**1. Introduction**

Of the 37.6 million waterbodies including canals, stream segments, ponds, and lakes in the United States, fewer than three million are monitored in situ, with only about 60,000 monitoring sites providing information that can be compared with remotely sensed data (Figure 1a) [1,2]. Of the millions of waterbodies, only 430,893 have been assessed for water quality impairments as of 2022 (Figure 1b) [3,4]. Within the recent past, even for the assessed waterbodies, ambient monitoring of general state of the water quality and synoptic data collection of the load, transport, and fate processes associated with impairment happens sporadically in only about a third of the cases [5]. Remote sensing, both with airborne and space-borne platforms offers tremendous potential for bridging these massive data gaps. But there are significant data gaps to even be able to use remote sensing everywhere in the United States (Figure 1a). Our contribution herein is, albeit with several simplifying assumptions, to map out the potential for remote sensing to be a viable monitoring tool for all the subwatersheds—the smallest catchment class associated with waterbodies as defined by the United States Geological Survey (USGS)—within the conterminous United

**Citation:** Sridharan, V.K.; Kumar, S.; Madhur Kumar, S. Can Remote Sensing Fill the United States' Monitoring Gap for Watershed Management?. *Water* **2022**, *14*, 1985. https://doi.org/10.3390/w14131985

Academic Editor: Thomas Meixner

Received: 1 May 2022 Accepted: 20 June 2022 Published: 21 June 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

States. As a corollary to these maps, we also present a geospatial estimate of where remote sensing is likely to have the greatest impact in the country by mapping the intersection of remote sensing potential and high risk of impairment and low data coverage. These maps will serve as decision support tools for researchers and practitioners to plan where to invest their energies and resources with respect to data collection for water quality management approaches.

**Figure 1.** Assessment and in-situ monitoring gap for water quality in the United States: (**a**) 39.6 million stream segments, ponds, lakes and estuaries in the National Hydrography Dataset and 603,433 in-situ monitoring locations from the AquaSat database for Secchi disk depth, Chlorophyll-*a*, total suspended solids and dissolved organic carbon that have been matched to landSat scenes, and (**b**) the 430,893 assessed waterbodies in the United States obtained from the United states Environmental Protection Agency's Assessment and Total Maximum Daily Load Tracking and Implementation System (ATTAINS) database. Red lines indicate state borders.

Our focus in this paper is not to limit our treatment to a specific land surface process, water quality parameter, type of impairment or remote sensing data source. This is because water quality is a complex system response within the watershed to surface, subsurface and hydrologic processes that are modulated by human influence. Therefore, rather than focusing on a specific aspect of watershed processes, we develop a comprehensive scoping tool for adopting remote sensing in monitoring and data collection. Scoping tools such as the one presented here can be developed for specific land surface process monitoring, or water quality management of given impairments by using information relevant to certain remote sensing platforms and data collection methods.

Over the past two decades, a large number of many aerial and space-borne platforms and sensors that sample different parts of the electromagnetic spectrum have come online [6]. Platforms such as LandSat, GeoEye, WorldView, and technologies such as Light Detection and Ranging (LiDAR) are useful for water body delineation (e.g., [7]), while bathymetry of waterbodies can be obtained either from LiDaR, or by analyzing spectral band ratios of satellite and drone imagery [8,9]. Streamflow can be obtained using Radio Detection and Ranging (RaDAR) altimetry and rating curves from platforms such as Jason-3, Sentinel-3, and Saral/ALtika [10]. The terrestrial water budget can be quantified using gravimetric data, while floods can be characterized using optical sensors such as the Advanced Very-High-Resolution Radiometer (AVHRR), the Visible Infrared Imaging Radiometer Suite (VIIRS), the Moderate Resolution Imaging Spectroradiometer (MODIS), Sentinel, Landsat, Satellite Pour l'Observation de la Terre (SPOT), the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), IKONOS, Worldview, RapidEye, Ziyuan 3 and Gaofen and synthetic aperture radar data [10]. Water quality parameters that are directly observable can be monitored using LandSat, Sentinel, MODIS, the MEdium Resolution Imaging Spectrometer (MERIS) and other specialized sensors such as Hyperion (chlorophyll-*a*), SPOT, AVHRR (Total suspended sediments), IKONOS (water clarity) and the Coastal Zone Color Scanner (CZCS) [colored dissolved organic matter] imagery [10]. Hyperspectral imagery [11], LiDAR and airborne laser scanning [12] can be used to infer crop characteristics and tree canopies in agrarian and forested watersheds. Within watersheds, indices developed by using reflectances from different spectral bands are useful for classifying land cover and land use types [13]. Also within watersheds, the state of best management practices such as low impact developments and green infrastructure can be monitored using visible and thermal imagery and spectral reflectance-based indices [14].

Authors have developed an array of methods ranging from the well-established to the experimental frontier have emerged to quantify watershed loading and receiving water quality pertaining to various types of impairment [15,16]. Broadly, for land surface processes such as crop cycles, land cover and land use changes, observations of surface reflectances and ratios of reflectance across multiple spectral bands are used with validation datasets such as land use classification maps or photographs are common (e.g., [17]). For water quality parameters that affect the inherent and apparent optical properties of the water column and spectral properties of the reflectances, such as total suspended solids, Chlorophyll-a, colored dissolved organic carbon, turbidity, and water surface temperature, statistical models linking observations of surface reflectances and ratios of reflectance across multiple spectral bands with the water quality parameters are popular [18]. For such processes, bio-optical and bio-geo-optical models that mechanistically link the irradiancereflectance ratio across multiple spectral bands to the bulk water column properties are a reliable alternative [19]. For impairments such as harmful algal blooms and oxygen depletion, more sophisticated statistical (e.g., [20]) or mechanistic models (e.g., [21]) are used to relate watershed nutrient loads with impairment. For water quality parameters that do not directly affect the optical properties of the water column and spectral properties of the reflectances (e.g., mercury and heavy metals), statistical, machine learning and artificial intelligence models relating the quantities of interest with other water constituents (e.g., turbidity) are becoming increasingly common [15].

Two unifying requirements of all of these approaches are that for a given location, (i) it must be possible to obtain remotely sensed spectral imagery in the first place, and (ii) the land surface process such as fertilizer application, presence of a pollutant load attenuation best management practice like a detention pond or bioswale, and the receiving water quality must all be ground-truthed with in-situ observations [22–24]. This means that the vast potential of remote sensing can only truly be exploited in those areas where atmospheric conditions are generally conducive to remote sensing, and where some access to the watershed is possible. We leverage geospatial datasets containing information on atmospheric conditions and accessibility to rank subwatersheds according to their propensity to being amenable to remote sensing for water quality monitoring under a variety of cost-payoff scenarios.

Reasons for the gap between the number of this assessment and monitored waterbodies and the total number of waterbodies in the United states can include (i) applicability, (ii) accessibility, (iii) resource constrains and (iv) socioeconomics. First, in most cases, waterbodies might be located in remote, locations with minimal human development and limited beneficial uses to society apart from preserving pristine natural beauty and bounty. Second, there may be access issues due to rugged landforms, impassable land use features and limited mobility networks which make monitoring difficult or even impossible. Third, state, tribal and local agencies tasked with assessing and programmatically attaining designated uses for waterbodies may be hampered by resource constraints that make regular ambient monitoring and synoptic data collection difficult or even impossible in all watersheds. Fourth, there may be environmental justice issues with impairment being linked to socioeconomic and demographic conditions (e.g., [25]). In this paper, we only look at the first three factors, as we are only interested in the technical constraints of feasibility of using remote sensing for water quality monitoring and data collection here. The advantage of remote sensing over conventional monitoring programs is that for areas where the former is applicable, ground-truthing need not be performed very often, and in

the overall life cycle of monitoring and data collection programs, can be significantly less expensive to implement, particularly if public domain data and tools are used.

The paper is organized as follows: in Section 2, we describe our classification approach, the datasets we used and the geostatistical methods we adopted. In Section 3, we showcase the remote sensing potential map. In Section 4, we discuss where in the country gaps need to be filled, given the potential for remote sensing by looking at the applicability and accessibility of the subwatersheds and the intersection of quantified risks of impairment, ecological vulnerability and lack of coverage. Finally, also in Section 4, we outline the caveats in our approach, and chart our future trajectory in making these classifications more robust and readily accessible to the watershed management community.
