Skip to main content
NCBI home page
NASA Author Manuscripts logoLink to NASA Author Manuscripts
. Author manuscript; available in PMC: 2022 Jan 27.
Published in final edited form as: Geophys Res Lett. 2018 Jun 12;45(12):6092–6099. doi: 10.1029/2018gl078343

Automatic correction of contaminated images for assessment of reservoir surface area dynamics

Gang Zhao 1, Huilin Gao 1,*
PMCID: PMC8793793  NIHMSID: NIHMS975418  PMID: 35095126

Abstract

The potential of using Landsat for assessing long-term water surface dynamics of individual reservoirs at a global scale has been significantly hindered by contaminations from clouds, cloud shadows, and terrain shadows. A novel algorithm was developed towards the automatic correction of these contaminated image classifications. By applying this algorithm to the dataset by Pekel et al. (2016), time series of area values for 6817 global reservoirs (with an integrated capacity of 6099 km3) were generated from 1984 to 2015. The number of effective images that can be used in each time series has been improved by 81% on average. The long-term average area for these global reservoirs was corrected from 1.73×105 km2 to 3.94×105 km2. The results were proven to be robust through validation using observations, synthetic data, and visual inspection. This continuous reservoir surface area dataset can provide benefit to various applications (both at continental and local scales).

Keywords: Landsat, Reservoir area, Contamination, Image enhancement, Google Earth Engine

Plain Language Summary

Understanding of water surface area dynamics is important for modern water resources management. Due to the difficulties collecting data from ground, remote sensing images from satellites have been widely used to map the water coverage and then to analyze the dynamics. However, one typical problem when using satellite images is that they are frequently contaminated by clouds, cloud shadows, and terrain shadows, which result in the underestimation of the water area. Thus, we developed a novel algorithm to remove the impacts of these contaminations for monitoring water area accurately. After comprehensive evaluations, the algorithm was proved to be able to effectively enhance the Landsat-based water area results. A dataset containing monthly surface area time series for 6817 global reservoirs from 1984 to 2015 was subsequently generated using the algorithm. This work (both the dataset and the algorithm) can support many applications on both global and local scales to benefit the water management, hydrology, and remote sensing communities.

1. Introduction

Reservoir storage variation is a key factor in regional water resources management. However, data access to most global reservoirs is very limited—especially for developing countries (Medina et al., 2008; Zhang et al., 2006). By combining remotely sensed water surface area and radar/lidar altimeter data, time series datasets of reservoir storage can be derived (Gao et al., 2012; Lettenmaier et al., 2015; Zhang et al., 2014). Continuous time series of reservoir surface area is also critical for many other applications. For example, seasonal/inter-annual variations of reservoir evaporation—which directly affect local water availability—are primarily driven by changes of reservoir surface area (Zhang et al., 2017). Additionally, evaluation of reservoir surface area dynamics can facilitate inundated area estimations (Jim et al., 2010), ecology and phenology studies (Gorham and Boyce, 1989), climate change assessments (Benson and Paillet, 1989; St. Louis et al., 2000), and hydrological model validations (Zhou et al., 2016).

Reservoir surface area variations can be captured using different satellite sensors—including the widely used Moderate Resolution Imaging Spectroradiometer (MODIS; Deus and Gloaguen, 2013; Klein et al., 2017; Zhang and Gao, 2016), and the Thematic Mapper (TM) on the Landsat 4–5 satellites (Donchyts et al., 2016; Pekel et al., 2016). Compared to MODIS, Landsat has a much finer spatial resolution (30m), and thus can detect significantly smaller water bodies with high accuracy (Khandelwal et al., 2017). More importantly—as one of the earliest satellite missions for Earth observations—Landsat has a unique advantage for assessing long-term water surface dynamics.

Leveraging the long term Landsat missions, Pekel et al. (2016) generated a consistent global surface water dataset (GSWD) from 1984 to 2015. In theory, monthly time series of water surface area for any Landsat-detectable reservoirs/lakes can be extracted from this dataset. However, the results can be significantly underestimated for some reservoirs/lakes because all of the contaminated pixels (due to clouds, cloud shadows, terrain shadows, and the Scan Line Corrector (SLC) failure) have been classified as “no data” in GSWD. For quality control purposes, most previous studies chose to discard all contaminated images (Huang et al., 2010; Vicente-Serrano et al., 2008). However, this approach does not work well for regions with frequent cloud cover, nor for areas with Landsat 7 images containing SLC errors. For instance, if all images with more than 5% contaminated pixels are removed, then only 30% of the images collected from 1984 to 2015 would be available for the Danjiangkou Reservoir in China.

Even though partially contaminated images lead to underestimated water surface areas, they do contain valuable information. For instance, the uncontaminated part of the reservoir water area can offer a clue about the total water area. By removing these images, such information is discarded rather than leveraged. To our best knowledge, there are very few studies that have used this information to improve water surface area estimations. While several relevant studies have employed elevation information (Li et al., 2013; Sun et al., 2012) and multi-temporal analysis (Yamazaki et al., 2015; Zhang et al., 2014) to correct classifications of contaminated pixels, there is still a lack of an automatic and robust algorithm to accurately correct all types of contaminated pixels.

The overarching objective of this study is to generate a high-quality time series dataset of surface area for the global reservoirs. This effort is built upon the GSWD, which has provided raw global monthly water coverage data from March 1984 to October 2015. To achieve this goal, we developed a novel algorithm to automatically repair the contaminated water classification images from GSWD. Using this algorithm, a dataset containing surface area time series values of 6817 global reservoirs was generated. Although our dataset only focuses on reservoirs, the algorithm is also applicable to any natural lakes that can be detected by Landsat. All of the calculations were implemented on the Google Earth Engine (GEE) platform, which is a cloud-computing platform used to facilitate planetary scale geo-data processing (Gorelick et al., 2017).

2. Data and method

2.1 Data for estimating reservoir area

The reservoirs of interest were identified from the Global Reservoir and Dam dataset (GRanD; Lehner et al., 2011), which contains 6862 reservoirs. By excluding 45 reservoirs without reported shapefiles, 6817 reservoirs (with an integrated capacity of 6099 km3) were selected for this study.

To generate reservoir surface area time series, two types of data from the GSWD were used. The first is the monthly water coverage map, which includes three classes: “water”, “not water”, and “no data”. A pixel is considered “no data” if it meets one of these three conditions: the pixel is contaminated by cloud, cloud shadow, terrain shadow, or SLC failure; the Landsat scene for the location is unavailable; or the pixel falls in a region which was not well illuminated when the Landsat image was collected. To ensure the accuracy of its water detection algorithm, GSWD removed the low illuminated regions (see Figure S1 in Supporting Information). Although our image enhancement algorithm focuses on reclassifying the first type of “no data”, it can also correct the second and the third types – if the reservoir is only partially covered by such “no data” pixels. Because the GSWD monthly water coverage maps serve as the inputs of our algorithm, they are referred to as “raw water area” hereinafter.

The second source of GSWD data used is the water occurrence, which was generated by stacking all of the monthly water coverage images mentioned above. Each pixel value represents the frequency (in percentage) that the pixel was covered by water during the 32-year period. The water occurrence at a given reservoir contains information similar to a bathymetry image (see Figure S2 for an example). This is valid because there is a unique area-elevation curve for any given reservoir.

2.2 Algorithm for water classification enhancement

This algorithm is based on the assumption that if all uncontaminated pixels with the same water occurrence value of α are classified as water, then all contaminated pixels that have water occurrence values greater than or equal to α should also be covered by water. For each reservoir, a mask is first generated by buffering 500 m outwards from the GRanD shapefile. The algorithm is executed within the masked area for computation efficiency and classification accuracy. Using a 41.7% contaminated image (randomly selected) as an example (Figure 1a), the steps involved in the algorithm are described as follows (flowchart as shown in Figure S3):

Figure 1.

Figure 1.

Open in a new tab

Automatic correction of a contaminated image: a) Contaminated image; b) Raw water area; c) Water occurrence; d) Clipped occurrence; e) Histogram; and f) Enhanced water area.The image used in this example is for the E.V. Spence reservoir in Texas, and was acquired on June 15, 2003.

1) Clipping the raw monthly water coverage maps from GSWD. The reservoir water coverage for each month is clipped out using the reservoir mask from the GSWD monthly data (Figure 1b).

2) Filtering contaminated images. If the percentage of “no data” pixels within the mask is larger than or equal to 95%, the image is discarded due to extreme contamination. If the percentage is lower than 5%, the image is treated as a high-quality image and the enhancement is not applied to it. Otherwise, Steps 3–7 are executed.

3) Clipping the water occurrence map from GSWD. The water occurrence is clipped out from the GSWD global water occurrence map using the reservoir mask (Figure 1c).

4) Clipping the occurrence image with raw water coverage. The pixels from the water occurrence that overlapped with the raw water area are extracted (Figure 1d).

5) Generating the histogram. The histogram of the occurrence image from Step 4 is generated, which shows the count of pixels at each water occurrence value (Figure 1e).

6) Calculating key thresholds. The average value of occurrence counts is first calculated. Then it is multiplied by a weighting factor ω (0.17) to generate the “Count threshold”. The first occurrence count that is greater than or equal to the “Count threshold” – which is defined as the “Occurrence threshold” – is then identified. Here the weighting factor (ω) 0.17 is found to perform well for most reservoirs, even though it shows limited sensitivity (see Figure S4).

7) Enhancing the water area classifications. Any “no data” pixel in the raw image which corresponds to an occurrence value larger or equal to the “Occurrence threshold” is reclassified as “water” to construct the “Enhanced water area” (Figure 1f).

In this algorithm, the enhancement adopted the “Occurrence threshold”, rather than the lowest occurrence value (𝛼𝑚𝑖𝑛). Theoretically, all pixels in the water occurrence that have values greater than or equal to 𝛼𝑚𝑖𝑛 should be covered by water. However, high spatial frequency noise (e.g., commission error due to shadow, or tiny water bodies surrounding the reservoir) can result in unrealistically low 𝛼𝑚𝑖𝑛 value. As a result, using 𝛼𝑚𝑖𝑛 as the threshold to extract real water coverage can lead to significant overestimation (Figure S5). In contrast, the influence of high spatial frequency noise from the raw water area image can be avoided by adopting the “Occurrence threshold”.

2.3 Surface area time series for the global reservoirs

Using the aforementioned algorithm, the reservoir surface area can be calculated accurately from a contaminated water coverage map. However, surface area cannot be estimated for months when more than 95% of the reservoir mask area is covered by “no data” pixels. Therefore, the missing monthly area estimates were filled in using linear interpolation.

3. Results

3.1 Algorithm validation

The algorithm performance was evaluated from three perspectives: 1) comparing calculated surface area with observed storage/elevation values; 2) testing the enhancements of synthetically contaminated images at a global scale; and 3) inspecting the enhancement results visually.

First, the resultant time series were validated over nine reservoirs using elevation or storage measurements from in-situ or satellite altimeters (Figure 2; Text S2 for data sources). These reservoirs represent different continents with a wide range of sizes. The time series of the raw area frequently show erroneous low values caused by contaminations. In the enhanced time series, the coefficient of determination (R2) between the surface areas and the observed elevation/storage values has increased for all reservoirs. For example, the R2 for the Waranga Basin reservoir (Australia) was improved from 0.22 to 0.83.

Figure 2.

Figure 2.

Open in a new tab

Comparison of calculated surface area with observed storage/elevation for nine reservoirs. The two numbers in the parenthesis indicate the values of R2 between observed storages (or elevations) and water areas before and after enhancement.

Second, validations at a global scale using synthetic data were carried out using the following steps:

1) For each of the 6817 reservoirs, a clear image (i.e., with less than 5% contamination) was randomly selected from March 1984 to October 2015. The water area of this image was regarded as the true state of the reservoir.

2) A contaminated image (i.e., between 5% and 95% contamination) of this reservoir was randomly selected.

3) A synthetic contaminated image was constructed by overlaying the contaminated pixels (from Step 2) to the clear image (from Step 1).

4) The algorithm was applied to the synthetic image to calculate the enhanced water area.

5) Relative bias at each reservoir was quantified by comparing the enhanced water area with the water area from the clear image.

Over each reservoir, we also calculated the surface area bias of the synthetic image from the clear image for comparison. Figure 3a shows that the R2 was improved from 0.735 to 0.998.

Figure 3.

Figure 3.

Open in a new tab

a) Comparisons among the area estimations from the clear, raw, and enhanced images. b) Relationship between contamination percentage and relative bias. c) Relationship between the occurrence with maximum contamination and the relative biases.

Specifically, the R2 value for the reservoirs that are larger than 10 km2 was improved from 0.598 to 0.997. The impacts of contamination percentage and reservoir size on the algorithm performance are shown in Figure 3b. Overall, the average relative bias is 3.7% (4.0% positive and −3.6% negative). The algorithm tends to slightly overestimate the area when contamination percentage is low, and underestimate the area when the contamination percentage is high, especially for the small reservoirs less than 1 km2. In addition, the occurrence with the maximum contamination fraction was identified for each synthetic image. This maximum contaminated occurrence was compared with the corresponding relative biases from the enhanced area. Results from Figure 3c suggest that the algorithm’s performance is not affected by the locations where the contamination occurs.

Third, the validity of the image enhancement algorithm was also examined by visually comparing the raw and enhanced water classifications. Classification images from eight reservoirs —across different continents on different dates—are used to show the corrections of a variety of issues associated with the raw water coverage maps (Figure S6). For instance, the case of Lake Mead shows that the algorithm can improve the results by repairing the raw image with a missing Landsat scene (i.e., the second type “no data”). The examples of the Kamianske Reservoir and the Toledo Bend Reservoir show the effectiveness of correcting SLC failures.

3.2 Improvements over the raw water area estimations

By repairing the raw water area associated with the 6817 reservoirs, the resultant surface area time series have been improved significantly—especially for the reservoirs that are located in the regions prone to cloudy conditions. The improvement ratio (I), which represents the ratio of the number of repaired images (NR) to the number of initially effective images (NC) for a given reservoir, is defined in Equation 1,

I​=NRNC⋅100%

Here an effective image refers to a raw image with less than 5% contamination. For all the images collected by Landsat over these 6817 global reservoirs (during the study period), the averaged value of the improvement ratio is 81%.

3.3 Surface area dynamics of global reservoirs

Results from this study can provide an improved understanding about the surface area dynamics of global reservoirs. In Figure 5, the four time series – raw, enhanced, interpolated raw, and interpolated enhanced – of the total surface area of the 6817 reservoirs from 1984 to 2015 are compared.

Figure 5.

Figure 5.

Open in a new tab

Time series of global reservoir surface area variations from March 1984 to October 2015 (6817 reservoirs in total).

There is a clear underestimation with regard to global area time series based on the raw water coverage. This is caused by both image contaminations (the first type of “no data”) and the partial coverage of GSWD (the second and third types). In particular, partial coverage due to low illumination (third type) has resulted in missing area values (i.e., zeros) for the high latitude reservoirs in the northern hemisphere during the low illumination winter months.

By using the linear interpolation technique to fill in the missing values in the time series, the underestimation during the low-data-coverage months in the raw area time series are notably reduced. Still, the effectiveness of interpolation depends on the quality of the raw area estimations. If any of the area values used for interpolation are estimated from contaminated raw images, underestimations of the raw images will be transferred to the interpolated results.

In contrast, the time series after the application of the image enhancement algorithm does not suffer from the underestimation due to contamination. However, improvements for the low-data-coverage months are quite small (particularly before 2000) because many reservoirs in these months have no raw water area to enhance. With the launch of Landsat 7 in 1999, the number of images collected has increased significantly (Pekel et al., 2016). This has directly helped to reduce the second type of “no data”. As a result, not only has the global reservoir area based on the raw water coverage increased, but the benefits of the enhancement during the winter months have become evident (in the Northern Hemisphere).

With both the image enhancement and the interpolation applied over each reservoir, global area estimations are significantly improved. The underestimations from both contamination and data coverage issues have been corrected. From 1984 to 2015, the average global reservoir area from this new dataset is 3.94×105 km2, while its counterpart from the raw water maps is only 1.73×105 km2.

From 1984 to around 2000, there is a clear increasing trend in the total surface area. This trend is mainly caused by the 849 new reservoirs (with a total capacity of 903 km3) constructed during this period. From 2001 to 2011 (i.e., the last year documented in the GRanD dataset), the trend of total surface area barely increased because there were only 63 new reservoirs (with a total capacity of 121 km3) constructed. In the most recent years (from 2011 to 2015), the area has actually decreased. This is mainly caused by the reduced precipitation over land (according to our calculation using the PERSIANN-CDR dataset). The area dynamics are also consistent with the reservoir storage variations modeled by Zhou et al. (2016).

4. Discussion and Conclusions

The GSWD dataset has offered a first-of-its-kind record of global surface water coverage. However, due to multi-source contaminations of Landsat images, the raw water area estimations for a given reservoir directly extracted from the GSWD are not accurate enough to support decision making. In this study, a novel algorithm was developed to repair the contaminated water areas for high quality area estimations. This algorithm and its resulting enhanced area estimations were proven robust through comprehensive validation exercises. A dataset containing monthly area time series for 6817 global reservoirs was then generated. Compared with the raw water area images, the average improvement ratio is 81% for these reservoirs after image enhancement. Furthermore, the gaps in the time series for each reservoir (caused by partial coverage of GSWD) were filled in using linear interpolation. Consequently, our dataset has corrected the global reservoir area from 1.73×105 km2 to 3.94×105 km2.

As explained in the Introduction, the continuous reservoir surface area dataset can provide benefit to various applications (both on continental and local scales). For instance, more accurate reservoir storage values can be calculated using this dataset to better assess water availability for decisions related to municipal, agricultural, and industrial water supplies. Additionally, detailed reservoir surface area dynamics—including both seasonal/inter-annual variability and long-term trends—are essential information for studies involving ecosystem assessments, hydrologic modelling, water-energy-food nexus evaluations, and geological processes. An example of how the recent California drought impacted reservoir surface areas is given in the Figure S7. Furthermore, as shown in Figure 2, reservoir surface area dynamics can be evaluated. The example of Lake Mead (Figure 2a) shows continuous and unsustainable depletion of the reservoir storage, which is mainly caused by the changing climate and increasing consumptive water use (Barnett and Pierce, 2008).

The enhancement algorithm not only works for the GSWD dataset, it can also be applied to the classification results from Landsat raw scenes and products from other sensors (e.g., MODIS). Even though our algorithm has performed well over most reservoirs, it has a few limitations: First, the enhancement algorithm is inoperable for the months that have no data at all over a given reservoir, or when the contamination percentage is greater than 95%. Second, the precision of the enhanced time series can be slightly affected by the selection of the empirical weighting factor (ω).

Supplementary Material

1

Figure 4.

Figure 4.

Open in a new tab

Improvement ratios for the 6817 global reservoirs. Average cloud fraction was calculated using the MOD08_M3 monthly product from March 2000 to December 2015.

Key points.

  • Image contamination significantly hinders the continuous and accurate assessment of water dynamics using Landsat imagery

  • A novel algorithm is developed to automatically repair contaminated Landsat images for generating more reliable surface area time series

  • The number of effective images that can be used in the time series is improved by 81% on average for 6817 global reservoirs

Acknowledgment

The dataset containing the surface area time series of the 6817 global reservoirs will be publicly available online at https://ceprofs.civil.tamu.edu/hgao/.

This research was supported by the NASA Science of Terra, Aqua, and Suomi NPP (TASNPP) Program (80NSSC18K0939) and the Earth and Space Science Fellowship (NESSF) Program (17-EARTH17F-0297) provided to Texas A&M University. This research has also benefitted from the usage of the Google Earth Engine platform (https://earthengine.google.com) and the Texas A&M Supercomputing Facility (http://sc.tamu.edu). We would also like to thank the Editor, Valeriy Ivanov, and the two anonymous reviewers for helping to improve the manuscript.

References

  1. Abatzoglou JT, Dobrowski SZ, Parks SA, & Hegewisch KC (2018). TerraClimate, a high-resolution global dataset of monthly climate and climatic water balance from 1958– 2015. Scientific data, 5, 170191. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Barnett TP, & Pierce DW (2008). When will Lake Mead go dry?, Water Resources Research, 44(3). [Google Scholar]
  3. Benson L, & Paillet FL (1989). The use of total lake-surface area as an indicator of climatic change: examples from the Lahontan Basin. Quaternary Research, 32(3), 262–275. [Google Scholar]
  4. Deus D, & Gloaguen R (2013). Remote sensing analysis of lake dynamics in semi-arid regions: Implication for water resource management. Lake Manyara, East African Rift, Northern Tanzania. Water, 5(2), 698–727. [Google Scholar]
  5. Donchyts G, Baart F, Winsemius H, Gorelick N, Kwadijk J, & van de Giesen N (2016). Earth’s surface water change over the past 30 years. Nature Climate Change, 6(9), 810–813. [Google Scholar]
  6. Gao H, Birkett C, & Lettenmaier DP (2012). Global monitoring of large reservoir storage from satellite remote sensing. Water Resources Research, 48(9). [Google Scholar]
  7. Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, & Moore R (2017). Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202, 18–27. [Google Scholar]
  8. Gorham E, & Boyce FM (1989). Influence of lake surface area and depth upon thermal stratification and the depth of the summer thermocline. Journal of Great Lakes Research, 15(2), 233–245. [Google Scholar]
  9. Huang C, Goward SN, Masek JG, Thomas N, Zhu Z, & Vogelmann JE (2010). An automated approach for reconstructing recent forest disturbance history using dense Landsat time series stacks. Remote Sensing of Environment, 114(1), 183–198. [Google Scholar]
  10. Jim C, Yang FY, & Wang L (2010). Social-ecological impacts of concurrent reservoir inundation and reforestation in the three gorges region of China. Annals of the Association of American Geographers, 100(2), 243–268. [Google Scholar]
  11. Khandelwal A, Karpatne A, Marlier ME, Kim J, Lettenmaier DP, & Kumar V (2017). An approach for global monitoring of surface water extent variations in reservoirs using MODIS data. Remote Sensing of Environment, 202, 113–128. [Google Scholar]
  12. Klein I, Gessner U, Dietz AJ, & Kuenzer C (2017). Global WaterPack–A 250 m resolution dataset revealing the daily dynamics of global inland water bodies. Remote Sensing of Environment, 198, 345–362. [Google Scholar]
  13. Lehner B, Liermann CR, Revenga C, Vörösmarty C, Fekete B, Crouzet P, . . . Magome J. (2011). High‐resolution mapping of the world’s reservoirs and dams for sustainable river‐flow management. Frontiers in Ecology and the Environment, 9(9), 494–502. [Google Scholar]
  14. Lettenmaier DP, Alsdorf D, Dozier J, Huffman GJ, Pan M, & Wood EF (2015). [Google Scholar]
  15. Inroads of remote sensing into hydrologic science during the WRR era. Water Resources Research, 51(9), 7309–7342. [Google Scholar]
  16. Li S, Sun D, Goldberg M, & Stefanidis A (2013). Derivation of 30-m-resolution water maps from TERRA/MODIS and SRTM. Remote Sensing of Environment, 134, 417–430. [Google Scholar]
  17. Medina CE, Gomez-Enri J, Alonso JJ, & Villares P (2008). Water level fluctuations derived from ENVISAT Radar Altimeter (RA-2) and in-situ measurements in a subtropical waterbody: Lake Izabal (Guatemala). Remote Sensing of Environment, 112(9), 3604–3617. [Google Scholar]
  18. Ozesmi SL, & Bauer ME (2002). Satellite remote sensing of wetlands. Wetlands ecology and management, 10(5), 381–402. [Google Scholar]
  19. Pekel J-F, Cottam A, Gorelick N, & Belward AS (2016). High-resolution mapping of global surface water and its long-term changes. Nature, 540(7633), 418–422. [DOI] [PubMed] [Google Scholar]
  20. St. Louis VL, Kelly CA, Duchemin É, Rudd JW, & Rosenberg DM. (2000). Reservoir Surfaces as Sources of Greenhouse Gases to the Atmosphere: A Global Estimate: Reservoirs are sources of greenhouse gases to the atmosphere, and their surface areas have increased to the point where they should be included in global inventories of anthropogenic emissions of greenhouse gases. AIBS Bulletin, 50(9), 766–775. [Google Scholar]
  21. Sun F, Sun W, Chen J, & Gong P (2012). Comparison and improvement of methods for identifying waterbodies in remotely sensed imagery. International journal of remote sensing, 33(21), 6854–6875. [Google Scholar]
  22. Bureau of Reclamation US (2011), 2009 Lake Mead LiDAR Survey, Boulder City, Nevada [Google Scholar]
  23. Vicente-Serrano SM, Pérez-Cabello F, & Lasanta T (2008). Assessment of radiometric correction techniques in analyzing vegetation variability and change using time series of Landsat images. Remote Sensing of Environment, 112(10), 3916–3934. [Google Scholar]
  24. Yamazaki D, Trigg MA, & Ikeshima D (2015). Development of a global~ 90 m water body map using multi-temporal Landsat images. Remote Sensing of Environment, 171, 337–351. [Google Scholar]
  25. Zhang S, & Gao H (2016). A novel algorithm for monitoring reservoirs under all‐weather conditions at a high temporal resolution through passive microwave remote sensing. Geophysical Research Letters, 43(15), 8052–8059. [Google Scholar]
  26. Zhang S, Gao H, & Naz BS (2014). Monitoring reservoir storage in South Asia from multisatellite remote sensing. Water Resources Research, 50(11), 8927–8943. [Google Scholar]
  27. Zhang J, Xu K, Yang Y, Qi L, Hayashi S, & Watanabe M (2006). Measuring Water Storage Fluctuations in Lake Dongting, China, by Topex/Poseidon Satellite Altimetry. Environmental Monitoring and Assessment, 115(1), 23–37. [DOI] [PubMed] [Google Scholar]
  28. Zhang H, Gorelick SM, Zimba PV, & Zhang X (2017). A remote sensing method for estimating regional reservoir area and evaporative loss. Journal of Hydrology, 555, 213–227. [Google Scholar]
  29. Zhou T, Nijssen B, Gao H, & Lettenmaier DP (2016). The contribution of reservoirs to global land surface water storage variations. Journal of Hydrometeorology, 17(1), 309–325. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS975418-supplement-1.pdf (1.3MB, pdf)

RESOURCES