is concentrated in few crops: wheat; potatoes; and three vegetables (cabbage, carrot, and turnip). Climate change does not only affect food production but can exacerbate malnutrition by removing food and nutrients in all stages of the food value chain. To identify perceived effects of climate change and measures to reduce climate change impacts in Mongolia's key food value chains, we implemented focus group discussions with 214 livestock and vegetable producers, traders, and food consumers. We also conducted 30 key informant interviews at the soum, provincial, and national levels across four agroecosystems in three provinces. Based on this community engagement analysis, we identify interventions that the government and private sector, including herders and farmers, should undertake to increase the food security and nutrition of the country's prioritized food value chains under climate change.
... In the last century, Mongolia transitioned from a centralised to a free market economy, and rapid pastoralism growth occurred (Guo et al., 2021). Overgrazing and nomadic practices led to a continuous decline in grassland quality and even desertification, resulting in the conversion of grasslands to sand in central Mongolia (Sankey et al., 2018). With the proposal of regional revitalisation policies such as the Belt and Road, Grassland Road, and Development of the Far East initiatives, the frequent foreign trade in transnational regions has led to the further development of land resources, and the land dynamics are higher than those of surrounding areas, indicating that human activities have increasingly disturbed land uses. ...
Understanding the spatiotemporal characteristics of land use, ecosystem services, and their relationship provides a basis for the sustainable development of ecosystems. However, the effect of land use change on multiple ecosystem services has not been widely quantified, especially in transnational areas. This study analysed the spatiotemporal characteristics of land use change in the China–Mongolia–Russia Economic Corridor using the ESA-Global Land Cover product from 1992 to 2019. We used the InVEST model to evaluate the changes in carbon storage, habitat quality, water yield, and soil retention and quantified the contribution of land use to the changes in ecosystem services by calculating the ecosystem service contribution index (ESCI). The results showed that from 1992 to 2019, the urban/built-up environment and croplands increased by 22,010 km² and 32,946 km², respectively, which led to the continuous decline of forests. The implementation of ecological projects significantly decreased the barren environment. The value of four ecosystem services gradually decreased depending on the distance from the coastline and improved overall from 1992 to 2019. Land use conversion significantly affected ecosystem services. The expansion of croplands, urbanisation, and desertification had a significant negative impact on ecosystem services and the conversion to grasslands. Forests had a significant positive impact on ecosystem services. The findings will enrich our understanding of ecosystem services in transnational areas and help balance the relationship between ecological conservation and social-economic development in the China-Mongolia-Russia Economic Corridor.
... In this study, we observed some transitions between cropland and grassland by assessing the cover types around the edges of croplands (Figure 7), which may be attributable to our pixelbased image classification method. Across the study landscapes, we did not find similar 'all-or-none' pattern in cropland conversion to those of Sankey et al. (2018) who adopted an infusion of object-based and pixel-based image classification methods, as well as kernels to filter out misclassified land cover classes [31]. ...
... Because of mass land and sparse population, planting croplands in Mongolia is mostly opportunistic and based on funding availability of individual household. The Mongolian Statistical Office only records total harvest area annually [31]. This most likely leads to the discrepancy between the area of classified cropland and agricultural statistics that we discussed earlier. ...
... This most likely leads to the discrepancy between the area of classified cropland and agricultural statistics that we discussed earlier. There have been research efforts to map cropland abandonment using satellite imagery in Mongolia [31], European Russia [45], and Kazakhstan [46], but the limitations in spatial and temporal coverage persist. The success of mapping cropland abandonment relies on developing a long time series of satellite images, which is computationally demanding and labor intensive due to the need of large training samples. ...
As the two largest landlocked countries, Kazakhstan and Mongolia have similar biophys-ical conditions and socioeconomic roots in the former Soviet Union. Our objective is to investigate the direction, extent, and spatial variation of land cover change at three administrative levels over three decades (1990-2020). We selected three provinces from each country (Aktobe, Akmola, and Almaty province in Kazakhstan, and Arkhangai, Tov, and Dornod in Mongolia) to classify the land cover into forest, grassland, cropland, barren, and water. Altogether, 6964 Landsat images were used in pixel-based classification method with random forest model for image processing. Six thousand training data points (300 training points × 5 classes × 4 periods) for each province were collected for classification and change detection. Land cover changes at decadal and over the entire study period for five land cover classes were quantified at the country, provincial, and county level. High classification accuracy indicates localized land cover classification have an edge over the latest global land cover product and reveal fine differences in landscape composition. The vast steppe landscapes in these two countries are dominated by grasslands of 91.5% for Dornod in Mongolia and 74.7% for Aktobe in Kazakhstan during the 30-year study period. The most common land cover conversion was grassland to cropland. The cyclic land cover conversions between grassland and cropland reflect the impacts of the Soviet Union's largest reclamation campaign of the 20th century in Kazakhstan and the Atar-3 agriculture redevelopment in Mongolia. Kazakhstan experienced a higher rate of land cover change over a larger extent of land area than Mongolia. The spatial distribution of land use intensity indicates that land use hotspots are largely influenced by policy and its shifts. Future research based on these large-scale land use and land cover changes should be focused the corresponding ecosystem and society functions.
... The risk introduced by poor farm management can deteriorate the quality and yield of farm products [12], resulting in the abandonment of these croplands. Cropland abandonment is expected to result in land degradation including soil erosion, reduced species richness, and a decrease in the cover of perennial grasses [13][14][15][16][17]. ...
In recent years, Mongolia has witnessed an increase in not only wheat fields, which have been present for a long time, but also rapeseed fields. This has led to increasing concerns about soil degradation due to inappropriate cultivation. This study aims to determine the impacts of rapeseed production on soil water storage in Mongolia. The soil water content and matric potential were measured in wheat and rapeseed fields and adjacent steppe rangeland for five years, including crop production and fallow years, and the soil water storages in the fields were compared. The results demonstrated that the matric potential below the root zone in the rapeseed field and both rangelands was drier than the wilting point, whereas the potential in the wheat field was usually almost the same or wetter than this point. The comparison of the amount of soil water storage during the fallow year with that of the adjacent rangeland showed it to be 5–10% higher for the wheat field and almost equal for the rapeseed field. Field management must consider the fact that rapeseed fields use more water than is required by wheat fields and that less water is stored during fallow periods.
... Existing studies on land transformation in the Eurasian steppes so far have predominantly investigated post-Soviet agricultural change, including massive cropland abandonment and partial recultivation; for instance, in Kazakhstan (Dara et al., 2018;de Beurs & Henebry, 2004;Kraemer et al., 2015;Löw et al., 2015), Mongolia (Sankey et al., 2018), and Russia (Nguyen et al., 2018;Rogova & Skvortsov, 2014). Several studies have also quantified the decline in livestock numbers, including a decrease in grazing pressure in Kazakhstan Hankerson et al., 2019;Kerven et al., 2016;Kraemer et al., 2015) while grazing pressure increased in Mongolia (Maasri & Gelhaus, 2011). ...
... The United States Geological Survey (USGS) optical 30-m Landsat programme and the recent European Space Agency Copernicus optical 10-and 20-m Sentinel-2 satellite programmes are particularly relevant for monitoring land-use and land-cover change (LULCC) in steppe landscapes due to their unprecedented spatial resolution, revisit time, and available sensors and bands, which are wellsuited to mapping the transformation of broad land-cover classes in steppe landscapes (Wulder et al., 2016). For instance, Landsat imagery helped to reveal massive agricultural land abandonment in the steppes of northern Kazakhstan (Dara et al., 2018;Kraemer et al., 2015;Löw et al., 2015) and Mongolia (Sankey et al., 2018), the forest-steppe region of Siberia, Russia (Nguyen et al., 2018) and war-induced steppe recovery due to cropland abandonment in Ukraine (Skakun et al., 2019). Landsat imagery has also been used to assess grassland degradation and recovery due to the changed livestock herding regimes in northern Kazakhstan and Turkmenistan Kaplan et al., 2014). ...
Over the 20th century, the Eurasian steppes underwent drastic land‐cover changes. Much progress was made studying cropland expansion and the post‐1990 (i.e., post‐Soviet) agricultural land abandonment in Eurasia. However, the alteration of steppe landscapes may include other disturbances, such as oil and gas development, formal and informal roads, garbage dumps, which were not systematically documented. Considering the example of the steppe Orenburg province in Russia, we reconstructed agricultural land‐cover change dynamics using Landsat and Sentinel‐2 imagery from 1990 to 2018. Furthermore, we used very high‐resolution imagery and assessed the patterns and determinants of other steppe landscape anthropogenic disturbances. Our study showed that, despite steppe recovery due to widespread cropland abandonment from 1990 to 2018, the steppes, including the recovered steppe patches, underwent fragmentation due to informal roads, oil and gas development, shrub encroachment, and garbage dumps. Only 6.4% of the sampled 7,859 1*1 km blocks in 2018 grassland extent had no documented disturbances. The mapped disturbances occurred primarily near settlements and roads, while some disturbances occurred in remote areas. Given the accessibility of steppes, our study calls for a strong need to systematically document alternatives to agricultural land uses in the steppes of Eurasia and other parts of the global grassland biome. This article is protected by copyright. All rights reserved.
... The snow-series images were used to first classify SCA across the study site in each image and subsequently delineate persistent snow cover patches. All images (n = 11) were classified into five classes: vegetation, sunlit bare ground, shaded bare ground, sunlit snow, and shaded snow, using a random forest supervised classification performed in Google Earth Engine [60,61]. The random forest classifier was parameterized based on previous remotely sensed image classification examples, with the number of trees set to 100 and the number of variables per split set to the square root of the number of variables [60,61]. ...
... All images (n = 11) were classified into five classes: vegetation, sunlit bare ground, shaded bare ground, sunlit snow, and shaded snow, using a random forest supervised classification performed in Google Earth Engine [60,61]. The random forest classifier was parameterized based on previous remotely sensed image classification examples, with the number of trees set to 100 and the number of variables per split set to the square root of the number of variables [60,61]. The training dataset consisted of manually digitized polygons (n = 492 total) for each of the five classes. ...
Seasonal snow cover in the dry forests of the American West provides essential water resources to both human and natural systems. The structure of trees and their arrangement across the landscape are important drivers of snow cover distribution across these forests, varying widely in both space and time. We used unmanned aerial vehicle (UAV) multispectral imagery and Structure-from-Motion (SfM) models to quantify rapidly melting snow cover dynamics and examine the effects of forest structure shading on persistent snow cover in a recently thinned ponderosa pine forest. Using repeat UAV multispectral imagery (n = 11 dates) across the 76 ha forest, we first developed a rapid and effective method for identifying persistent snow cover with 90.2% overall accuracy. The SfM model correctly identified 98% (n = 1280) of the trees, when compared with terrestrial laser scanner validation data. Using the SfM-derived forest structure variables, we then found that canopy shading associated with the vertical and horizontal metrics was a significant driver of persistent snow cover patches (R2 = 0.70). The results indicate that UAV image-derived forest structure metrics can be used to accurately predict snow patch size and persistence. Our results provide insight into the importance of forest structure, specifically canopy shading, in the amount and distribution of persistent seasonal snow cover in a typical dry forest environment. An operational understanding of forest structure effects on snow cover will help drive forest management that can target snow cover dynamics in addition to forest health.
... Our study includes a relatively proportionate distribution of field-based training samples, where rare species are represented by fewer samples, whereas the dominant species are represented by more samples. In contrast, previous studies with coarse resolution satellite images disproportionately oversampled the rare cover types to improve detection accuracies of such classes (Congalton and Green, 2002;Massey et al., 2018;Sankey et al., 2018). Such a strategy would require greater numbers of field-based samples (Foody, 2002;Olofsson et al., 2014), which would also require much greater image extent covered by the UAV platform. ...
Rangelands cover 70% of the world's land surface, and provide critical ecosystem services of primary production, soil carbon storage, and nutrient cycling. These ecosystem services are governed by very fine-scale spatial patterning of soil carbon, nutrients, and plant species at the centimeter-to-meter scales, a phenomenon known as “islands of fertility”. Such fine-scale dynamics are challenging to detect with most satellite and manned airborne platforms. Remote sensing from unmanned aerial vehicles (UAVs) provides an alternative option for detecting fine-scale soil nutrient and plant species changes in rangelands tn0020 smaller extents. We demonstrate that a model incorporating the fusion of UAV multispectral and structure-from-motion photogrammetry classifies plant functional types and bare soil cover with an overall accuracy of 95% in rangelands degraded by shrub encroachment and disturbed by fire. We further demonstrate that employing UAV hyperspectral and LiDAR fusion greatly improves upon these results by classifying 9 different plant species and soil fertility microsite types (SFMT) with an overall accuracy of 87%. Among them, creosote bush and black grama, the most important native species in the rangeland, have the highest producer's accuracies at 98% and 94%, respectively. The integration of UAV LiDAR-derived plant height differences was critical in these improvements. Finally, we use synthesis of the UAV datasets with ground-based LiDAR surveys and lab characterization of soils to estimate that the burned rangeland potentially lost 1474 kg/ha of C and 113 kg/ha of N owing to soil erosion processes during the first year after a prescribed fire. However, during the second-year post-fire, grass and plant-interspace SFMT functioned as net sinks for sediment and nutrients and gained approximately 175 kg/ha C and 14 kg/ha N, combined. These results provide important site-specific insight that is relevant to the 423 Mha of grasslands and shrublands that are burned globally each year. While fire, and specifically post-fire erosion, can degrade some rangelands, post-fire plant-soil-nutrient dynamics might provide a competitive advantage to grasses in rangelands degraded by shrub encroachment. These novel UAV and ground-based LiDAR remote sensing approaches thus provide important details towards more accurate accounting of the carbon and nutrients in the soil surface of rangelands.
... However, by building a large and dense time-series images from satellite data, we might detect gradual trends in forest health, vigor, and canopy moisture, and thus resiliency. Google Earth Engine (GEE) provides a unique opportunity to synthesize such trends with its large archives of Landsat satellite images and high-performance computing capabilities (Gorelick et al., 2017;Sankey et al., 2018). GEE offers complex analysis capabilities for image classification, change detection, time-series analysis, and vector-based extraction of image statistics (Erickson, 2014;Hancher, 2013;Moore & Hansen, 2011). ...
... GEE offers complex analysis capabilities for image classification, change detection, time-series analysis, and vector-based extraction of image statistics (Erickson, 2014;Hancher, 2013;Moore & Hansen, 2011). As a result, GEE has been successfully used in several largescale studies (Dong et al., 2015;Hansen et al., 2013;Johansen et al., 2015;Massey et al., 2018;Patel et al., 2015;Sankey et al., 2018;Simonetti et al., 2015;Teluguntla et al., 2017). ...
... The computational efficiency of GEE combined with the complete archives of 40 years of pre-processed Landsat data and the GRID-MET surface meteorological dataset allowed the entire workflow and data merging within the same consistent environment. Several other studies have similarly produced Landsat-based analysis at regional and continental scales using GEE (Azzari and Lobell, 2017;Dong et al., 2015;Massey et al., 2018;Patel et al., 2015;Sankey et al., 2018;Simonetti et al., 2015;Padarian et al., 2015). We first evaluated long-term trends in annual total precipitation across our study region and observed that much of the study region (up to 50% of the region in some seasons) was already experiencing a significantly decreasing trend in annual total precipitation prior to treatments. ...
Large‐scale changes in forest structure and ecological function throughout western North America have led to increased frequency, size, and severity of wildfires. The US Forest Service is implementing state‐wide forest restoration initiatives to reduce wildfire hazards and improve forest health. We provide a synthesis of pre‐ and post‐treatment forest vegetation and ecosystem moisture trends between 1990 and 2017 in Arizona, the first US state where the initiative has implemented a variety of thinning and burning methods in over 1,200 polygon areas across 3.5 million ha. Using 4,426 Landsat satellite images on Google Earth Engine, we calculated normalized difference moisture index (NDMI), normalized difference water index (NDWI), and normalized difference vegetation index (NDVI) to create dense time‐series datasets. The indices and 1990‐2017 annual total precipitation dataset were then examined using a Mann–Kendall tau test to identify statistically significant upward and downward trends for each pixel. Our results indicate that much of the study region were experiencing drought conditions prior to restoration treatments and NDVI values were significantly decreasing, especially during the dry spring season. However, both NDMI and NDWI trends indicate that the forest restoration treatments have contributed to increased total ecosystem moisture, while precipitation in the post‐treatment period exhibit stable trends. Forest restoration treatments appear to have improved the overall forest health and resiliency to drought, especially during the dry spring season, when forests are most vulnerable to water stress and wildfire risks. Our results of the spatial patterns and long‐term trends in these variables can inform the currently ongoing and future restoration treatments to better target the treatment strategy across the southwestern USA. Google Earth Engine enabled our synthesis of these long‐term trends over the large region and will enhance our continued monitoring in the coming decade. We provide a synthesis of the regional‐scale forest restoration treatments implemented across 3.5 million ha in Arizona, USA. Our analysis of over 4,400 Landsat images on Google Earth Engine indicates that forest thinning and burning treatments have significantly improved forest health and resiliency to drought.
... This kind of abandonment often occurs in the aftermath of substantial rural depopulation, perhaps because of rural-urban migration and an aging labor force (Feldhoff, 2012). Previous research has shown that temporal variations in cropland abandonment are mostly influenced by a range of socioeconomic factors (Rey Benayas, 2007;Sankey et al., 2018;Ustaoglu & Collier, 2018), including sociopolitical regime changes (Sankey et al., 2018), industrialization (Rey , shifts in farming practices and agricultural trade policy reforms (Renwick et al., 2013), as well as regulations for nature conservation (Lambin et al., 2014;Renwick et al., 2013). In one example, following the collapse of socialism, many rural areas in the former Soviet Union experienced large rural exoduses and agricultural commodity production declines as a consequence of economic transition from state-run to market-driven. ...
... This kind of abandonment often occurs in the aftermath of substantial rural depopulation, perhaps because of rural-urban migration and an aging labor force (Feldhoff, 2012). Previous research has shown that temporal variations in cropland abandonment are mostly influenced by a range of socioeconomic factors (Rey Benayas, 2007;Sankey et al., 2018;Ustaoglu & Collier, 2018), including sociopolitical regime changes (Sankey et al., 2018), industrialization (Rey , shifts in farming practices and agricultural trade policy reforms (Renwick et al., 2013), as well as regulations for nature conservation (Lambin et al., 2014;Renwick et al., 2013). In one example, following the collapse of socialism, many rural areas in the former Soviet Union experienced large rural exoduses and agricultural commodity production declines as a consequence of economic transition from state-run to market-driven. ...
Cropland abandonment because of rural depopulation or policy interventions has become a key issue in Chinese mountainous areas. One such region is the Guangxi Karst Mountainous Area (GKMA), a zone where more than 59% of total land area is hilly. Although depopulation and declining agriculture since 2000 within the GKMA have led to vast areas of abandoned cropland, the spatiotemporal distribution that underlies this pattern as well as its causes remain little understood. We therefore estimated the extent of cropland abandonment since 2001 within this region using land use trajectories derived from Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) data alongside phenology metrics. The results of this analysis show that 4.89% (0.56 × 10⁴ km²) of land within the GKMA has been abandoned since 2001. Specifically, within four sub-periods (between 2001 and 2005, between 2005 and 2010, between 2011 and 2015, and between 2016 and 2019), overall trends were characterized by an initial increase and then slight decrease in cropland abandonment rate (CAR); 10.67%, 14.71%, 23.13%, and 21.26% over these time periods, respectively. Data also show that CAR spatial distribution tends to be similar to adjacent areas. We then utilized a geographical detector model (GDM) and Spearman correlation analysis to assess the effects of environmental and socioeconomic variables on cropland abandonment patterns. Data show that environmental factors have generally exerted a more significant effect on cropland abandonment within the GKMA than socioeconomic variables. Interactions between determinants have also either exerted non-linear-enhanced or bi-enhanced effects on cropland abandonment over time.
