Next Article in Journal
The Comparison of Different Methods of Texture Analysis for Their Efficacy for Land Use Classification in Satellite Imagery
Next Article in Special Issue
Harmonizing Multi-Source Remote Sensing Images for Summer Corn Growth Monitoring
Previous Article in Journal
Characterization of Electromagnetic Properties of In Situ Soils for the Design of Landmine Detection Sensors: Application in Donbass, Ukraine
Previous Article in Special Issue
An Improved CASA Model for Estimating Winter Wheat Yield from Remote Sensing Images
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 11
Issue 10
10.3390/rs11101234
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Long-Term Monitoring of Cropland Change near Dongting Lake, China, Using the LandTrendr Algorithm with Landsat Imagery

by
Lihong Zhu
,
Xiangnan Liu
*,
Ling Wu
,
Yibo Tang
and
Yuanyuan Meng
School of Information Engineering, China University of Geosciences, Beijing 100083, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2019, 11(10), 1234; https://doi.org/10.3390/rs11101234
Submission received: 13 April 2019 / Revised: 6 May 2019 / Accepted: 22 May 2019 / Published: 24 May 2019
(This article belongs to the Special Issue Quantitative Remote Sensing for Agricultural Monitoring in the Big Data Era)
Browse Figures
Versions Notes

Abstract

:
Tracking cropland change and its spatiotemporal characteristics can provide a scientific basis for assessments of ecological restoration in reclamation areas. In 1998, an ecological restoration project (Converting Farmland to Lake) was launched in Dongting Lake, China, in which original lake areas reclaimed for cropland were converted back to lake or to poplar cultivation areas. This study characterized the resulting long-term (1998–2018) change patterns using the LandTrendr algorithm with Landsat time-series data derived from the Google Earth Engine (GEE). Of the total cropland affected, ~447.48 km2 was converted to lake and 499.9 km2 was converted to poplar cultivation, with overall accuracies of 87.0% and 83.8%, respectively. The former covered a wider range, mainly distributed in the area surrounding Datong Lake, while the latter was more clustered in North and West Dongting Lake. Our methods based on GEE captured cropland change information efficiently, providing data (raster maps, yearly data, and change attributes) that can assist researchers and managers in gaining a better understanding of environmental influences related to the ongoing conversion efforts in this region.

1. Introduction

Dongting Lake, once the largest freshwater lake in China, has shrunk to the second-largest due to land reclamation for agriculture; this has resulted in serious ecological degradation of the area’s wetlands [1]. After the devastating flood in 1998, the government promoted the project of Converting Farmland to Lake (CFTL) and planned to abandon 785.7 km2 of cropland area, where conversion to lake or poplar trees cultivation could establish a new balance between economic benefits and the regulation of wetland ecological functions. These efforts have created a need for long-term monitoring of cropland change processes to provide scientific data for ongoing ecological management after project completion.
Although field surveys are the most accurate monitoring method for this purpose, they are impractical over large areas due to time constraints and the inability to assess certain inaccessible sites (such as shoal areas). In contrast, remote sensing techniques enable frequent imaging, easier access, and larger-scale monitoring, becoming important global data sources for detecting environmental changes (including those to cropland) [2,3,4]. The moderate resolution of publicly available USGS Landsat data [5] makes them suitable for monitoring changes in agriculture fields [6,7,8]. Detection of cropland change can be based on comparisons between two or more Landsat images [9,10,11], and to date, many studies have applied Landsat time-series analysis to long-term monitoring of cropland change [12,13]. Such remote sensing time series can be used to separate long-term and short-term land-use changes more reliably in highly resilient land systems. Zhe [14] provided a comprehensive review of change detection approaches using Landsat time-series data, including the description of frequencies, preprocessing, algorithms, and applications.
Choosing a suitable algorithm is a critical step for conducting change detection [15]. Several change-detection techniques based on Landsat time series have been developed to be robust against spectral variations arising from topography and phenology [16]. For example, Landsat-based Detection of Trends in Disturbance and Recovery (LandTrendr) and Breaks For Additive Season and Trend (BFAST) have been applied to monitor cropland change [17,18]. BFAST has been used to investigate cropping systems and temporal paddy crop dynamics in Sidoarjo Regency, Indonesia, and provide accurate and up-to-date information on agricultural land-use changes [19,20]. LandTrendr can be used to identify cropland abandonment and re-cultivation based on annual time series of cropland probabilities; it can also contribute to the identification of different smallholder cultivation patterns [21,22].
The above change-detection algorithms, and others such as Continuous Change Detection and Classification (CCDC) [23] are subject to considerable computational time in data-preprocessing and are themselves time-consuming. In order to reach to a broader research community, the LandTrendr algorithm (https://emapr.github.io/LT-GEE/) was implemented on the Google Earth Engine (GEE) platform, which provides full access to the Landsat archive and parallel processing to increase computational speed [24]. Compared to previous change detection algorithms run on local servers [25,26], LandTrendr in GEE detects cropland change more efficiently with the advantages of cloud computing and offers an opportunity to map large-scale cropland change [27].
Previous change assessment studies have focused on cropland abandonment and re-cultivation and detecting changes in their cultivation patterns. Thus, cropland conversion to other land cover classes in areas such as Dongting Lake in China has not received much attention. Using LandTrendr with GEE support, we can observe the process of cropland conversion to lake or poplar trees and its spatial and temporal characteristics efficiently, understanding this process can provide the policy enlightenment for ecological conservation in Dongting Lake.
Our overarching goal therefore was to use LandTrendr with Landsat time-series in GEE to detect cropland change processes in the Dongting Lake area, we analyzed a conceptual model of the spectral feature trajectories related to cropland conversion to lake or poplar cultivation. We also parameterized the LandTrendr algorithm and tested it based on Landsat time-series data to track these changes, then compared their spatial and temporal characteristics.

2. Materials and Methods

2.1. Study Area

Dongting Lake (28°42′–29°38′ N, 111°52′–113°08′ E), located in the north part of Hunan Province, is the second largest freshwater lake in China with an area of 2794.7 km2 [28]. We defined our study area by administrative divisions in the lake’s surrounding area, including 20 counties (districts) in Yueyang, Yiyang, and Changde (Figure 1). This area included reclamation zones outside the Dongting Lake embankment with internal lakes used mainly for agricultural production; while the area within the embankment included the outer lake (East Dongting Lake, South Dongting Lake, Hengling Lake, and West Dongting Lake).
There has been excessive reclamation of the lake for almost 100 years, resulting in serious ecological degradation of the wetland ecosystem associated with the lake. Following the great Yangtze River flood in 1998, the CFTL was launched to help restore the lake’s ecology and guide the area’s development in a more benign direction. After the project was initiated, agricultural land focused on two change patterns: conversions back to lake were implemented mainly around in the outer lake area to improve ecological functions, while poplar trees cultivation was implemented primarily in part of the outer lake for economic benefits.
Two typical CFTL areas (Figure 1), the Qingshan Polder (28°51′ N, 112°12′ E) with an area of 11.1 km2 and the Jicheng polder (29°41′ N, 112°56′ E) with an area of 33.7 km2, are taken as examples (Figure 2). The local government led its cropland conversion to natural lake after 1998 with Qingshan Polder, aiming to restore a wetland ecosystem. As a typical demonstration area for the CFTL project, Qingshan developed a green aquatic culture that did not require fertilizer or feed for recovery. At Jicheng Polder, many poplars were planted to form a plantation forest system, based on the pattern of poplar cultivation after local residents moved outside the village and abandoned the cropland. Poplar cultivation can alleviate the tight supply of medicinal materials and promote the development of a local economy [29].

2.2. Data Preparation

We used the GEE platform’s full archive to build stacks of Surface Reflectance Tier 1 Landsat TM/ETM+/OLI images from 1998–2018 for use with LandTrendr (Table 1). In order to minimize variations caused by phenology or changes in solar geometry, a total of 320 images across three path/row numbers were acquired in the crop-growing season (June 1 to September 1) within the whole study period. As the Landsat 8/OLI has higher 12-bit radiometric resolution than the previous Landsat 7/ETM+, we applied statistical harmonization functions between the spectral values of both sensors to normalize the reflectance [30]. Cloud, cloud shadow, and snow masks were produced from the Fmask band. Finally, 21 annual composites from 320 images with minimal cloud cover were created using the median reflectance values of the collection [31].

2.3. Cropland Change Conceptual Model

Figure 3 presents the conceptual model of two cropland change patterns, from which the pattern of conversion to either lake cover over poplar cultivation can be seen; this results in a mixed pattern of changes. To focus on the pattern of conversions to lakes, we removed poplar cultivations by subtracting the superposition results of the two patterns using ArcGIS 10.4. Normalized Difference Vegetation Index (NDVI) is a useful indicator for monitoring vegetation changes in agricultural change areas [32,33], so we used this to detect cropland changes in this study. We captured as many examples of cropland conversion as possible to know how NDVI signatures would represent the cropland conversions. When cropland is converted to lake, the initially higher NDVI should drop to a much lower level (Figure 3a). When cropland is converted to poplar cultivation, NDVI should drop temporarily but recover to a higher level than before (Figure 3b). These proposed trajectories reflect occurrence, duration, and magnitude attributes, which respectively express the year of occurrence, duration time, and the range of NDVI variation during conversion.

2.4. Initial Land Cover Classification

Prior to running the LandTrendr algorithm, we applied the Random Forest (RF) classifier [34] to pre-CFTL Landsat images in GEE to extract cropland areas (Figure 4a). First, Landsat images from 1997 were used with the cloud and shadow masks to produce a cloud-free composite, from which the NDVI was calculated. Next, 250 training samples for each land-cover class were selected for interpretation by high-resolution imagery from Google Earth. As NDVI can be used to distinguish impervious surfaces, bare soil, and water bodies from forest or croplands, this along with bands one through five, and seven were used as feature inputs into the RF classifier with 20 trees. The resulting imagery was classified into five land-cover categories: water, cropland, forest, urban, and shoal. Finally, elevation and slope derived from GEE DEM data were used to modify the classification results for higher accuracy. In 1997, the major land cover types were cropland and water (Figure 1c).

2.5. Monitoring Cropland Change with LandTrendr

We used the LandTrendr algorithm developed by Kennedy [35] to map and characterize post-CFTL cropland changes (Figure 4b). The core of the LandTrendr algorithm is a temporal segmentation method used to capture both long-term gradual and short-term drastic changes; this approach can monitor cropland change by analyzing the temporal-spectral trajectory of each pixel. The input for each pixel is the annual time series of one spectral band or index, plus the date. The processing procedure for finding the best model involves removing noise-induced spikes (outliers), identifying potential vertices (breakpoints), fitting trajectories, and setting the optimal number of segments [23].
LandTrendr requires the setting of control parameters to ensure the quality of change detection, so we analyzed examples of cropland conversion event, tested different combinations of parameter values to determine the optimal combination of parameters. We also assessed the characteristics of NDVI in the study area to exclude other change patterns in order to make LandTrendr more precise (Figure 5). NDVI before conversion ranged from 0.55–0.73, dropped sharply after conversion to lake (–0.25 to –0.16), and rose slightly after conversion to poplar cultivation (0.57–0.77). Therefore, for the event of conversion to lake, we set the pre-lake-conversion NDVI to be >0.55 and the magnitude of NDVI decrease should be >0.71. While for the event of poplar cultivation, the pre-poplar-cultivation NDVI > –0.25 and the magnitude of NDVI increase should be >0.73.

2.6. Accuracy Assessment and Validation

To assess the accuracy of the 1997 classification results, we selected validation sites different from those used for training the classification algorithm, randomly generating 180 samples for each land-cover class. We then produced a confusion matrix and estimates of overall accuracy, user accuracy, and producer accuracy for each class following the methods defined by Foody [36] and Congalton and Green [37].
To assess the accuracy of the cropland conversions, during the study period we selected 300 pixels representing conversion and another 300 pixels representing unchanged areas for each pattern in which the unchanged pixels were selected for the probability that they were breaking points. For each pixel, we visually inspected all 21 composite images and used high-spatial-resolution Google Earth imagery for further manual interpretation of land cover. We thus determined and recorded whether and when (occurrence year) cropland conversion occurred at these pixels; if multiple changes occurred within a single pixel, only the greatest change was used in the accuracy assessment.

3. Results

3.1. Historical Change Process of Conversion Patterns

The disturbance map (Figure 6) was rendered as a gradient according to the occurrence years to present the spatial and temporal distribution of conversion to lake during 1998–2018. The conversion was widely dispersed but most prevalent in four regions: Datong Lake, West Dongting Lake, South Dongting Lake, and East Dongting Lake (Figure 6). Overall, 447.48 km2 of cropland was converted to lake from 1998–2018 (Table 2). Annual change peaked in 1998 due to the CFTL project’s initiation, after which conversion generally declined until 2012, when it began to increase again due to new policies aimed at improving the protection and restoration of Dongting Lake.
Figure 7 was also rendered as a gradient according to the occurrence years to present the spatial and temporal distribution of poplar cultivation. Conversion to poplar cultivation was particularly prominent in West and North Dongting Lake (P1, and P2, P3, P4 in Figure 7). The patterns of poplar cultivation events in inner lakes around Datong Lake did not follow the disturbance trends of the outer West and North Dongting Lake. The total area of poplar conversion was ~12% larger than that of lake conversion over the past 21 years (Table 2). Annual conversion spiked in two periods, 2004–2005 with disturbance rates which amounted to 4.44% and 4.61%, and 2013–2014 with disturbances rates which amounted to 7.80% and 6.92%.

3.2. Characterization of Two Conversion Patterns

NDVI spectral–temporal characteristics discriminated between two cropland change patterns (Figure 8). Of the pixels in which cropland was converted to lake, 82.4% showed a decline in NDVI value between 0.70–0.90. The area near Datong Lake experienced a similar decline in NDVI between 0.77–0.88 (Figure 8a), with even higher NDVI declines (0.99–1.10) during flooding events in East Dongting Lake. The average duration of conversion to lake was 2.71 years; nearly 89.9% of relevant pixels ranged from 1–5 years while 52.1% were very short (two years or less) (Figure 8b). Moreover, East Dongting Lake was dominated by one-year durations [27]. For every change pixel of conversion to poplar cultivation, we analyzed magnitude and duration characteristics of the poplar growth process for the second half of each LandTrendr trajectory. 75.6% of the relevant pixels showed an increase in NDVI value between 0.73–1.02. Higher NDVI increases clustered at the centralized poplar cultivation areas in North and West Dongting Lake (Figure 8c). The average duration of conversion to poplar cultivation was 3.17 years, while the most common durations were 1 (23.5%) and 21 (17.1%) years. Generally, longer change durations occurred in North Dongting Lake (Figure 8d) where the humid climate was suitable for poplar growth [27].

3.3. Accuracy Assessment

3.3.1. Accuracy Assessment for Initial Classification

The confusion matrix for the initially classified 1997 imagery (Table 3) indicated that the overall accuracy, user accuracy, and producer accuracy for each land-cover class were mostly >80%. Three categories (cropland, forest, and shoal) were most frequently misclassified. Cropland and forests were difficult to distinguish between at the 30 m Landsat resolution, because forests are often small and scattered in the study area. In addition, forests and shoals were often misclassified likely because other vegetation growing in shoals resemble forest in their spectral characteristics. Although the lower producer accuracies are indicative of errors related to the classification algorithm, the overall accuracy was greater than 85% for each classification period, indicating the RF classifier was sufficient for this study area.

3.3.2. Accuracy Assessment for LandTrendr

The results of the segmentation and fitting algorithms suggest that both conversion patterns were successfully captured by LandTrendr. Both accuracy assessments (Table 4) indicated high producer and user accuracies, with the overall accuracy of conversion to lake (87.0%) being slightly higher than conversion to poplar cultivation (83.8%). One typical area of each conversion type (Qingshan polder and Jicheng polder described in Figure 2) was further studied to validate the algorithm, RF classification (Figure 9a2,b2) in specific years finished to help us better analyze the relevant changes. In the first area (Figure 9a), original cropland was abandoned after the CFTL and converted to lake from 1998 onward, a change conducive to wetland ecological recovery [28]. In the second area (Figure 9b), cropland was abandoned before poplar planting prior to 2002 but these plantings were decreasing by 2010.

4. Discussion

4.1. Mapping Approach

LandTrendr in this study detected different spatial distribution characteristics accurately under both conversion patterns. Conversion to lake detected most prevalent around Datong Lake, because of the Government encouraged the improvement of water storage capacity of the inner lake by excavating deep ponds near Datong Lake (L1 in Figure 6), while the East, South, and West Dongting Lake areas (L2, L3, and L4 in Figure 6 respectively) have also undergone large-scale improvements to improve their interconnectivity and improve the ecological function of Dongting Lake as a whole [38]. Conversion to poplar cultivation was particularly prominent in West and North Dongting Lake due to raw material supply bases set up in this regions to develop the local economy, resulting in large-scale poplar planting in this district.
Using LandTrendr resulted in producer and user accuracies high enough to detecting different spatial and temporal characteristics under both conversion patterns. Compare the user’s accuracy between changed and stable pixels, the relatively low accuracy for changed pixels indicated that commission errors did occur; we identified four types (Figure 10). First, errors can be caused by transient change events like floods (such as the major flooding in 1998), which can lead to incorrect classification as conversion to lake (Figure 10a). Second, non-abrupt conversion to lake, which occurs when the NDVI signals are unstable, leads to advance or lag segments in the fitted trajectories (Figure 10b). Third, reed plantings have the same NDVI fluctuation range as poplars and the two can be difficult to distinguish (Figure 10c). Fourth, non-abrupt poplar cultivation events produce gradual changes while the NDVI before or after the conversion have strong fluctuations (Figure 10d). In these cases, the fitted trajectory does not reach the set p-value parameter (0.05), making the segmentation of the LandTrendr return a straight line as a segment for whole trajectory. Despite these potential errors, the high producer and user accuracies for both conversion patterns imply that the LandTrendr approach used in this research was robust.

4.2. Benefits of Change Detection Using LandTrendr in GEE

Accurate information about the timing and extent of cropland change is crucial for environmental assessment and policy making [39]. In contrast to analyzed cropland abandonment and re-cultivation via LandTrendr [40], we used this approach to study cropland conversion to other land cover classes. By using this temporal segmentation algorithm with a time series of Landsat data, we were able to map the spatial and temporal characteristics of different cropland conversion type. One clear advantage of Landsat is the availability of consistent data for over three decades, allowing long-term analyses. Compared to multi-date change mapping, the Landsat time series used here allowed us to map more deviations in detail [41].
Although change detection-related studies have seen a considerable amount of research using LandTrendr using data on local servers [25,35,42], the algorithm is still limited by time-consuming data management work. In contrast, there are a few advantages of LandTrendr with GEE support to monitor cropland change. Firstly, images are mostly prepared and updated by Google for cropland change analysis. Secondly, GEE had an increased speed and great advantages in terms of vast data handling and management cost [43,44], after data processing, the actual LandTrendr computational cost and time are also reduced. Moreover, implementation in GEE makes it available to a much broader base of users for those lack the technical capacity, experience, or financial means.

4.3. Research Limitations

Vegetation indices have been regarded as better variables than individual spectral bands, in terms of monitoring land cover change, because they can reduce the impacts of external factors such as topography and atmosphere on the surface reflectance. Selecting suitable vegetation indices is critical for successfully detecting cropland conversion in our study; different vegetation indices, such as Enhanced Vegetation Index (EVI) [20] and NDVI [32,45], have been used for this purpose. Research shows that NDVI distinguishes cropland change better than other indices [46,47], but there are still some limitations to applying the LandTrendr algorithm with NDVI. One example is the disturbance caused by other vegetation, such as the reeds in our study region that produced an NDVI signal similar to poplar and thus influenced the detection result. It is unclear whether or not NDVI is the best index to use for the Dongting Lake region; more research is needed to identify an optimum.

5. Conclusions

In this study, we showed the effectiveness of a trajectory-based change detection approach to characterize two cropland change patterns in China’s Dongting Lake region. By using the LandTrendr algorithm with Landsat imagery derived from the Google Earth Engine, we detected and mapped the conversion of cropland to lake and poplar cultivation with overall accuracies of 87.0% and 83.8%, respectively. Conversion to lake covered a wider range, mainly distributed in areas surrounding the outer lake and Datong Lake, while conversion to poplar cultivation was more prevalent in North and West Dongting Lake. Over the entire study period, almost 947.38 km2 of cropland was converted, with poplar accounting for 52.42 km2 more than lake. Some differences were apparent between the two conversion patterns when their spectral–temporal characteristics were compared. For lake conversion pixels, 82.4% had a decline in NDVI value between 0.70–0.90, while 75.6% of poplar-conversion pixels increased by 0.73–1.02. We also found a high proportion of short-duration (two years or less) pixels for conversion to lake, while the average duration of poplar cultivation pixels was a bit longer. These results can assist researchers and managers in better understanding the conversion processes over time in this region while providing baseline information for assessing the environmental influences of cropland conversion in the Dongting Lake region.

Author Contributions

Conceptualization, X.L. and L.Z.; Data curation, L.Z., L.W. and Y.T.; Formal analysis, L.Z., L.W. and Y.T.; Methodology, L.Z. and Y.T.; Supervision, X.L.; Validation, and L.W.; Visualization, Y.M.; Writing—Original Draft Preparation, L.Z.; Writing—Review & Editing, X.L., L.W. and Y.M.

Funding

This research was funded by the National Natural Science Foundation of China under Grant 41871223 and the Fundamental Research Funds for the Central Universities under Grant 2652017116.

Acknowledgments

The authors would like to thank the anonymous reviewers and the editor for their constructive comments and suggestions for this paper.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Xu, C.; Mcgowan, S.; Lei, X.; Zeng, L.; Yang, X. Effects of hydrological regulation and anthropogenic pollutants on Dongting Lake in the Yangtze floodplain. Ecohydrology 2016, 9, 315–325. [Google Scholar]
  2. Hereher, M.E. Environmental monitoring and change assessment of Toshka lakes in southern Egypt using remote sensing. Environ. Earth Sci. 2015, 73, 3623–3632. [Google Scholar] [CrossRef]
  3. Qi, S.H.; Brown, D.G.; Tian, Q.; Jiang, L.G.; Zhao, T.T.; Bergen, K.M. Inundation extent and flood frequency mapping using LANDSAT imagery and digital elevation models. Mapp. Sci. Remote Sens. 2009, 46, 101–127. [Google Scholar] [CrossRef]
  4. Kraemer, R.; Prishchepov, A.V.; Müller, D.; Kuemmerle, T.; Radeloff, V.C.; Dara, A.; Terekhov, A.; Frühauf, M. Long-term agricultural land-cover change and potential for cropland expansion in the former Virgin Lands area of Kazakhstan. Environ. Res. Lett. 2015, 10, 054012. [Google Scholar] [CrossRef]
  5. Woodcock, C.E.; Richard, A.; Martha, A.; Alan, B.; Robert, B.; Warren, C.; Feng, G.; Goward, S.N.; Dennis, H.; Eileen, H. Free access to Landsat imagery. Science 2008, 320, 1011. [Google Scholar] [CrossRef] [PubMed]
  6. Röder, A.; Stellmes, M.; Hill, J.; Kuemmerle, T.; Tsiourlis, G.M. Analysing land cover change using time series analysis of Landsat data and geoinformation processing. A natural experiment in Northern Greece. Proc. SPIE Int. Soc. Opt. Eng. 2008, 7104, 43–56. [Google Scholar]
  7. Rogan, J.; Franklin, J.; Roberts, D.A. A comparison of methods for monitoring multitemporal vegetation change using Thematic Mapper imagery. Remote Sens. Environ. 2002, 80, 143–156. [Google Scholar] [CrossRef]
  8. Wohlfart, C.; Mack, B.; Liu, G.; Kuenzer, C. Multi-faceted land cover and land use change analyses in the Yellow River Basin based on dense Landsat time series: Exemplary analysis in mining, agriculture, forest, and urban areas. Appl. Geogr. 2017, 85, 73–88. [Google Scholar] [CrossRef]
  9. Liao, C.; Feng, Z.; Peng, L.I.; Zhang, J. Monitoring the spatio-temporal dynamics of swidden agriculture and fallow vegetation recovery using Landsat imagery in northern Laos. Acta Geogr. Sin. 2015, 25, 1218–1234. [Google Scholar] [CrossRef] [Green Version]
  10. Forkuor, G.; Conrad, C.; Thiel, M.; Zoungrana, B.; Tondoh, J. Multiscale Remote Sensing to Map the Spatial Distribution and Extent of Cropland in the Sudanian Savanna of West Africa. Remote Sens. 2017, 9, 839. [Google Scholar] [CrossRef]
  11. Falkowski, M.J.; Manning, J.A. Parcel-based classification of agricultural crops via multitemporal Landsat imagery for monitoring habitat availability of western burrowing owls in the Imperial Valley agro-ecosystem. Can. J. Remote Sens. 2010, 36, 750–762. [Google Scholar] [CrossRef]
  12. Justice, C.J. Landsat-derived cropland mask for Tanzania using 2010–2013 time series and decision tree classifier methods. In Proceedings of the Agu Fall Meeting, College Park, MD, USA, 17 December 2015. [Google Scholar]
  13. Xu, Y.; Yu, L.; Zhao, F.R.; Cai, X.; Zhao, J.; Lu, H.; Gong, P. Tracking annual cropland changes from 1984 to 2016 using time-series Landsat images with a change-detection and post-classification approach: Experiments from three sites in Africa. Remote Sens. Environ. 2018, 218, 13–31. [Google Scholar] [CrossRef]
  14. Zhe, Z. Change detection using landsat time series: A review of frequencies, preprocessing, algorithms, and applications. Isprs J. Photogramm. Remote Sens. 2017, 130, 370–384. [Google Scholar] [CrossRef]
  15. Lu, D.; Li, G.; Moran, E. Current situation and needs of change detection techniques. Int. J. Image Data Fusion 2014, 5, 13–38. [Google Scholar] [CrossRef]
  16. Griffiths, P.; Kuemmerle, T.; Kennedy, R.E.; Abrudan, I.V.; Knorn, J.; Hostert, P. Using annual time-series of Landsat images to assess the effects of forest restitution in post-socialist Romania. Remote Sens. Environ. 2012, 118, 199–214. [Google Scholar] [CrossRef]
  17. Novo-Fernández, A.; Franks, S.; Wehenkel, C.; López-Serrano, P.M.; Molinier, M.; López-Sánchez, C.A. Landsat time series analysis for temperate forest cover change detection in the Sierra Madre Occidental, Durango, Mexico. Int. J. Appl. Earth Obs. Geoinf. 2018, 73, 230–244. [Google Scholar] [CrossRef]
  18. Verbesselt, J.; Herold, M.; Hyndman, R.; Zeileis, A.; Culvenor, D. A robust approach for phenological change detection within satellite image time series. In Proceedings of the Analysis of Multi-Temporal Remote Sensing Images, Trento, Italy, 12–14 July 2011. [Google Scholar]
  19. Fatikhunnada, A.; Seminar, K.B.; Solahudin, M.; Buono, A. Optimization of Parallel K-means for Java Paddy Mapping Using Time-series Satellite Imagery. Telkomnika 2018, 16, 1409–1415. [Google Scholar] [CrossRef]
  20. Huang, K.; Tao, Z.; Xiang, Z. Extreme Drought-induced Trend Changes in MODIS EVI Time Series in Yunnan, China. IOP Conf. Ser. Earth Environ. Sci. 2014, 17, 012070. [Google Scholar] [CrossRef] [Green Version]
  21. Schneibel, A.; Stellmes, M.; Röder, A.; Frantz, D.; Kowalski, B.; Haß, E.; Hill, J. Assessment of spatio-temporal changes of smallholder cultivation patterns in the Angolan Miombo belt using segmentation of Landsat time series. Remote Sens. Environ. 2017, 195, 118–129. [Google Scholar] [CrossRef] [Green Version]
  22. Dara, A.; Baumann, M.; Kuemmerle, T.; Pflugmacher, D.; Rabe, A.; Griffiths, P.; Hölzel, N.; Kamp, J.; Freitag, M.; Hostert, P. Mapping the timing of cropland abandonment and recultivation in northern Kazakhstan using annual Landsat time series. Remote Sens. Environ. 2018, 213, 49–60. [Google Scholar] [CrossRef]
  23. Zhe, Z.; Woodcock, C.E. Continuous change detection and classification of land cover using all available Landsat data. Remote Sens. Environ. 2014, 144, 152–171. [Google Scholar] [CrossRef] [Green Version]
  24. Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 2017, 202, 18–27. [Google Scholar] [CrossRef]
  25. Watts, L.M.; Laffan, S.W. Effectiveness of the BFAST algorithm for detecting vegetation response patterns in a semi-arid region. Remote Sens. Environ. 2014, 154, 234–245. [Google Scholar] [CrossRef]
  26. Li, Z.; Wu, W.; Liu, X.; Fath, B.D.; Sun, H.; Liu, X.; Xiao, X.; Cao, J. Land use/cover change and regional climate change in an arid grassland ecosystem of Inner Mongolia, China. Ecol. Model. 2017, 353, 86–94. [Google Scholar] [CrossRef]
  27. Shelestov, A.; Lavreniuk, M.; Kussul, N.; Novikov, A.; Skakun, S. Exploring Google Earth Engine Platform for Big Data Processing: Classification of Multi-Temporal Satellite Imagery for Crop Mapping. Front. Earth Sci. 2017, 5, 17. [Google Scholar] [CrossRef]
  28. Hu, Y.; Huang, J.; Yun, D.; Han, P.; Wei, H. Monitoring Spatial and Temporal Dynamics of Flood Regimes and Their Relation to Wetland Landscape Patterns in Dongting Lake from MODIS Time-Series Imagery. Remote Sens. 2015, 7, 7494–7520. [Google Scholar] [CrossRef] [Green Version]
  29. Li, Y.; Chen, X.; Xie, Y.; Li, X.; Li, F.; Hou, Z. Effects of young poplar plantations on understory plant diversity in the Dongting Lake wetlands, China. Sci. Rep. 2014, 4, 6339. [Google Scholar] [CrossRef]
  30. Roy, D.P.; Kovalskyy, V.; Zhang, H.K.; Vermote, E.F.; Yan, L.; Kumar, S.S.; Egorov, A. Characterization of Landsat-7 to Landsat-8 reflective wavelength and normalized difference vegetation index continuity. Remote Sens. Environ. 2016, 185, 57–70. [Google Scholar] [CrossRef] [Green Version]
  31. Simonetti, D.; Simonetti, E.; Szantoi, Z.; Lupi, A.; Eva, H.D. First Results From the Phenology-Based Synthesis Classifier Using Landsat 8 Imagery. IEEE Geosci. Remote Sens. Lett. 2015, 12, 1496–1500. [Google Scholar] [CrossRef]
  32. Zhang, Y.; Gao, J.; Liu, L.; Wang, Z.; Ding, M.; Yang, X. NDVI-based vegetation changes and their responses to climate change from 1982 to 2011: A case study in the Koshi River Basin in the middle Himalayas. Glob. Planet. Chang. 2013, 108, 139–148. [Google Scholar] [CrossRef]
  33. Bai, J.J.; Bai, J.T.; Wang, L. Spatio-temporal Change of Vegetation NDVI and Its Relations with Regional Climate in Northern Shaanxi Province in 2000–2010. Sci. Geogr. Sin. 2014, 34, 882–888. [Google Scholar]
  34. Mainknorn, M.; Cohen, W.B.; Kennedy, R.E.; Grodzki, W.; Pflugmacher, D.; Griffiths, P.; Hostert, P. Monitoring coniferous forest biomass change using a Landsat trajectory-based approach. Remote Sens. Environ. 2013, 139, 277–290. [Google Scholar] [CrossRef]
  35. Kennedy, R.E.; Yang, Z.; Cohen, W.B. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr—Temporal segmentation algorithms. Remote Sens. Environ. 2010, 114, 2897–2910. [Google Scholar] [CrossRef]
  36. Foody, G.M. Status of land cover classification accuracy assessment. Remote Sens. Environ. 2002, 80, 185–201. [Google Scholar] [CrossRef]
  37. Congalton, R.G.; Green, K. Practical look at the sources of confusion in error matrix generation. Photogramm. Eng. Remote Sens. 1993, 59, 641–644. [Google Scholar]
  38. Bo, Y.; Zeng, F.; Yuan, M.; Li, D.; Qiu, Y.; Li, J. Measurement of Dongting Lake Area Based on Visual Interpretation of Polders. Proced. Environ. Sci. 2011, 10, 2684–2689. [Google Scholar] [Green Version]
  39. Zhang, P.L.; Bo, L.R.; Chao, J. An Object-based Basic Farmland Change Detection Using High Spatial Resolution Image and GIS Data of Land Use Planning. Key Eng. Mater. 2012, 500, 492–499. [Google Scholar] [CrossRef]
  40. Yin, H.; Prishchepov, A.V.; Kuemmerle, T.; Bleyhl, B.; Buchner, J.; Radeloff, V.C. Mapping agricultural land abandonment from spatial and temporal segmentation of Landsat time series. Remote Sens. Environ. 2018, 210, 12–24. [Google Scholar] [CrossRef]
  41. Schmidt, M.; Pringle, M.; Devadas, R.; Denham, R.; Dan, T. A Framework for Large-Area Mapping of Past and Present Cropping Activity Using Seasonal Landsat Images and Time Series Metrics. Remote Sens. 2016, 8, 312. [Google Scholar] [CrossRef]
  42. Schwantes, A.M.; Swenson, J.J.; Jackson, R.B. Quantifying drought-induced tree mortality in the open canopy woodlands of central Texas. Remote Sens. Environ. 2016, 181, 54–64. [Google Scholar] [CrossRef] [Green Version]
  43. Long, T.; Zhang, Z.; He, G.; Jiao, W.; Tang, C.; Wu, B.; Zhang, X.; Wang, G.; Yin, R. 30 m resolution Global Annual Burned Area Mapping based on Landsat images and Google Earth Engine. Remote Sens. 2019, 11, 489. [Google Scholar] [CrossRef]
  44. Dong, J.; Xiao, X.; Menarguez, M.A.; Zhang, G.; Qin, Y.; Thau, D.; Biradar, C.; Iii, B.M. Mapping paddy rice planting area in northeastern Asia with Landsat 8 images, phenology-based algorithm and Google Earth Engine. Remote Sens. Environ. 2016, 185, 142–154. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  45. Shrestha, R.; Di, L.; Yu, E.G.; Kang, L.; Shao, Y.Z.; Bai, Y.Q. Regression model to estimate flood impact on corn yield using MODIS NDVI and USDA cropland data layer. J. Integr. Agric. 2017, 16, 398–407. [Google Scholar] [CrossRef] [Green Version]
  46. Tian, H.; Wu, M.; Wang, L.; Niu, Z. Mapping Early, Middle and Late Rice Extent Using Sentinel-1A and Landsat-8 Data in the Poyang Lake Plain, China. Sensors 2018, 18, 185. [Google Scholar] [CrossRef]
  47. Tong, X.; Brandt, M.; Hiernaux, P.; Herrmann, S.M.; Feng, T.; Prishchepov, A.V.; Fensholt, R. Revisiting the coupling between NDVI trends and cropland changes in the Sahel drylands: A case study in western Niger. Remote Sens. Environ. 2017, 191, 286–296. [Google Scholar] [CrossRef] [Green Version]
Remotesensing 11 01234 g001 550
Figure 1. Geographic location and land cover map of the study area: (a) general location in China; (b) location in Hunan Province; (c) 1997 land-use classification (prior to Converting Farmland to Lake (CFTL) implementation) from Google Earth Engine (GEE) imagery (see Section 2.3). The red frames show the location of the two typical CFTL areas.
Figure 1. Geographic location and land cover map of the study area: (a) general location in China; (b) location in Hunan Province; (c) 1997 land-use classification (prior to Converting Farmland to Lake (CFTL) implementation) from Google Earth Engine (GEE) imagery (see Section 2.3). The red frames show the location of the two typical CFTL areas.
Remotesensing 11 01234 g001
Remotesensing 11 01234 g002 550
Figure 2. Two examples areas of each cropland change pattern. Corresponding photos a1 and a2, b1 and b2 before and after CFTL are from google earth to demonstrate the process of conversion to lake, poplar cultivation respectively. Figures a1 and b1 represent the years 1997 and figures a2 and b2 the years 2018. Field photographs c1 and c2 were collected in 2018.
Figure 2. Two examples areas of each cropland change pattern. Corresponding photos a1 and a2, b1 and b2 before and after CFTL are from google earth to demonstrate the process of conversion to lake, poplar cultivation respectively. Figures a1 and b1 represent the years 1997 and figures a2 and b2 the years 2018. Field photographs c1 and c2 were collected in 2018.
Remotesensing 11 01234 g002
Remotesensing 11 01234 g003 550
Figure 3. Conceptual models and several examples for cropland conversion in the CFTL: (a) Normalized Difference Vegetation Index (NDVI) trajectory of conversion to lake; (b) NDVI trajectory of conversion to poplar cultivation.
Figure 3. Conceptual models and several examples for cropland conversion in the CFTL: (a) Normalized Difference Vegetation Index (NDVI) trajectory of conversion to lake; (b) NDVI trajectory of conversion to poplar cultivation.
Remotesensing 11 01234 g003
Remotesensing 11 01234 g004 550
Figure 4. Methodological flowchart for (a) pre-CFTL land-use classification and (b) post-CFTL analysis of conversion to lake and poplar cultivation.
Figure 4. Methodological flowchart for (a) pre-CFTL land-use classification and (b) post-CFTL analysis of conversion to lake and poplar cultivation.
Remotesensing 11 01234 g004
Remotesensing 11 01234 g005 550
Figure 5. NDVI range of cropland, conversion to lake, and conversion to poplar cultivation in the study area.
Figure 5. NDVI range of cropland, conversion to lake, and conversion to poplar cultivation in the study area.
Remotesensing 11 01234 g005
Remotesensing 11 01234 g006 550
Figure 6. Year of conversion from cropland to lake with four areas shown in detail: (L1) Datong Lake; (L2) West Dongting Lake; (L3) South Dongting Lake; (L4) East Dongting Lake (scale of the four areas is 1:130,000).
Figure 6. Year of conversion from cropland to lake with four areas shown in detail: (L1) Datong Lake; (L2) West Dongting Lake; (L3) South Dongting Lake; (L4) East Dongting Lake (scale of the four areas is 1:130,000).
Remotesensing 11 01234 g006
Remotesensing 11 01234 g007 550
Figure 7. Year of conversion from cropland to poplar cultivation with four areas shown in detail: (P1) West Dongting Lake; (P2); (P3); (P4) portions of North Dongting Lake (scale of the four areas is 1:130,000).
Figure 7. Year of conversion from cropland to poplar cultivation with four areas shown in detail: (P1) West Dongting Lake; (P2); (P3); (P4) portions of North Dongting Lake (scale of the four areas is 1:130,000).
Remotesensing 11 01234 g007
Remotesensing 11 01234 g008 550
Figure 8. Magnitude and duration of conversion from cropland to (a,b) lake and (c,d) poplar cultivation in the Dongting Lake region. Close up looks for the prevalent regions of two patterns are displayed.
Figure 8. Magnitude and duration of conversion from cropland to (a,b) lake and (c,d) poplar cultivation in the Dongting Lake region. Close up looks for the prevalent regions of two patterns are displayed.
Remotesensing 11 01234 g008
Remotesensing 11 01234 g009 550
Figure 9. Landsat spectral trajectories and LandTrendr fitted trajectories in typical areas (crosses indicate trajectory locations) for (a,C) conversion to lake and (b,D) conversion to poplar cultivation. Related imagery (a1,b1,bands 4–5–3) and corresponding Landsat classification results(a2,b2) are also shown at left.
Figure 9. Landsat spectral trajectories and LandTrendr fitted trajectories in typical areas (crosses indicate trajectory locations) for (a,C) conversion to lake and (b,D) conversion to poplar cultivation. Related imagery (a1,b1,bands 4–5–3) and corresponding Landsat classification results(a2,b2) are also shown at left.
Remotesensing 11 01234 g009
Remotesensing 11 01234 g010 550
Figure 10. Four types of commission errors possible when assessing cropland conversion with LandTrendr: (a) transient floods misclassified as conversion to lake; (b) non-abrupt conversion to lake; (c) reed cultivation producing a similar NDVI signal as poplar cultivation; (d) non-abrupt poplar cultivation. The typical trajectory for each error (blue or purple solid line) and real cropland conversion (black dotted line) were extracted from NDVI series.
Figure 10. Four types of commission errors possible when assessing cropland conversion with LandTrendr: (a) transient floods misclassified as conversion to lake; (b) non-abrupt conversion to lake; (c) reed cultivation producing a similar NDVI signal as poplar cultivation; (d) non-abrupt poplar cultivation. The typical trajectory for each error (blue or purple solid line) and real cropland conversion (black dotted line) were extracted from NDVI series.
Remotesensing 11 01234 g010
Table
Table 1. Datasets used in this study.
Table 1. Datasets used in this study.
DataDescriptionSource
Landsat5
Landsat7
Landsat8 (path/row: 123/040,124/039,124/040)		Annual atmospherically corrected Surface Reflectance Collection from June to September, 1998–2018, for use with LandTrendrPre-Collection 1 archive available from Google Earth Engine
DEMAdvanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data used for modification of classificationASTER Global Emissivity Dataset 100 m V003 available from Google Earth Engine
Field surveyInventory data within the study area, collected in 2018
High-resolution imagesImagery used for
validation of LandTrendr disturbance		Google Earth
Table
Table 2. Yearly change area for both conversions from 1998 to 2018 by area and the conversion area per year as a percentage of the total converted area for all years.
Table 2. Yearly change area for both conversions from 1998 to 2018 by area and the conversion area per year as a percentage of the total converted area for all years.
YearConversion to LakeConversion to Poplar Cultivation
Area (km2)PercentageArea (km2)Percentage
1998258.9858.21%221.8844.39%
19993.080.68%4.730.95%
20000.820.19%8.671.73%
20011.130.28%9.541.91%
20023.550.87%4.030.81%
20035.521.24%6.321.26%
20042.490.54%22.194.44%
20050.910.21%23.034.61%
20061.260.30%7.371.47%
20073.750.84%13.242.65%
20085.931.33%6.501.30%
20092.420.52%7.381.48%
20105.121.10%7.941.59%
20112.910.62%19.533.91%
201214.373.31%8.021.61%
20138.251.80%39.007.80%
201451.3811.07%34.586.92%
201513.833.13%7.771.55%
201631.766.76%9.011.80%
201714.623.47%19.974.00%
201815.403.52%19.193.84%
Total447.48100%499.90100%
Table
Table 3. Confusion matrix for the 1997 classified imagery.
Table 3. Confusion matrix for the 1997 classified imagery.
Validation Data
ClassWaterCroplandForestUrbanShoalUser Accuracy
Water1790000100%
Cropland115324103070.2%
Forest01415311782.7%
Urban031168197.1%
Shoal
Total		0
180		10
180		2
180		1
180		132
180		91.0%
Producer Accuracy99.4%85.0%85.0%93.3%73.3%
Overall accuracy = 87.2%.
Table
Table 4. Accuracy assessments for change detection in both conversion patterns.
Table 4. Accuracy assessments for change detection in both conversion patterns.
Conversion to Lake
Changed PixelsStable PixelsTotalUser Accuracy
Changed pixels2584230086.0%
Stable pixels3626430088.0%
Total294306
Producer Accuracy87.886.3%Overall87.0%
Conversion to Poplar Cultivation
Changed PixelsStable PixelsTotalUser Accuracy
Changed pixels2455530081.7%
Stable pixels4225830086.0%
Total287313
Producer Accuracy85.4%82.4%Overall83.8%
Note: Stable pixels means areas where land cover was persistent throughout the time of the analysis (1998–2018).

Share and Cite

MDPI and ACS Style

Zhu, L.; Liu, X.; Wu, L.; Tang, Y.; Meng, Y. Long-Term Monitoring of Cropland Change near Dongting Lake, China, Using the LandTrendr Algorithm with Landsat Imagery. Remote Sens. 2019, 11, 1234. https://doi.org/10.3390/rs11101234

AMA Style

Zhu L, Liu X, Wu L, Tang Y, Meng Y. Long-Term Monitoring of Cropland Change near Dongting Lake, China, Using the LandTrendr Algorithm with Landsat Imagery. Remote Sensing. 2019; 11(10):1234. https://doi.org/10.3390/rs11101234

Chicago/Turabian Style

Zhu, Lihong, Xiangnan Liu, Ling Wu, Yibo Tang, and Yuanyuan Meng. 2019. "Long-Term Monitoring of Cropland Change near Dongting Lake, China, Using the LandTrendr Algorithm with Landsat Imagery" Remote Sensing 11, no. 10: 1234. https://doi.org/10.3390/rs11101234

APA Style

Zhu, L., Liu, X., Wu, L., Tang, Y., & Meng, Y. (2019). Long-Term Monitoring of Cropland Change near Dongting Lake, China, Using the LandTrendr Algorithm with Landsat Imagery. Remote Sensing, 11(10), 1234. https://doi.org/10.3390/rs11101234

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop