added figure

Author: ssmahto

paper/paper.md

@@ -36,16 +36,16 @@ bibliography: paper.bib
 
 # Statement of need
 
-Accurate information concerning the amount of water stored by dams is crucial for a variety of engineering applications, such as water resources management (@Giuliani:2021), hydrological modelling (@Dang:2020), or hydropower systems management (@Galelli:2022). Unfortunately, these data are rarely available, particularly in developing countries and transboundary river basins, where dam operations are often a point of contention (@Warner:2012). A user necessitating information about dam storage is, in general, left with two alternatives. The first and simplest one is to retrieve such information from a database containing actual observations, such as ResOpsUS (@Steyaert:2022). The problem is that the creation of these databases is still at its infancy; at this stage, their spatial coverage is limited to the United States. The second alternative is to bank on the information contained in satellite observations. By combining information on a reservoir’s bathymetry (elevation-area-storage curve) with information
-on water level (measured by an altimeter), or water surface (retrieved from satellite images), one can, in principle, infer the corresponding time series of water storage (@Gao:2015). The availability of such data has indeed opened up the possibility of monitoring reservoirs in ungauged areas (@Bonnema:2017, @Busker:2019), and also led to the development of online monitoring tools (@Das:2022).
+Accurate information concerning the amount of water stored by dams is crucial for a variety of engineering applications, such as water resources management [@Giuliani:2021], hydrological modelling [@Dang:2020], or hydropower systems management [@Galelli:2022]. Unfortunately, these data are rarely available, particularly in developing countries and transboundary river basins, where dam operations are often a point of contention [@Warner:2012]. A user necessitating information about dam storage is, in general, left with two alternatives. The first and simplest one is to retrieve such information from a database containing actual observations, such as ResOpsUS [@Steyaert:2022]. The problem is that the creation of these databases is still at its infancy; at this stage, their spatial coverage is limited to the United States. The second alternative is to bank on the information contained in satellite observations. By combining information on a reservoir’s bathymetry (elevation-area-storage curve) with information
+on water level (measured by an altimeter), or water surface (retrieved from satellite images), one can, in principle, infer the corresponding time series of water storage [@Gao:2015]. The availability of such data has indeed opened up the possibility of monitoring reservoirs in ungauged areas [@Bonnema:2017; @Busker:2019], and also led to the development of online monitoring tools [@Das:2022].
 
-Creating storage time series for a given reservoir network is, however, not simple: water level observations provided by altimeters are relatively easy to retrieve and use, but are not available for every dam; moreover, satellite missions that included altimeters were launched only recently (@Schwatke2015). Alternatively, satellite images, such as those provided by the Landsat missions, are available for longer periods of time (almost 40 years) and with global coverage. Yet, working with Landsat images requires familiarity with geospatial analysis software (e.g., ArcGIS, QGIS) as well as the statistical or traditional image analysis tools that are needed to process the images affected by cloud cover (@Zhao:2018)—a rather common problem in many regions, particularly during the monsoon seasons. 
+Creating storage time series for a given reservoir network is, however, not simple: water level observations provided by altimeters are relatively easy to retrieve and use, but are not available for every dam; moreover, satellite missions that included altimeters were launched only recently [@Schwatke2015]. Alternatively, satellite images, such as those provided by the Landsat missions, are available for longer periods of time (almost 40 years) and with global coverage. Yet, working with Landsat images requires familiarity with geospatial analysis software (e.g., ArcGIS, QGIS) as well as the statistical or traditional image analysis tools that are needed to process the images affected by cloud cover [@Zhao:2018]—a rather common problem in many regions, particularly during the monsoon seasons. 
 
-Motivated by these modelling challenges, we sought to develop a package that could be potentially used to study any reservoir for a period of time that goes beyond the availability of altimeter observations. This is achieved by developing a package that can extract information from Landsat images, which have a spatial resolution of 30 meters and frequency of 16 days. `InfeRes` was created with this goal: based on the methodology first presented in (@Vu:2022), the NASA Landsat Collection 2 top-of-atmosphere (TOA) reflectance, the OSGeo package (@Coetzee:2020), and scikit-learn library (@Hao:2019). `InfeRes` automates three key modelling steps, namely data download, image processing, and time series reconstruction. With minimal input data required (dam location and design specifications), `InfeRes` is easy to use, even for users unfamiliar with remote sensing and image classification techniques and thus expected to provide an entry point for many users. 
+Motivated by these modelling challenges, we sought to develop a package that could be potentially used to study any reservoir for a period of time that goes beyond the availability of altimeter observations. This is achieved by developing a package that can extract information from Landsat images, which have a spatial resolution of 30 meters and frequency of 16 days. `InfeRes` was created with this goal: based on the methodology first presented in [@Vu:2022], the NASA Landsat Collection 2 top-of-atmosphere (TOA) reflectance, the OSGeo package [@Coetzee:2020], and scikit-learn library [@Hao:2019]. `InfeRes` automates three key modelling steps, namely data download, image processing, and time series reconstruction. With minimal input data required (dam location and design specifications), `InfeRes` is easy to use, even for users unfamiliar with remote sensing and image classification techniques and thus expected to provide an entry point for many users. 
 
 # Functionality
 
-`InfeRes` is available on GitHub [https://github.com/Critical-Infrastructure-Systems-Lab/InfeRes]. Its documentation [https://inferes-test.readthedocs.io/en/latest/index.html] provides a detailed explanation of the installation steps and guidelines for running the code. This includes the preparation of the required modules, which can be easily imported in the Python environment via the pip, conda, or conda-forge package manager.
+`InfeRes` is available on GitHub ['https://github.com/Critical-Infrastructure-Systems-Lab/InfeRes']. Its documentation ['https://inferes-test.readthedocs.io/en/latest/index.html'] provides a detailed explanation of the installation steps and guidelines for running the code. This includes the preparation of the required modules, which can be easily imported in the Python environment via the pip, conda, or conda-forge package manager.
 
 The package's core functionality is divided into two main modules, which are run in sequence. The first module (`data_download.py`) downloads the Landsat imageries using the Earth Engine Python API. To that purpose, the user needs to install the *earthengine-api* package and authenticate with the Google Earth Engine account. In principle, the user can download any set of data from the Google Earth Engine using `data_download.py`; however, for the current objective, we simply use the Normalised Difference Water Index (NDWI) and Quality Assessment Bands (QA_PIXEL) from the Landsat data collection. Alternatively, the user can also download the GREEN and NIR bands to calculate NDWI, instead of downloading NDWI directly from the Earth Engine. Note that more storage and time would be required in such case. The data download module also helps users change the data specifications, such as satellite sensor, area of interest, spatial resolution, and selection of bands. The time needed for data download depends on the size of the Landsat images. For instance, when tested on one of the biggest reservoir area (3010 x 5413 pixels of 30 m resolution), it took around 6-8 hours to download 1330 Landsat images (Landsat 5, 7, and 8), which required nearly 45 GB of storage.