Search

Wetlands are important CO2 sinks and methane sources, and their seasonality and phenological cycle play an essential role in understanding the carbon budget. However, given the spatial heterogeneity of wetland landscapes and the coarser spatial resolution of satellites, the phenological retrievals of wetlands are challenging. Here we examined the phenology of wetlands from 30 m harmonized Landsat/Sentinel-2 (LandSent30) and 500 m MODIS satellite observations using the ground phenology network PhenoCam as a benchmark. This study used all 11 available wetland PhenoCam sites (about 30 site years), covering diverse wetland types from different climate zones. We found that the LandSent30-based phenology results were in overall higher consistency with the PhenoCam results compared to MODIS, which could be related to the better explanation capacity of LandSent30 data in the heterogeneous landscapes of wetlands. This also means that the LandSent30 has an advantage over the 500 m MODIS regarding wetland vegetation phenological retrievals. It should be noted that the LandSent30 did not show a greatly improved performance, which could be related to the specificity and complexity of the wetlands landscape. We also illustrated the potential effects of the location and observation direction of PhenoCam cameras, the selection of Region of Interest (ROI), as well as the landscape composition of the site. Overall, this study highlights the complexity of wetland phenology from both ground and remote sensing observations at different scales, which paves the road for understanding the role of wetlands in global climate change and provides a basis for understanding the real phenological changes of wetland surfaces.

Deciduous broadleaf forests (DBF) are an extremely widespread vegetation type in the global ecosystem and an indicator of global environmental change; thus, they require accurate phenological monitoring. However, there is still a lack of systematic understanding of the sensitivity of phenological retrievals for DBF in terms of different spatial resolution data and proxy indices. In this study, 79 globally distributed DBF PhenoCam Network sites (total 314 site-years, 2013–2018) were used as the reference data (based on green chromaticity coordinates, GCC). Different spatial resolutions (30 m Landsat and Sentinel-2 data, and 500 m MCD43A4 data) and satellite remote sensing vegetation indices (normalized difference vegetation index, NDVI; enhanced vegetation index, EVI; and near-infrared reflectance of vegetation, NIRV) were compared to find the most suitable data and indices for DBF phenological retrievals. The results showed that: (1) for different spatial resolutions, both 30 m Landsat–Sentinel-2 data and 500 m MODIS data accurately captured (R2 > 0.8) DBF phenological metrics (i.e., the start of the growing season, SOS, and the end of the growing season, EOS), which are associated with the comparatively homogeneous landscape pattern of DBF; (2) for SOS, the NIRv index was closer to GCC than EVI and NDVI, and it showed a slight advantage over EVI and a significant advantage over NDVI. However, for EOS, NDVI performed best, outperforming EVI and NIRv; and (3) for different phenological metrics, the 30 m data showed a significant advantage for detecting SOS relative to the 500 m data, while the 500 m MCD43A4 outperformed the 30 m data for EOS. This was because of the differences between the wavebands used for GCC and for the satellite remote sensing vegetation indices calculations, as well as the different sensitivity of spatial resolution data to bare soil. This study provides a reference for preferred data and indices for broad scale accurate monitoring of DBF phenology.

Implementation of a measuring, reporting, and verifying (MRV) framework is essential for reducing emissions from deforestation and forest degradation (REDD+). According to the United Nations Framework Convention on Climate Change, MRV can be regarded as an important mechanism to mitigate global warming. Upland Myanmar, with an elevation of ~80–2600 m, is experiencing tropical deforestation, which is commonly explained by the expansion of shifting cultivation. The vegetation change tracker algorithm, with its high-automation and wild-adaptation features, and the enhanced integrated forest z-score were applied in this elevation-based study of time series deforestation monitoring in upland Myanmar using data from 2003 to 2015. Four spatial patterns of deforestation, namely stripes, adjacent, filled, and staggered, were found in the research area. Moreover, our work showed that the center of elevation of deforestation was ~1000 m. Further analysis revealed that this center tended to shift to a higher elevation over time; a “golden cross”/changeover could be deciphered at ~1000 m, indicating that the scale and intensity of shifting cultivation continue to expand vertically. The results suggest the need to track the elevation-based signature of vegetation clearings to help achieve the goals of REDD+ at the regional level in tropical rainforest countries.

Urban green space (e.g., parks, farmland, gardens, etc.) design in different urban functional regions (e.g., residential land, commercial land, etc.) depends on different planning purposes. The changes in urban green spaces are highly related to urban land-use changes (e.g., from residential land to commercial land). However, the investigation of urban green space patterns in response to urban land-use changes has been ignored. This research takes Hangzhou city, a typical example in terms of urbanization, population growth, economic development, and land-use changes, as the study site, aiming to explore the landscape patterns of urban green space changes with different urban land-use changes. The results showed that urban green spaces increased from 2017 to 2021, and the growth was mainly concentrated in the urban core area, indicating that the city has made remarkable achievements in the planning of green spaces. Specifically, the increase in urban green spaces in the first ring belt was mainly related to the old town transformation program in the residential land. The change from the residence parcels to the business parcels determined the increase in green spaces in the second and third ring belts, probably because of the attractiveness of customers. In addition, a large number of open-space parcels have changed to business parcels around the urban periphery, which might be due to the transformation from farmland to impervious surfaces. Combined with the urban land-use and green-space policies, the findings highlighted that a reasonable urban land-use layout can promote the optimization and layout of urban green spaces. The private sector (e.g., shopping malls) can also contribute to the increase in green spaces. The understanding of urban green landscapes with different urban land-use changes can provide references for analyzing and optimizing green space in other cities experiencing rapid urban land-use changes.

With advancements in big geospatial data and artificial intelligence, multi-source data and diverse data-driven methods have become common in dengue risk prediction. Understanding the current state of data and models in dengue risk prediction enables the implementation of efficient and accurate prediction in the future. Focusing on predictors, data sources, spatial and temporal scales, data-driven methods, and model evaluation, we performed a literature review based on 53 journal and conference papers published from 2018 to the present and concluded the following. (1) The predominant predictors include local climate conditions, historical dengue cases, vegetation indices, human mobility, population, internet search indices, social media indices, landscape, time index, and extreme weather events. (2) They are mainly derived from the official meteorological agency satellite-based datasets, public websites, department of health services and national electronic diseases surveillance systems, official statistics, and public transport datasets. (3) Country-level, province/state-level, city-level, district-level, and neighborhood-level are used as spatial scales, and the city-level scale received the most attention. The temporal scales include yearly, monthly, weekly, and daily, and both monthly and weekly are the most popular options. (4) Most studies define dengue risk forecasting as a regression task, and a few studies define it as a classification task. Data-driven methods can be categorized into single models, ensemble learning, and hybrid learning, with single models being further subdivided into time series, machine learning, and deep learning models. (5) Model evaluation concentrates primarily on the quantification of the difference/correlation between time-series observations and predicted values, the ability of models to determine whether a dengue outbreak occurs or not, and model uncertainty. Finally, we highlighted the importance of big geospatial data, data cloud computing, and other deep learning models in future dengue risk forecasting.

Convolutional neural network (CNN)-based remote sensing (RS) image segmentation has become a widely used method for building footprint mapping. Recently, DeeplabV3+, an advanced CNN architecture, has shown satisfactory performance for building extraction in different urban landscapes. However, it faces challenges due to the large amount of labeled data required for model training and the extremely high costs associated with the annotation of unlabelled data. These challenges encouraged us to design a framework for building footprint mapping with fewer labeled data. In this context, the published studies on RS image segmentation are reviewed first, with a particular emphasis on the use of active learning (AL), incremental learning (IL), transfer learning (TL), and their integration for reducing the cost of data annotation. Based on the literature review, we defined three candidate frameworks by integrating AL strategies (i.e., margin sampling, entropy, and vote entropy), IL, TL, and DeeplabV3+. They examine the efficacy of AL, the efficacy of IL in accelerating AL performance, and the efficacy of both IL and TL in accelerating AL performance, respectively. Additionally, these frameworks enable the iterative selection of image tiles to be annotated, training and evaluation of DeeplabV3+, and quantification of the landscape features of selected image tiles. Then, all candidate frameworks were examined using WHU aerial building dataset as it has sufficient (i.e., 8188) labeled image tiles with representative buildings (i.e., various densities, areas, roof colors, and shapes of the building). The results support our theoretical analysis: (1) all three AL strategies reduced the number of image tiles by selecting the most informative image tiles, and no significant differences were observed in their performance; (2) image tiles with more buildings and larger building area were proven to be informative for the three AL strategies, which were prioritized during the data selection process; (3) IL can expedite model training by accumulating knowledge from chosen labeled tiles; (4) TL provides a better initial learner by incorporating knowledge from a pre-trained model; (5) DeeplabV3+ incorporated with IL, TL, and AL has the best performance in reducing the cost of data annotation. It achieved good performance (i.e., mIoU of 0.90) using only 10–15% of the sample dataset; DeeplabV3+ needs 50% of the sample dataset to realize the equivalent performance. The proposed frameworks concerning DeeplabV3+ and the results imply that integrating TL, AL, and IL in human-in-the-loop building extraction could be considered in real-world applications, especially for building footprint mapping.

The principal difficulty in extracting urban surface water using remote-sensing techniques is the influence of noise from complex urban environments. Although various methods exist, there are still many sources of noise interference when extracting urban surface water, and automatic cartographic methods with long time series are especially scarce. Here, we construct an automatic urban surface water extraction method from the combination of traditional water index, urban shadow index (USI), and land surface temperature (LST) by using the Google Earth Engine cloud computing platform and Landsat imagery. The three principal findings derived from the application of the method were as follows. (i) In comparison with autumn and winter, LST in spring and summer could better distinguish water from high-reflection ground objects, shadows, and roads and roofs covered by asphalt. (ii) The overall accuracy of Automated Water Extraction Index (AWEIsh) in Zhengzhou was 77.5% and the Kappa coefficient was 0.55; with consideration of the USI and LST, the overall accuracy increased to 96.0% and the Kappa coefficient increased to 0.92. (iii) During 1990–2020, the area of urban surface water in Zhengzhou increased, with an evident trend in expansion from 11.51 km2 in 2008 to 49.28 km2 in 2020. Additionally, possible omissions attributable to using 30m-resolution imagery to extract urban water areas were also discussed. The method proposed in this study was proven effective in eliminating the influence of noise in urban areas, and it could be used as a general method for high-accuracy long-term mapping of urban surface water.

The phenology-based approach has proven effective for paddy rice mapping due to the unique flooding and transplanting features of rice during the early growing season. However, the method may be greatly affected if no valid observations are available during the flooding and rice transplanting phase. Here, we compare the effects of data availability of different sensors in the critical phenology phase, thereby supporting paddy rice mapping based on phenology-based approaches. Importantly, our study further analyzed the effects of the spatial pattern of the valid observations related to certain factors (i.e., sideslips, clouds, and temporal window lengths of flooding and rice transplanting), which supply the applicable area of the phenology-based approach indications. We first determined the flooding and rice transplanting phase using in situ observational data from agrometeorological stations and remote sensing data, then evaluated the effects of data availability in this phase of 2020 in China using all Landsat-7 and 8 and Sentinel-2 data. The results show that on the country level, the number of average valid observations during the flooding and rice transplanting phase was more than ten for the integration of Landsat and Sentinel images. On the sub-country level, the number of average valid observations was high in the cold temperate zone (17.4 observations), while it was relatively lower in southern China (6.4 observations), especially in Yunnan–Guizhou Plateau, which only had three valid observations on average. Based on the multicollinearity test, the three factors are significantly correlated with the absence of valid observations: (R2 = 0.481) and Std.Coef. (Std. Err.) are 0.306 (0.094), −0.453 (0.003) and −0.547 (0.019), respectively. Overall, these results highlight the substantial spatial heterogeneity of valid observations in China, confirming the reliability of the integration of Landsat-7 and 8 and Sentinel-2 imagery for paddy rice mapping based on phenology-based approaches. This can pave the way for a national-scale effort of rice mapping in China while further indicating potential omission errors in certain cloud-prone regions without sufficient optical observation data, i.e., the Sichuan Basin.

Convolutional neural network (CNN)-based very high-resolution (VHR) image segmentation has become a common way of extracting building footprints. Despite publicly available building datasets and pre-trained CNN models, it is still necessary to prepare sufficient labeled image tiles to train CNN models from scratch or update the parameters of pre-trained CNN models to extract buildings accurately in real-world applications, especially the large-scale building extraction, due to differences in landscapes and data sources. Deep active learning is an effective technique for resolving this issue. This study proposes a framework integrating two state-of-the-art (SOTA) models, U-Net and DeeplabV3+, three commonly used active learning strategies, (i.e., margin sampling, entropy, and vote entropy), and landscape characterization to illustrate the performance of active learning in reducing the effort of data annotation, and then understand what kind of image tiles are more advantageous for CNN-based building extraction. The framework enables iteratively selecting the most informative image tiles from the unlabeled dataset for data annotation, training the CNN models, and analyzing the changes in model performance. It also helps us to understand the landscape features of iteratively selected image tiles via active learning by considering building as the focal class and computing the percent, the number of patches, edge density, and landscape shape index of buildings based on labeled tiles in each selection. The proposed method was evaluated on two benchmark building datasets, WHU satellite dataset II and WHU aerial dataset. Models in each iteration were trained from scratch on all labeled tiles. Experimental results based on the two datasets indicate that, for both U-Net and DeeplabV3+, the three active learning strategies can reduce the number of image tiles to be annotated and achieve good model performance with fewer labeled image tiles. Moreover, image tiles with more building patches, larger areas of buildings, longer edges of buildings, and more dispersed building distribution patterns were more effective for model training. The study not only provides a framework to reduce the data annotation efforts in CNN-based building extraction but also summarizes the preliminary suggestions for data annotation, which could facilitate and guide data annotators in real-world applications.

Timely and accurate forecasts of dengue cases are of great importance for guiding disease prevention strategies, but still face challenges from (1) time-effectiveness due to time-consuming satellite data downloading and processing, (2) weak spatial representation capability due to data dependence on administrative unit-based statistics or weather station-based observations, and (3) stagnant accuracy without the application of historical case information. Geospatial big data, cloud computing platforms (e.g., Google Earth Engine, GEE), and emerging deep learning algorithms (e.g., long short term memory, LSTM) provide new opportunities for advancing these efforts. Here, we focused on the dengue epidemics in the urban agglomeration of the Federal District of Brazil (FDB) during 2007–2019. A new framework was proposed using geospatial big data analysis in the Google Earth Engine (GEE) platform and long short term memory (LSTM) modeling for dengue case forecasts over an epidemiological week basis. We first defined a buffer zone around an impervious area as the main area of dengue transmission by considering the impervious area as a human-dominated area and used the maximum distance of the flight range of Aedes aegypti and Aedes albopictus as a buffer distance. Those zones were used as units for further attribution analyses of dengue epidemics by aggregating the pixel values into the zones. The near weekly composite of potential driving factors was generated in GEE using the epidemiological weeks during 2007–2019, from the relevant geospatial data with daily or sub-daily temporal resolution. A multi-step-ahead LSTM model was used, and the time-differenced natural log-transformed dengue cases were used as outcomes. Two modeling scenarios (with and without historical dengue cases) were set to examine the potential of historical information on dengue forecasts. The results indicate that the performance was better when historical dengue cases were used and the 5-weeks-ahead forecast had the best performance, and the peak of a large outbreak in 2019 was accurately forecasted. The proposed framework in this study suggests the potential of the GEE platform, the LSTM algorithm, as well as historical information for dengue risk forecasting, which can easily be extensively applied to other regions or globally for timely and practical dengue forecasts.

Plantations of Panax notoginseng (PN), traditional herbal medicine for the prevention and treatment of vascular diseases, are expanding rapidly in China, especially in the Yunnan province of China, due to its increasing demands and prices and causing dramatic environmental concerns. However, existing information on its planting area and spatial distribution are limited. Here, we mapped the PN planting area by using a new integrated pixel- and object-based (IPOB) approach, the Random Forest (RF) classifier, and the high-resolution ZiYuan-3 (ZY-3) imagery. We improved the procedures of classification in three aspects: (1) a new spectral index—Normalized Difference PN Index (NDPI)—was proposed, (2) the efficiency and scale of segmentation were optimized by using the Bi-level Scale-sets Model (BSM), and (3) feature variables were selected through an iteration analysis from 99 feature variables (spectral, textural, geometric, and geographic). Compared with the pixel- and the object-based methods, the IPOB has the highest F1 score of 0.98 and also has high robustness in terms of user and producer accuracies (97% and 99%, respectively), following by the object-based method (F1 = 0.94) and the pixel-based method (F1 = 0.93). The high accuracy was expected since the target class has very distinctive spectral and textural characteristics. Although all three approaches showed reasonably high accuracies due to the application of the NDPI and optimized procedures, the result showed the outperformance of the proposed IPOB approach. The framework established in this study expects to apply for regional or national PN surveys extensively. The information on the area and spatial distribution of PN can guide the government on policy making for the planting and exporting of traditional Chinese medicine resources.

Information about urban land use is important for urban planning and sustainable development. The emergence of geospatial big data (GBD), increased the availability of remotely sensed (RS) data and the development of new methods for data integration to provide new opportunities for mapping types of urban land use. However, the modes of RS and GBD integration are diverse due to the differences in data, study areas, classifiers, etc. In this context, this study aims to summarize the main methods of data integration and evaluate them via a case study of urban land use mapping in Hangzhou, China. We first categorized the RS and GBD integration methods into decision-level integration (DI) and feature-level integration (FI) and analyzed their main differences by reviewing the existing literature. The two methods were then applied for mapping urban land use types in Hangzhou city, based on urban parcels derived from the OpenStreetMap (OSM) road network, 10 m Sentinel-2A images, and points of interest (POI). The corresponding classification results were validated quantitatively and qualitatively using the same testing dataset. Finally, we illustrated the advantages and disadvantages of both approaches via bibliographic evidence and quantitative analysis. The results showed that: (1) The visual comparison indicates a generally better performance of DI-based classification than FI-based classification; (2) DI-based urban land use mapping is easy to implement, while FI-based land use mapping enables the mixture of features; (3) DI-based and FI-based methods can be used together to improve urban land use mapping, as they have different performances when classifying different types of land use. This study provides an improved understanding of urban land use mapping in terms of the RS and GBD integration strategy.

China launched the Three-North Shelterbelt Forest Program (TNSFP) in 1978 in northern China to combat desertification and dust storms, but it is still controversial in ecologically fragile arid and semi-arid areas, which is partly due to the uncertainties of monitoring of the spatial-temporal changes of forest distribution. In this study, we aim to provide an overall retrospect of the forest changes (i.e., forest gain and forest loss) in northern China during 2007–2017, and to analyze the forest changes in different precipitation zones. We first generated annual forest maps at 30 m spatial resolution during 2007–2017 in northern China through integrating Landsat and PALSAR/PALSAR-2 data. We found the PALSAR/Landsat-based forest maps outperform other four existing products (i.e., JAXA F/NF, FROM-GLC, GlobeLand30, and NLCD-China) from either PALSAR or Landsat data, with a higher overall accuracy 96% ± 1%. The spatial-temporal analyses of forests showed a substantial forest expansion from 316,898 ± 34,537 km2 in 2007 to 384,568 ± 35,855 km2 in 2017 in the central and eastern areas. We found a higher forest loss rate (i.e., 35%) in the precipitation zones with the annual mean precipitation less than 400 mm (i.e., the arid and semi-arid areas) comparing to that (i.e., 25%) in the zones with more than 400 mm (i.e., the humid areas), which suggests the lower surviving rate in the drylands. This study provides satellite-based evidence for the forest changes in different precipitation zones, and suggests that the likely impacts of precipitation on afforestation effectiveness should be considered in future implementation of ecological restoration projects like TNSFP.

The authors would like to change the authors’ affiliation in the recent published paper [...]

Human land use intensity affects the surface energy balance by changing the biogeophysical parameters. This study used Moderate Resolution Imaging Spectroradiometer remote sensing data and surface energy balance algorithms to quantify changes in surface energy budgets corresponding to changes in land use in Beijing from 2000 to 2015. Land use was reclassified by considering land use intensity. The difference in the latent heat flux (LE) and net radiation (Rn) (LE−Rn) expressed the warming or cooling effect. The results showed that: (i) The increasing trend of net longwave radiation in Beijing offset the decreasing trend of net shortwave radiation. The Rn changed slightly, while the LE and LE−Rn showed a significant increase of 0.55 and 0.56 W/(m²∙year), respectively. The findings indicated that considering only radiative forcing, or even Rn, was not enough to measure the impacts of land use change on the energy budget. (ii) The order of Rn, LE, and LE−Rn values from high to low were natural and seminatural areas, cropland, mixed pixel areas, urban expansion areas, and old urban areas. Compared with natural and seminatural areas, the changing LE−Rn trend in the other four land use types decreased with the increase in human impact intensity, indicating that human activities weakened the positive change trend of LE−Rn and increased the warming effect. (iii) Although the temporal trend of LE increased in Beijing from 2000 to 2015, the effect of Rn on LE−Rn was greater than that of LE, especially in the four land use types affected by human activities. The results for surface temperature in various land use types confirmed this point. This study highlights the energy budget differences of various land use types affected by human activities. It makes an important contribution to understanding the urban heat island effect from a biogeophysical perspective.

The impact of human activities on vegetation has been the focus of much research, but the impact on radiation energy through surface albedo associated with vegetation greenness and length of the growth season is still not well documented. Based on the land cover data for the years 2000 and 2015, this study first divided the land cover change in Beijing from 2000 to 2015 into five types according to the impact of human activities and vegetation resilience, namely, old urban areas (OU), urban expansion areas (UE), cropland (CP), mixed pixel areas (MP, which means the land covers other than urban expansion which had changed from 2000 to 2015), and the residual vegetation cover areas (pure pixels (PP), dominated by natural and seminatural vegetation, such as grassland, forest, and wetland). Then, we calculated the direct radiative forcing from the albedo change from 2000 to 2015 and analyzed the effect of vegetation on the albedo under different land cover types based on multi-resource Moderate Resolution Imaging Spectroradiometer (MODIS) products of vegetation, albedo, and solar radiation. The results showed that the most typical changes in land cover were from urban expansion. By comparing the PP with the four human-affected land cover types (OU, UE, MP, and CP), we confirmed that the radiative forcing increment between 2001−2003 and 2013−2015 in PP (0.01 W/m2) was much smaller than that in the four human-affected land cover types (the mean increment was 0.92 W/m2). This study highlights that human activities affected vegetation growth. This, in turn, brought changes in the albedo, thereby enhancing radiative forcing in Beijing during 2000−2015.

Accurate acquisition of the spatiotemporal distribution of urban forests and fragmentation (e.g., interior and intact regions) is of great significance to contributing to the mitigation of climate change and the conservation of habitat biodiversity. However, the spatiotemporal pattern of urban forest cover changes related with the dynamics of interior and intact forests from the present to the future have rarely been characterized. We investigated fragmentation of urban forest cover using satellite observations and simulation models in the Nanjing Laoshan Region of Jiangbei New Area, Jiangsu, China, during 2002−2023. Object-oriented classification-based land cover maps were created to simulate land cover changes using the cellular automation-Markov chain (CA-Markov) model and the state transition simulation modeling. We then quantified the forest cover change by the morphological change detection algorithm and estimated the forest area density-based fragmentation patterns. Their relationships were built through the spatial analysis and statistical methods. Results showed that the overall accuracies of actual land cover maps were approximately 83.75−92.25% (2012−2017). The usefulness of a CA-Markov model for simulating land cover maps was demonstrated. The greatest proportion of forest with a low level of fragmentation was captured along with the decreasing percentage of fragmented area from 81.1% to 64.1% based on high spatial resolution data with the window size of 27 pixels × 27 pixels. The greatest increase in fragmentation (3% from 2016 to 2023) among the changes between intact and fragmented forest was reported. However, intact forest was modeled to have recovered in 2023 and restored to 2002 fragmentation levels. Moreover, we found 58.07 km2 and 0.35 km2 of interior and intact forests have been removed from forest area losses and added from forest area gains. The loss rate of forest interior and intact area exceeded the rate of total forest area loss. However, their approximate ratio (1) implying the loss of forest interior and intact area would have slight fragmentation effects on the remaining forests. This analysis illustrates the achievement of protecting and restoring forest interior; more importantly, excessive human activities in the surrounding area had been avoided. This study provides strategies for future forest conservation and management in large urban regions.

Accurate crop planting area information is of significance for understanding regional food security and agricultural development planning. While increasing numbers of medium resolution satellite imagery and improved classification algorithms have been used for crop mapping, limited efforts have been made in feature selection, despite its vital impacts on crop classification. Furthermore, different crop types have their unique spectral and phenology characteristics; however, the different features of individual crop types have not been well understood and considered in previous studies of crop mapping. Here, we examined an optimized strategy to integrate specific features of individual crop types for mapping an improved crop type layer in the Sanjiang Plain, a new food bowl in China, by using all Sentinel-2 time series images in 2018. First, an automatic spectro-temporal feature selection (ASTFS) method was used to obtain optimal features for individual crops (rice, corn, and soybean), including sorting all features by the global separability indices for each crop and removing redundant features by accuracy changes when adding new features. Second, the ASTFS-based optimized feature sets for individual crops were used to produce three crop probability maps with the Random Forest classifier. Third, the probability maps were then composited into the final crop layer by considering the probability of each crop at every pixel. The resultant crop layer showed an improved accuracy (overall accuracy = 93.94%, Kappa coefficient = 0.92) than the other classifications without such a feature optimizing process. Our results indicate the potential of the ASTFS method for improving regional crop mapping.

Accurate and updated finer resolution maps of rubber plantations and stand ages are needed to understand and assess the impacts of rubber plantations on regional ecosystem processes. This study presented a simple method for mapping rubber plantation areas and their stand ages by integration of PALSAR 50-m mosaic images and multi-temporal Landsat TM/ETM+ images. The L-band PALSAR 50-m mosaic images were used to map forests (including both natural forests and rubber trees) and non-forests. For those PALSAR-based forest pixels, we analyzed the multi-temporal Landsat TM/ETM+ images from 2000 to 2009. We first studied phenological signatures of deciduous rubber plantations (defoliation and foliation) and natural forests through analysis of surface reflectance, Normal Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Land Surface Water Index (LSWI) and generated a map of rubber plantations in 2009. We then analyzed phenological signatures of rubber plantations with different stand ages and generated a map, in 2009, of rubber plantation stand ages (≤5, 6–10, >10 years-old) based on multi-temporal Landsat images. The resultant maps clearly illustrated how rubber plantations have expanded into the mountains in the study area over the years. The results in this study demonstrate the potential of integrating microwave (e.g., PALSAR) and optical remote sensing in the characterization of rubber plantations and their expansion over time.

Oil palm plantations have expanded rapidly. Estimating either positive effects on the economy, or negative effects on the environment, requires accurate maps. In this paper, three classification algorithms (Support Vector Machine (SVM), Decision Tree and K-Means) were explored to map oil palm plantations in Cameroon, using PALSAR 50 m Orthorectified Mosaic images and differently sized training samples. SVM had the ideal performance with overall accuracy ranging from 86% to 92% and a Kappa coefficient from 0.76 to 0.85, depending upon the training sample size (ranging from 20 to 500 pixels per class). The advantage of SVM was more obvious when the training sample size was smaller. K-Means required the user’s intervention, and thus, the accuracy depended on the level of his/her expertise and experience. For large-scale mapping of oil palm plantations, the Decision Tree algorithm outperformed both SVM and K-Means in terms of speed and performance. In addition, the decision threshold values of Decision Tree for a large training sample size agrees with the results from previous studies, which implies the possible universality of the decision threshold. If it can be verified, the Decision Tree algorithm will be an easy and robust methodology for mapping oil palm plantations.

An unprecedented reforestation process happened in the Loess Plateau, China due to the ecological restoration project ‘Grain for Green Project’, which has affected regional carbon and water cycles as well as brought climate feedbacks. Accurately mapping the area and spatial distribution of emerged forests in the Loess Plateau over time is essential for forest management but a very challenging task. Here we investigated the changes of forests in the Loess Plateau after the forest reconstruction project. First, we used a pixel and rule-based algorithm to identify and map the annual forests from 2007 to 2017 in the Loess Plateau by integrating 30 m Landsat data and 25 m resolution PALSAR data in this study. Then, we carried out the accuracy assessment and comparison with several existing forest products. The overall accuracy (OA) and Kappa coefficient of the resultant map, were about 91% and 0.77 in 2010, higher than those of the other forest products (FROM-GLC, GlobeLand30, GLCF-VCF, JAXA, and OU-FDL) with OA ranging from 83.57% to 87.96% and Kappa coefficients from 0.52 to 0.68. Based on the annual forest maps, we found forest area in the Loess Plateau has increased by around 15,000 km2 from 2007 to 2017. This study clearly demonstrates the advantages of data fusion between PALSAR and Landsat images for monitoring forest cover dynamics in the Loess Plateau, and the resultant forest maps with lower uncertainty would contribute to the regional forest management.

Unprecedented human-induced land cover changes happened in China after the Reform and Opening-up in 1978, matching with the era of Landsat satellite series. However, it is still unknown whether Landsat data can effectively support retrospective analysis of land cover changes in China over the past four decades. Here, for the first time, we conduct a systematic investigation on the availability of Landsat data in China, targeting its application for retrospective and continuous monitoring of land cover changes. The latter is significant to assess impact of land cover changes, and consequences of past land policy and management interventions. The total and valid observations (excluding clouds, cloud shadows, and terrain shadows) from Landsat 5/7/8 from 1984 to 2017 were quantified at pixel scale, based on the cloud computing platform Google Earth Engine (GEE). The results show higher intensity of Landsat observation in the northern part of China as compared to the southern part. The study provides an overall picture of Landsat observations suitable for satellite-based annual land cover monitoring over the entire country. We uncover that two sub-regions of China (i.e., Northeast China-Inner Mongolia-Northwest China, and North China Plain) have sufficient valid observations for retrospective analysis of land cover over 30 years (1987−2017) at an annual interval; whereas the Middle-Lower Yangtze Plain (MLYP) and Xinjiang (XJ) have sufficient observations for annual analyses for the periods 1989−2017 and 2004−2017, respectively. Retrospective analysis of land cover is possible only at a two-year time interval in South China (SC) for the years 1988−2017, Xinjiang (XJ) for the period 1992−2003, and the Tibetan Plateau (TP) during 2004−2017. For the latter geographic regions, land cover dynamics can be analyzed only at a three-year interval prior to 2004. Our retrospective analysis suggest that Landsat-based analysis of land cover dynamics at an annual interval for the whole country is not feasible; instead, national monitoring at two- or three-year intervals could be achievable. This study provides a preliminary assessment of data availability, targeting future continuous land cover monitoring in China; and the code is released to the public to facilitate similar data inventory in other regions of the world.

As the biggest carbon flux of terrestrial ecosystems from photosynthesis, gross primary productivity (GPP) is an important indicator in understanding the carbon cycle and biogeochemical process of terrestrial ecosystems. Despite advances in remote sensing-based GPP modeling, spatial and temporal variations of GPP are still uncertain especially under extreme climate conditions such as droughts. As the only official products of global spatially explicit GPP, MOD17A2H (GPPMOD) has been widely used to assess the variations of carbon uptake of terrestrial ecosystems. However, systematic assessment of its performance has rarely been conducted especially for the grassland ecosystems where inter-annual variability is high. Based on a collection of GPP datasets (GPPEC) from a global network of eddy covariance towers (FluxNet), we compared GPPMOD and GPPEC at all FluxNet grassland sites with more than five years of observations. We evaluated the performance and robustness of GPPMOD in different grassland biomes (tropical, temperate, and alpine) by using a bootstrapping method for calculating 95% confident intervals (CI) for the linear regression slope, coefficients of determination (R2), and root mean square errors (RMSE). We found that GPPMOD generally underestimated GPP by about 34% across all biomes despite a significant relationship (R2 = 0.66 (CI, 0.63⁻0.69), RMSE = 2.46 (2.33⁻2.58) g Cm−2 day−1) for the three grassland biomes. GPPMOD had varied performances with R2 values of 0.72 (0.68⁻0.75) (temperate), 0.64 (0.59⁻0.68) (alpine), and 0.40 (0.27⁻0.52) (tropical). Thus, GPPMOD performed better in low GPP situations (e.g., temperate grassland type), which further indicated that GPPMOD underestimated GPP. The underestimation of GPP could be partly attributed to the biased maximum light use efficiency (εmax) values of different grassland biomes. The uncertainty of the fraction of absorbed photosynthetically active radiation (FPAR) and the water scalar based on the vapor pressure deficit (VPD) could have other reasons for the underestimation. Therefore, more accurate estimates of GPP for different grassland biomes should consider improvements in εmax, FPAR, and the VPD scalar. Our results suggest that the community should be cautious when using MODIS GPP products to examine spatial and temporal variations of carbon fluxes.

Knowing the stand age of rubber tree (Hevea brasiliensis) plantations is vitally important for best management practices, estimations of rubber latex yields, and carbon cycle studies (e.g., biomass, carbon pools, and fluxes). However, the stand age (as estimated from the establishment year of rubber plantation) is not available across large regions. In this study, we analyzed Landsat time series images from 1987–2015 and developed algorithms to identify (1) the establishment year of rubber plantations; and (2) the pre-conversion land cover types, such as old rubber plantations, evergreen forests, and cropland. Exposed soil during plantation establishment and linear increases in canopy closure during non-production periods (rubber seedling to mature plantation) were used to identify the establishment year of rubber plantations. Based on the rubber plantation map for 2015 (overall accuracy = 97%), and 1981 Landsat images since 1987, we mapped the establishment year of rubber plantations on Hainan Island (R2 = 0.85/0.99, and RMSE = 2.34/0.54 years at pixel/plantation scale). The results show that: (1) significant conversion of croplands and old rubber plantations to new rubber plantations has occurred substantially in the northwest and northern regions of Hainan Island since 2000, while old rubber plantations were mainly distributed in the southeastern inland strip; (2) the pattern of rubber plantation expansion since 1987 consisted of fragmented plantations from smallholders, and there was no tendency to expand towards a higher altitude and steep slope regions; (3) the largest land source for new rubber plantations since 1988 was old rubber plantations (1.26 × 105 ha), followed by cropland (0.95 × 105 ha), and evergreen forests (0.68 × 105 ha). The resultant algorithms and maps of establishment year and pre-conversion land cover types are likely to be useful in plantation management, and ecological assessments of rubber plantation expansion in China.

As the largest among terrestrial ecosystems, forests are vital to maintaining ecosystem services and regulating regional climate. The area and spatial distribution of trees in densely forested areas have been focused on in the past few decades, while sparse forests in agricultural zones, so-called agroforests or trees outside forests (TOF), have usually been ignored or missed in existing forest mapping efforts, despite their important role in regulating agricultural ecosystems. We combined Landsat and PALSAR data to map forests in a typical agricultural zone in the North China Plain. The resultant map, based on PALSAR and Landsat (PL) data, was also compared with five existing medium resolution (30–100 m) forest maps from PALSAR (JAXA forest map) and Landsat: NLCD-China, GlobeLand30, ChinaCover, and FROM-GLC. The results show that the PL-based forest map has the highest accuracy (overall accuracy of 95 ± 1% with a 95% confidence interval, and Kappa coefficient of 0.86) compared to those forest maps based on single Landsat or PALSAR data in the North China Plain (overall accuracy ranging from 85 ± 2% to 92 ± 1%). All forest maps revealed higher accuracy in densely forested mountainous areas, while the PL-based and JAXA forest maps showed higher accuracy in the plain, as the higher omission errors existed in only the Landsat-based forest maps. Moreover, we found that the PL-based forest map can capture more patched forest information in low forest density areas. This means that the radar data have advantages in capturing forests in the typical agricultural zones, which tend to be missing in published Landsat-based only forest maps. Given the significance of agroforests in regulating ecosystem services of the agricultural ecosystem and improving carbon stock estimation, this study implies that the integration of PALSAR and Landsat data can provide promising agroforest estimates in future forest inventory efforts, targeting a comprehensive understanding of ecosystem services of agroforests and a more accurate carbon budget inventory.

The increasing expansion of rubber plantations throughout East and Southeast Asia urgently requires improved methods for effective mapping and monitoring. The phenological information from rubber plantations was found effective in rubber mapping. Previous studies have mostly applied rule-pixel-based phenology approaches for rubber plantations mapping, which might result in broken patches in fragmented landscapes. This study introduces a new paradigm by combining phenology information with object-based classification to map fragmented patches of rubber plantations in Xishuangbanna. This research first delineated the time windows of the defoliation and foliation phases of rubber plantations by acquiring the temporal profile and phenological features of rubber plantations and natural forests through the Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) data. To investigate the ability of finer resolution images at capturing the temporal profile or phenological information, 30 m resolution Landsat image data were used to capture the temporal profile, and a phenology algorithm to separate rubber plantations and natural forests was then defined. The derived phenology algorithm was used by both the object-based and pixel-based classification to investigate whether the object-based approach could improve the mapping accuracy. Whether adding the phenology information to the object-based classification could improve rubber plantation mapping accuracy in mountainous Xishuangbanna was also investigated. This resulted in three approaches: rule-pixel-based phenology, rule-object-based phenology, and nearest-neighbor-object-based phenology. The results showed that the rule-object-based phenology approaches (with overall accuracy 77.5% and Kappa Coefficients of 0.66) and nearest-neighbor-object-based phenology approach (91.0% and 0.86) achieved a higher accuracy than that of the rule-pixel-based phenology approach (72.7% and 0.59). The results proved that (1) object-based approaches could improve the accuracy of rubber plantation mapping compared to the pixel-based approach and (2) incorporating the phenological information from vegetation improved the overall accuracy of the thematic map.

Forests play an important role in maintaining ecosystem services, especially in ecologically fragile areas such as the Loess Plateau (LP) in China. However, there is still great uncertainty in the spatial extent and distribution of forests in such a fragmented region. In order to examine the advantages and disadvantages of existing forest mapping products, we conducted a thorough accuracy assessment on the eight recent, medium resolution (30–50 m) products by using the LP in 2010 as the region of interest. These mapping products include Landsat and/or PALSAR images (including the forest products from GlobeLand30), FROM-GLC, Hansen, ChinaCover, NLCD-China, GLCF VCF, OU-FDL, and JAXA. The same validation data were used to assess and rank the accuracy of each product. Additionally, the spatial consistency of the different forest products and their dependence on the terrain were analyzed. The results showed that the overall accuracies of the eight forest products on the LP in 2010 were between 0.93 ± 0.003 and 0.97 ± 0.002 with a 95% confidence interval, and GlobeLand30 presented the highest overall accuracy (0.97 ± 0.002). Among them, the PALSAR-based products (OU-FDL and JAXA) indicated relatively high accuracies, while the six Landsat-based products showed a large diversity in the accuracy. According to the eight products, the total estimated forest area of the LP varied from 7.627 ± 0.077 to 10.196 ± 0.1 million ha with a 95% confidence interval. We also found that the consistency in the spatial distribution of forests between these maps: 1) increased substantially with increasing elevation until 2000m, but then decreased at higher elevations, and 2) showed mild variation along increasing slope, but had a slight rate of increase. Our findings implied that future forest mapping studies should consider topographical attributes such as elevation and slope in their final products. Our results are fundamental in guiding future applications of these existing forest maps.

Surface urban heat island (SUHI) in the context of urbanization has gained much attention in recent decades; however, the seasonal variations of SUHI and their drivers are still not well documented. In this study, the Beijing-Tianjin-Hebei (BTH) urban agglomeration, one of the most typical areas experiencing drastic urbanization in China, was selected to study the SUHI intensity (SUHII) based on remotely sensed land surface temperature (LST) data. Pure and unchanged urban and rural pixels from 2000 to 2010 were chosen to avoid non-concurrency between land cover data and LST data and to estimate daytime and nighttime thermal effects of urbanization. Different patterns of the seasonal variations were found in daytime and nighttime SUHIIs. Specifically, the daytime SUHII in summer (4 °C) was more evident than in other seasons while a cold island phenomenon was found in winter; the nighttime SUHII was always positive and higher than the daytime one in all the seasons except summer. Moreover, we found the highest daytime SUHII in August, which is the growing peak stage of summer maize, while nighttime SUHII showed a trough in the same month. Seasonal variations of daytime SUHII showed higher significant correlations with the seasonal variations of ∆LAI (leaf area index) (R2 = 0.81, r = −0.90) compared with ∆albedo (R2 = 0.61, r = −0.78) and background daytime LST (R2 = 0.69, r = 0.83); moreover, agricultural practices (double-cropping system) played an important role in the seasonal variations of daytime SUHII. Seasonal variations of the nighttime SUHII did not show significant correlations with either of seasonal variations of ∆LAI, ∆albedo, and background nighttime LST, which implies different mechanisms in nighttime SUHII variation needing future studies.

Accurately mapping the spatial distribution of forests in sub-humid to semi-arid regions over time is important for forest management but a challenging task. Relatively large uncertainties still exist in the spatial distribution of forests and forest changes in the sub-humid and semi-arid regions. Numerous publications have used either optical or synthetic aperture radar (SAR) remote sensing imagery, but the resultant forest cover maps often have large errors. In this study, we propose a pixel- and rule-based algorithm to identify and map annual forests from 2007 to 2010 in Oklahoma, USA, a transitional region with various climates and landscapes, using the integration of the L-band Advanced Land Observation Satellite (ALOS) PALSAR Fine Beam Dual Polarization (FBD) mosaic dataset and Landsat images. The overall accuracy and Kappa coefficient of the PALSAR/Landsat forest map were about 88.2% and 0.75 in 2010, with the user and producer accuracy about 93.4% and 75.7%, based on the 3270 random ground plots collected in 2012 and 2013. Compared with the forest products from Japan Aerospace Exploration Agency (JAXA), National Land Cover Database (NLCD), Oklahoma Ecological Systems Map (OKESM) and Oklahoma Forest Resource Assessment (OKFRA), the PALSAR/Landsat forest map showed great improvement. The area of the PALSAR/Landsat forest was about 40,149 km2 in 2010, which was close to the area from OKFRA (40,468 km2), but much larger than those from JAXA (32,403 km2) and NLCD (37,628 km2). We analyzed annual forest cover dynamics, and the results show extensive forest cover loss (2761 km2, 6.9% of the total forest area in 2010) and gain (3630 km2, 9.0%) in southeast and central Oklahoma, and the total area of forests increased by 684 km2 from 2007 to 2010. This study clearly demonstrates the potential of data fusion between PALSAR and Landsat images for mapping annual forest cover dynamics in sub-humid to semi-arid regions, and the resultant forest maps would be helpful to forest management.