**4. Discussion and Conclusions**

We have presented a set of maps that chart the potential of remote sensing to augment the monitoring and data collection in watersheds for land surface and water quality processes. These are available for open source download in a Git repo [35]. To enable easy access to, and visualization of these maps, the risk factors for impairment, and summary statistics of the potential of remote sensing under various assumptions within the costpayoff matrix, we have also developed an app [36] using Google Earth Engine [37]. This app contains an interactive map within which the user can click on a specific subwatershed of interest to obtain additional information (Figure 6). A summarization of this paper is available as a guide to using the app in a separate panel. Once a subwatershed has been clicked, a summary visual report will be generated, and the relevant information on the impairment risk factors and the potential for remote sensing can be downloaded as comma-separated value files.

**Figure 6.** Google Earth Engine application to demonstrate the potential of remote sensing in monitoring watershed management in the United States. Attribution: Map Data © 2022 Google INEGI Imagery © 2022 NASA, TerraMetrics.

Additionally, we have shown how such maps can be modified within a cost-payoff landscape to be truly customizable by watershed decision-makers. The cost-payoff analysis also shows that improving the data processing pipelines for occluded and poorly lit scenes can tremendously benefit the power of remote sensing (going left to right in Figure 4). By combining the regions of high remote sensing potential with other risk factors describing human footprint, impairment risk, ecosystem vulnerability, and conventional monitoring coverage gaps, we can identify those subwatersheds where remote sensing is likely to have the highest benefit, and in fact, likely be an integral source of primary data. Large areas in the Southern United States, particularly in California, New Mexico, Texas, Mississippi, Georgia, and the Carolinas, and sporadic watersheds in the Northeast and Northwest seaboards (Washington and Maine) and the Midwest would likely benefit most from using remote sensing for watershed monitoring (Figure 5f).

Our approach represents a seminal and necessary step in aiding the decision-making process for resource-constrained regulatory agencies and contractors. By augmenting these maps with other socio-economic geospatial data, deeper insights can be gained to tackle challenges of environmental justice and equity. Rather than focusing on specific watershed processes, impairments, and remote sensing platforms, we have developed a method of assessing the general potential for remote sensing for whatever applications are envisioned by system managers in their monitoring and data collection workflows. This approach allows us to explore the benefits of remote sensing further, as (i) not all parts of the country are monitored regularly, and even when they are, in-situ monitoring costs can typically exceed several hundred thousands of dollars a year within a small watershed [38], and (ii) remote sensing can provide long-term cost savings. Remote sensing potential maps could be constructed for specific applications outlined in the introduction by suitably modifying the factors considered in the model presented in Section 2.1. For example, if a consultant proposes a data fusion pipeline using Sentinel 2, MODIS, and Planet imagery [39] to study watershed health, heavy metal concentrations in a nearby lake, and the role of watershed best management practices, then the principal costs of image acquisition will likely be the cost of procuring high-resolution Planet imagery, while in-situ monitoring costs will likely be driven by personnel costs rather than access time.

There are several caveats to the maps we have produced here. First, the assumptions built into the EarthEnv cloud cover and the Malaria Atlas travel time databases carry over to our maps. While these data sources are excellent for global coverage at the kilometer-scale, our pipeline may be refined with source DEMs and spectral imagery for specific watersheds to systematically downscale our national coverage maps. The shortest time to reach a population center, while somewhat ad-hoc, is an appropriate metric to characterize the difficulty of ground-truthing remote sensing imagery because typically waterbodies outside major population centers are not monitored. However, as we show in Figure 5d, such areas are also at high risk of impairment. Second, our remote sensing potential model is extremely simple and steady state, and yet captures the underlying tradeoff between scene acquisition and ground-truthing. Within a watershed, multiple process-scales exist, ranging from storms and coastal upwelling events that happen on hourly to daily timescales, land use and land cover changes that happen on annual to decadal timescales, and landform changes that happen over many decades to centuries [40]. Our maps do not factor in these different timescales. More sophisticated models, including spatial kriging approaches and Gaussian process models that incorporate seasonal variability in scene acquisition and travel time can be developed. Cloud cover, for instance, can vary seasonally so that even in areas with many cloudy days, there can be periods with clear skies during which period field campaigns could be planned. We do not capture such nuances here. For simple applications that require only a single remote sensing image such as a National Agricultural Image Program [41] or LandSat scene, cloud cover may not even be an issue if the acquisition window can be carefully determined. Although many non-dimensional metrics for land use and land cover classification for characterizing the environment can be derived from multi- and hyperspectral imagery in most locations within the United States, as we showed in Figure 1, there are still significant data gaps that require physical ground-truthing for which our maps will be useful. Third, we have operated at the HUC-12 subwatershed scale in this project for computational tractability. However, much finer reach-scale resolution of the impaired waterbodies can be incorporated within our pipeline by combining the NHDplus and ATTAINS datasets using cloud platforms such as Google's Earth Engine [37]. Fourth, our thresholds for cloud cover and minimum travel time to population centers for the various cost-payoff scenarios are somewhat arbitrary, but still indicate sensible and meaningful trends in the remote sensing potential maps. This is a reflection of the fact that we have sacrificed specificity to different remote sensing technologies for the generality of our guidance maps. These thresholds can be more rigorously defined for different types of remote sensing platforms by combining the atmospheric conditions and ground-truthing needs of specific technologies. Fifth, while we have chosen risk metrics that largely encompass most risk factors for watershed impairments, a more thorough exploration of available datasets, particularly socioeconomic data, is possible to identify other, transdisciplinary benefits of remote sensing within a watershed. It is likely that such exhaustive inclusion must be done on a local municipal, county, or state scale than on the national scale for disparate factors to sensibly be combined at meaningful physical scales. In subsequent iterations of these products, we will tackle these advances. Sixth, we only considered terrestrial subwatersheds in our current pipeline, and excluded all coastal and

Great Lakes impairments. In subsequent iterations of these maps, we will include these areas as well.

While we have developed the pipeline for the United States, a similar workflow could be developed anywhere in the world, or even for the entire planet. The only consideration is that while Earth Explorer, the United States Environmental Protection Agency, and the EnviroAtlas catalog a wide array of geospatial datasets for immediate use, such data may be difficult to find in the international landscape, and may have to be stitched together from disparate sources of information. Nonetheless, for any serious, reproducible, and scalable application of remote sensing for monitoring and data collection on watershed loads and impairments, this is a necessary first step. For the first time ever, a product is available that will allow researchers and managers to evaluate the potential of remote sensing to augment, or even fulfill ambient monitoring and data collection needs throughout the United States. We hope that our maps will be useful for the watershed managers within the United States, even as we continue to improve them with finer granularity and additional factors.

**Author Contributions:** Conceptualization, V.K.S., S.K. and S.M.K.; methodology, V.K.S. and S.K.; software, V.K.S. and S.K.; validation, V.K.S. and S.K.; formal analysis, V.K.S.; investigation, V.K.S.; resources, V.K.S., S.K. and S.M.K.; data curation, V.K.S.; writing—original draft preparation, V.K.S.; writing—review and editing, V.K.S. and S.K.; visualization, V.K.S.; supervision, V.K.S. and S.K.; project administration, V.K.S.; funding acquisition, Unfunded study. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Institutional review and approval was not applicable for this study.

**Informed Consent Statement:** Informed consent was not applicable for this study as there were no human subjects.

**Data Availability Statement:** Data available in a publicly accessible repository that does not issue DOIs here: https://github.com/vamsiks2003/remoteSensingPotentialMapping (accessed on 20 April 2022). Publicly available datasets were analyzed in this study. This data can be found here: https:// www.usgs.gov/national-hydrography/national-hydrography-dataset, https://earthexplorer.usgs.gov, https://www.earthenv.org/cloud, https://malariaatlas.org/research-project/accessibility-to-cities/, https://www.census.gov/geographies/mapping-files/time-series/geo/cartographic-boundary.html, https://www.epa.gov/waterdata/attains, https://www.epa.gov/enviroatlas (accessed on 20 April 2022). Data citation: Sridharan, V.K., Kumar, S.N., and Madhur Kumar, S. 2021. GitHub Repo remoteSensingPotentialMapping (https://github.com/vamsiks2003/remoteSensingPotentialMapping (accessed on 20 April 2022)).

**Acknowledgments:** Discussions with Rocky Talchabadel and Santosh Palmate at Texas A&M University were useful in the development of the methods presented here. Critical review by Lee Harrison at the University of California, San Diego and recommendations by three anonymous reviewers and the Academic Editor at Water significantly improved the quality of this manuscript. The lead author acknowledges the Environmental and Water Resources Institutes Remote Sensing Task Committee, Total Maximum Daily Load Analysis and Modeling Task Committee and the Watershed Management Technical Committee whose deliberations motivated this study.

**Conflicts of Interest:** The authors declare no conflict of interest.
