First, we detected and quantified human activity expansions in highlands (mainly referring to the expansions of cultivated lands and artificial surfaces in highlands) from 2000 to 2020 in 48 Asian countries, including East Asia (China, Mongolia, North Korea, South Korea and Japan), Southeast Asia (Philippines, Vietnam, Laos, Cambodia, Myanmar, Thailand, Malaysia, Brunei, Singapore, Indonesia and East Timor), South Asia (Nepal, Bhutan, Bangladesh, India, Pakistan, Sri Lanka and Maldives), Central Asia (Kazakhstan, Kyrgyzstan, Tajikistan, Uzbekistan and Turkmenistan) and West Asia (Afghanistan, Iraq, Iran, Syria, Jordan, Lebanon, Israel, Palestine, Saudi Arabia, Bahrain, Qatar, Kuwait, United Arab Emirates, Oman, Yemen, Georgia, Armenia, Azerbaijan, Turkey and Cyprus). We then investigated the visible effects of human activity expansions in the highlands (mainly referring to the ecological land loss and habitat fragmentation, the contribution rate of cultivated land net growth and population capacity of artificial surface expansion in highlands).
Digital elevation model (DEM)-derived landforms
Freely available Advanced Spaceborne Thermal Emission and Reflection Radiometer Global DEM (ASTER GDEM) products with WGS-84 coordinate system and 30 m spatial resolution, which were captured around or after 2000 in Asia, were downloaded from the NASA website (http://reverb.echo.nasa.gov/reverb/). ASTER GDEM was jointly developed by the Ministry of Economy Trade and Industry (METI) of Japan and NASA of the USA by stacking all individual cloud- and non-cloud-masked scene DEMs while applying various algorithms to remove abnormal data40. ASTER GDEM products have been widely used in various fields, and their reliability have been proven41. We used ASTER GDEM (3935 tiles) to derive the major landforms (i.e. lowland and highland)42 in Asia as the base map for subsequent analyses. Lowland includes plains and terraces, while highland refers to the collection of upland (hills) and montane (mountains). The classification criteria of highland and lowland are based on elevation and slope (see details in Supplementary Table 10), which is summarised by Margono and colleagues according to large previous studies42. The spatial distributions of lowlands and highlands in Asia were identified with the batch processing module of ArcGIS 10.2 platform. Overall, highlands and lowlands are widely distributed in Asia, and their areas accounted for about 40.41% and 59.59%, respectively (Fig. 1).
Satellite-based high-resolution land cover products
Some freely available global land cover products with fine spatial resolutions (≤500 m) exist43,44, including the MODIS Land Cover Type (MLCT) series products from 2001 to 2016 (500 m), ESA Climate Change Initiative (ESA-CCI) land cover product from 1992 to 2015 (300 m), Finer Resolution Observation and Monitoring of Global Land Cover product (FROM-GLC) in 2010, 2015 and 2017 (30 m and 10 m), global 30 m land cover classification with a fine classification system (GLC_FCS30) in 2015 and 2020 (30 m) and global land cover (Globeland30) data product in 2000, 2010 and 2020 (30 m). Since human activities need to be detected in detail (such as the activities in small hilly areas), we tested all these products and found that the minimum geographical unit (a pixel) of MLCT (500 × 500 m2) and ESA-CCI (300 × 300 m2) could not completely detect some small hills, which may underestimate human activity areas. Moreover, FROM-GLC and GLC_FCS30 are limited by historical archives and are thus not suitable for studying long-term land cover change. Thus, we selected Globeland30 as the land cover data to quantify the human activity expansions in the highlands. Globeland30 datasets (http://www.globallandcover.com/) have the same coordinate and projection systems as DEM, and they were developed by combining several multispectral images, including Landsat Thematic Mapper (TM5) and Enhanced Thematic Mapper Plus (ETM + ), Chinese HJ-1 images and 16 m Gaofen (GF-1) multispectral images (2020 version). The classification was performed using a split-and-merge strategy, Pixel–Object–Knowledge approach was applied to classify each land cover type and a knowledge-based interactive verification procedure was applied to improve mapping accuracy45. Globeland30 datasets cover 10 land cover types: cultivated land, artificial surface, forest, grassland, shrubland, wetland, water, tundra, bare land, glaciers and permanent snow. The accuracy of Globeland30 was assessed using the landscape shape index-based sampling model and more than 230 thousand samples, and the overall accuracy and kappa coefficient, respectively, were 85.72% and 0.82 for the 2020 version and about 83.50% and 0.78 for the previous versions45,46, which meets the accuracy requirement of land cover change analysis47. Consequently, the Globeland30 products have been extensively used for a variety of applications48,49.
Socio-economic datasets
Socioeconomic datasets include population, economic level and high-resolution population density for all 48 Asian counties. Population and economic levels were collected from the World Bank database (https://data.worldbank.org.cn/) (see Supplementary Data 1). The World Bank divides the economic levels of different countries into low, lower-middle, upper-middle and high income based on the gross national income per capita. According to this criterion, Asia covers 12 high, 15 upper-middle, 16 lower-middle and 5 low-income countries (Supplementary Data 1). The high-resolution population density gridded dataset from 2020 with 100 m resolution was obtained from WorldPop (www.worldpop.org/project/categories?id=3). This dataset is currently the most accurate estimation of population density, and it includes constrained and unconstrained type data. In this study, we employed constrained type data as they were developed using a top-down constrained method and Random-Forests-based dasymetric redistribution (i.e. the output yields a more accurate population distribution without the prediction of small population numbers in likely uninhabited areas as it employs building footprints and/or built settlements as inputs)50,51. In contrast to unconstrained type data, the residential populations of small settlements and isolated buildings were identified in constrained type data50. WorldPop’s population density dataset holds the same coordinate system as DEM and Globeland30, and its unit is the number of people per pixel with country totals adjusted to match the corresponding official report of the Population Division of the Department of Economic and Social Affairs of the UN Secretariat. We applied the nearest neighbourhood algorithm to resample the 100 m resolution population density dataset to 30 m to match the spatial resolution of DEM and GlobeLand30.
Administrative boundary
The administrative boundary data of the 48 Asian countries were obtained from the Database of Global Administrative Areas (GADM) (https://gadm.org/) hosted by the University of California at Davis. The GADM data provide high-resolution shape files at administrative levels, including country and state or provincial level. The data with the latest version (v.3.6) were used in our study.
Measuring highland human activity expansions
We selected the expansions of cultivated lands and artificial surfaces as the representatives of human activity expansions. According to the classification system of GlobeLand30, cultivated land denotes the land used for planting crops, including paddy fields, irrigated dry land, rain-fed dry land, vegetable land, pasture land, greenhouse land and land mainly for planting crops with fruit trees and other economic trees and economic crops, such as tea and coffee gardens46. Artificial surface is impervious surface formed by artificial construction activities, including various residential settlements, such as towns, industrial and mining areas and transportation facilities, but excluding contiguous green spaces and water bodies within construction lands46. Therefore, based on the above definitions, the expansions of cultivated lands and artificial surfaces cover almost all intensive human activities that may occur in the highlands. To improve computational efficiency, the following strategies were applied. The Python platform with a target-level change detection method was applied to detect the change regions (i.e. changed pixels) of cultivated lands and artificial surfaces in Asia from 2000 to 2020, including increased (i.e. expanded areas of human activities), unchanged and decreased areas. The human activity expansions in lowlands and highlands were then classified and quantified according to landform information and change detection results. The human activity expansions in the highlands include the expansions of cultivated lands and artificial surfaces in the upland and montane (Fig. 1). Finally, we compared the rates of human activity expansions in the Asian highlands (i.e. highland/(lowland +highland) × 100%).
Assessing the effects of human activity expansions in highlands
This study quantified four human activity expansion effects in Asian highlands: ecological land loss (i.e. forest, grassland, wetland and shrub, which serve as fundamental ecological lands52), habitat fragmentation of ecological land, the contribution rate of cultivated land net growth in highland to the total cultivated land conservation and population capacity of artificial surface expansion. The ecological land loss in highlands induced by human activity expansions denotes the areas of ecological lands in highlands at time T that are transformed into cultivated land/artificial surfaces at T + 1. A transfer matrix with Python and ArcGIS was developed to detect the ecological land loss in the highlands. To calculate the habitat fragmentation degrees of ecological lands in the highlands caused by human activity expansions, a habitat fragmentation index (HFI) (Eq. 1) was employed. HFI can reflect the fragmentation degrees of habitat areas from a patch perspective. The HFI value varies from 0 to 1, 0 and 1 imply no and complete habitat fragmentation, respectively, and a value closer to 1 indicates a high habitat fragmentation and a strong impact of human activities53. The HFI values from 2000 to 2020 were calculated with Python and ArcGIS (Supplementary Fig. 10).
$$HF{I}_{i}=1-\,\max ({A}_{ij})/\mathop{\sum }\nolimits_{j=1}^{{N}_{i}}{A}_{ij}$$
(1)
where \(HF{I}_{i}\) denotes the habitat fragmentation degree of landscape class i (i refers to the ecological land in the highlands), \({A}_{ij}\) is the area of patch j of landscape class i and \({N}_{i}\) denotes the total patch mumble of landscape class i.
The contribution rate of highland cultivated land net growth to the total cultivated land conservation and population capacity of artificial surface expansion in the highlands were estimated using Eqs. 2 and 3, respectively. The contribution rate of the cultivated land net growth in the highlands for conserving the total net increase of cultivated lands was calculated by dividing the net growth area of highland cultivated land by the total net growth area of cultivated land (Eq. 2; Supplementary Tables 5 and 6). The population capacity of artificial surface expansions in the highlands was estimated by individually matching the population density pixels with artificial surface expansion area pixels (Eq. 3; Supplementary Tables 7 and 8).
$$Cr=({H}_{T+1}-{H}_{T})/({(H+L)}_{T+1}-{(H+L)}_{T})$$
(2)
$$Pc\,=\,w\mathop{\sum }\limits_{i=1}^{n}(N1+N2+N3+\cdot \cdot \cdot Nn)$$
(3)
where \(Cr\) is the contribution rate of cultivated land net growth in the highlands for conserving the total net growth of cultivated lands (a net growth of cultivated land occurs in the highlands and the total cultivated land is a net growth) or the contribution rate of cultivated land net growth in the highlands for preventing the net loss of total cultivated lands (a net growth of cultivated land occurs in highlands and the total cultivated land growth is negative); \({H}_{T+1}\) and \({H}_{T}\) represent the cultivated land area in highlands at time T + 1 and T, respectively; \({L}_{T+1}\)and \({L}_{T}\)represent the cultivated land area in lowlands at time T + 1 and T, respectively. \(Pc\) is the total population capacity; \(w\) represents the sampling coefficient of population density data; \(n\) is the class of the population density pixel, and \(N\) is the total number of pixels with different population density classes. Since the population density with 100 m resolution was sampled to 30 m, the sampling coefficient was 0.09 (\(w\) = 0.09) in this study.
Accuracy evaluation and analysis
A strategy was designed in our study to verify the result reliability. To assess human activity expansion results, we randomly selected 4008 lowland and 6676 highland (3420 in hills and 3256 in mountains) samples using random function (the sampling interval was a 5 × 5 km2 grid) from the human activity expansion areas in Asia (cultivated and artificial land expansions each accounted for about 50% of the sample size) (Supplementary Data 2 and 3; Supplementary Fig. 6). These samples were evenly distributed across the 48 Asian countries. All samples were tagged geographically with Keyhole Markup Language and placed into Google Earth, and the available historical high-resolution images of Google Earth, Landsat TM5 and ETM + satellite imagery from 2000 (around 2000 and cloud-free) were applied to interpret land cover types at all sample sites (total 10,684 samples; Supplementary Data 2 and 3). The collected human activity expansion samples using cloudless high-resolution (≤5 m) imageries around or after 2020 available in the Google Earth, Planet (www.planet.com), in situ measurements (Geo-Wiki, OpenStreetMap, Tencent and Baidu StreetMap) and other public geographic data (Geospatial Data Cloud) were examined to visually interpret whether each sample site was a cultivated land or an artificial surface (Supplementary Data 2 and 3). The visual interpretations of images were performed by a team with specialised knowledge and training in standard procedures and risk assessment. The samples were labelled as human activity expansions if they were a cultivated land or an artificial surface in 2020 but not in 2000 (i.e. correct results); otherwise, they were labelled as non-human activity expansions (e.g. human activity shrinks, stable human activities and stable non-human activities) (i.e. incorrect results) (Supplementary Data 2 and 3). The extremely high resolution satellite images provided by the above platforms include EarlyBird-1 (0.8–3.0 m), IKONOS (3.2 m), QuickBird (0.6–2.4 m), GeoEye-1 (0.41–1.65 m), WorldView (0.25–1.84 m), Pleiades-1A (0.5–2.0 m), Pleiades-1B (0.5–2.0 m), SPOT-6 and SPOT-7 (1.5–6.0 m), RapidEye (5 m), Doves (3 m; 4-band PlanetScope Scene), Gaofen-1 (2 m/8 m panchromatic band or multispectral bands), Chinese Gaofen-2(1–4 m), Chinese ZY1-02C HRC (2.36 m) and ZY-3 (2.1–5.8 m).
We repeated the sample selection of ecological land loss areas induced by human activity expansions, and finally 4474 lowland and 6275 highland samples of ecological land loss areas (the sampling interval was a 5 × 5 km2 grid; hills and mountains each held about 50% of the entire sample size) were randomly selected across Asia (the samples were evenly distributed in forest, grassland, wetland and shrubland loss areas) (Supplementary Data 4 and 5; Supplementary Fig. 7). To each sample, we checked all the cloudless historical high-resolution images collected in 2000 in Google Earth, Landsat TM5 and ETM + satellite imagery to determine whether the sample site comprised forests, grasslands, wetlands or shrublands (total 10,749 samples; Supplementary Data 4 and 5). Then, we examined cloudless high-resolution images collected around or after 2020 from Google Earth, Planet, in situ measurements and public geographic data to visually determine whether the sample sites comprised cultivated lands or artificial surfaces (Supplementary Data 4 and 5). A sample was labelled as ecological land loss induced by human activity expansion if it was an ecological land (i.e. forest, grassland, wetland or shrub) in 2000 but a cultivated land or an artificial surface around 2020 (i.e. correct results); otherwise, the sample was labelled as the non-ecological land loss or ecological land loss induced by non-human activity expansion (i.e. incorrect results) (Supplementary Data 4 and 5). Finally, the number of valid samples were accounted (the samples correctly identified after excluding uncertain or unknown samples), and the overall accuracy metrics54 (Eq. 4) were calculated to assess detection accuracy (human activity expansion in the highlands and lowlands in Supplementary Table 1; human activity expansion-induced ecological land loss in highlands and lowlands in Supplementary Table 2).
$$OA={\sum }_{i=1}^{n}{P}_{i}/P$$
(4)
where OA is overall accuracy (i.e. the accuracy of correct detection), \({P}_{i}\) represents the samples whose validation results are consistent with classification results (i.e. the correct samples) and P is the total number of samples.
Learn Crypto Trading, Yield Farms, Income strategies and more at CrytoAnswers
https://nov.link/cryptoanswers
Comments are closed.