**3. Results**

#### *3.1. Surface Water Area Validation Using SkySat Imagery*

The surface area obtained from PlanetScope, Basemap, and Planet Fusion showed great agreement (r2 ≥ 0.98) with the validation dataset. In addition, PlanetScope had the smallest MAPE (8.09%), followed by Basemap (8.21%) and Planet Fusion (9.17%) (Figure 3). When splitting the validation surface area observations into different size classes (Table 4), all three image sources presented similar agreement (r2 ≥ 0.87), and the highest r2 values were found for surface area observations between 10 and 30 ha (r2 ≥ 0.95). All three sources had a similar MAPE for observations between 0.1 and 5 ha (~7.55%) and between 10 and 30 ha (~7.98%), while the highest values were found for observations between 5 and 10 ha (~10.27%).

**Figure 3.** Pairwise comparisons between the SkySat validation surface area and the surface area obtained from PlanetScope, Basemap, and Planet Fusion for multiple observations in time.

**Table 4.** Pairwise comparisons between the SkySat validation surface area and the surface area obtained from PlanetScope, Basemap, and Planet Fusion for multiple observations in time and divided into three size classes.


#### *3.2. Number of Surface Area Observations per Dataset*

The number of PlanetScope observations varied throughout the year and varied across different OFRs (Figure 4). The months with the highest number of PlanetScope images were November–December 2020 (~17) and March–April 2021 (~14), while the months with the lowest numbers were July–September 2020 (<10) and February 2021 (<3) (Figure 4A). In addition, most of the OFRs (~60%) had 80–100 PlanetScope observations per year (Figure 4B). Basemap images were processed at a daily cadence, and we considered a new Basemap observation every time a new image composite was used. In this regard, the number of Basemap images followed a similar pattern found for PlanetScope; however, the mean number of Basemap observations per month was higher than of PlanetScope in 10 out of the 12 months analyzed, and most of the OFRs (~75%) had 90–120 Basemap observations per year (Figure 4A,B).

**Figure 4.** Number of surface area observations per month for PlanetScope and Basemap (**A**) and for Planet Fusion real, mixed, and synthetic (**C**). Frequency distribution of the total number of observations per OFR per year for PlanetScope and Basemap (**B**) and for Planet Fusion real, mixed, and synthetic (**D**).

Planet Fusion images were derived from real and synthetic pixel values, and the number of real and synthetic observations varied throughout the year (Figure 4C). The number of images derived from real pixels reached its peak (~13–15) between November and December 2020, and the lowest numbers were found in February 2021 (~2) and between May and June 2021 (<5). In general, most of the OFRs had ~ 80–100 real observations per year. The number of mixed images (i.e., composed by real and synthetic pixels) tended to be < 10 for all months, and most of the OFRs had <50 mixed observations per year (Figure 4C,D). The number of synthetic images is higher than real and mixed observations for all months of the year, and the highest values (~22–26) occurred in July 2021 and May–June 2020, with the lowest values between November and December 2020 (~14–16) (Figure 4C). In addition, most of the OFRs had ~250–260 synthetic observations per year (Figure 4D).

### *3.3. Planet Fusion Comparison with PlanetScope*

We found a high agreement (r<sup>2</sup> ≥ 0.90) for the same-day surface area pairwise comparisons between Planet Fusion and PlanetScope for all size classes (Figure 5). MAPE decreased as observations increased in size, and the highest MAPE values were found for synthetic, mixed, and real for all size classes (Figure 5). The number of pairwise comparisons for real was higher than mixed and synthetic, to a large extent (~60%). This finding is somewhat expected, as the Planet Fusion algorithm uses PlanetScope images as an input to generate daily Planet Fusion imagery.

**Figure 5.** Same-day pairwise comparisons between PlanetScope and Planet Fusion real, mixed, and synthetic for multiple observations in time and for all OFRs divided into three size classes (0.1–5 ha, 5–10 ha, and 10–30 ha). Brighter colors indicate higher point density.

#### *3.4. Monthly Comparisons between Basemap and Planet Fusion with PlanetScope*

When comparing each OFR surface area time series derived from Basemap and Planet Fusion with PlanetScope, for both datasets, most of the OFRs (63% and 61% for Basemap and Planet Fusion, respectively) showed good agreement with r2 ≥ 0.55, and 74% and 70% of the OFRs presented small uncertainties with MAPE <5% (Figure 6).

**Figure 6.** Frequency distribution of r2 and MAPE calculated by comparing the OFR time series from Basemap and Planet Fusion with PlanetScope.

The mean monthly percent error—calculated by comparing Basemap and Planet Fusion with PlanetScope—for Basemap and Planet Fusion varied between −2.45–1.48% and between −3.36%–1.66% for 0.1–5 ha, between −2.88–1.11% and between −3.56–0.51% for 5–10 ha, and between −2.23–0.53% and −3.13–0.76% for 10–30 ha. These values were stable throughout the year (Figure 7). The percent error variability decreased as the surface area observations increased in size, and the observations between 10 and 30 ha had the least variability. In addition, Planet Fusion presented smaller percent error variability when compared to Basemap for all size classes (Figure 7). The highest MAPE values for Basemap (4.73%) and Planet Fusion (5.80%) were found for observations between 0.1 and 5 ha, and the MAPE was <4.40% for all months for both Basemap and Planet Fusion for observations between 5 and 10 ha and 10 and 30 ha, respectively. This indicates that even when there are fewer PlanetScope images available to generate Basemap and Planet Fusion due to clouds, shadow, and haze, both products tend to have surface area uncertainties <5%.

**Figure 7.** Monthly percent error variability and MAPE calculated from the same-day pairwise comparisons between Basemap and Planet Fusion with PlanetScope for the three size classes (0.1–5 ha, 5–10 ha, and 10–30 ha).

#### *3.5. OFR Surface Area Time Series*

We selected six OFRs (Table 2) to illustrate the surface area time series derived from PlanetScope, Basemap, and Planet Fusion (Figure 8). The surface area time series show that different OFRs have different surface area change patterns. In general, the OFR surface area decreased between 20 July and 20 November (e.g., Figure 8, OFRs A–D), period of the year when farmers are irrigating their crops [23], and it increased between 21 January and 21 May, which are the months when the study region receives most of its annual precipitation [23].

When compared to PlanetScope and Basemap, Planet Fusion had a smoother surface area time series with less variability (e.g., Figure 8 OFRs A–D). In addition, the Planet Fusion time series was less affected by the presence of clouds and haze, which can increase or decrease OFR surface area. Even though we used a low cloud-cover threshold (<10%) for PlanetScope, there are several PlanetScope and Basemap images contaminated with cloud shadows and haze (e.g., Figure 8, OFR A, between 20 August and 20 September), indicating surface area ~20% larger than that of Planet Fusion. Other examples were observed between 20 July and 20 August and between 21 May and 21 June in Figure 8, OFR B, in which there were no PlanetScope images available and the Basemap shows abrupt changes in surface area—a drop of 20% and 15% for both dates—which were caused by the presence of cloud shadows and haze. In Figure 9, we highlighted the impact of clouds and haze for OFR A (16 August 2020) and OFR B (30 August 2020). For OFR A, PlanetScope and Basemap surface areas were ~20% larger than those of Planet Fusion, which is explained by the misclassification of water on the lower-right corner of the OFR. For OFR B, while the PlanetScope image had a surface area ~13% larger than that of Planet Fusion, the Basemap image indicated a surface area ~14% smaller. These discrepancies are caused by the presence of clouds in the PlanetScope image and haze in the Basemap image.

#### **Figure 8.** *Cont*.

**Figure 8.** OFRs (see Table 2) surface-area time series derived from PlanetScope, Basemap, and Planet Fusion and OFR shapefiles overlaid on high-resolution Google Satellite imagery.

**Figure 9.** OFRs A and B (see Table 2) PlanetScope, Basemap, and Planet Fusion false-color composites (blue: red, green: green, and red: NIR) and the surface-water classification for 16 August 2020 (OFR A) and 30 August 2020 (OFR B).

OFR surface water classification is impacted by the OFRs' environmental conditions and their shape geometry. OFRs with complex geometries (e.g., not circular or square and shapes with a large number of edges) tend to have higher surface area classification uncertainties [33]. For example, Figure 8, OFR D, shows a multi-part OFR that may not have all parts inundated at the same time, which can explain part of the variability in the surface area time series for PlanetScope, Basemap, and Planet Fusion. The surface area time series from OFR E and OFR F (Figure 8) are influenced by the presence of vegetation within the OFRs. The presence of vegetation impacts surface water classification [5,33,55], leading to noisy surface area time series and abrupt changes (e.g., OFR E between 20 September and 21 January). In addition, the high variability in surface area for OFRs E and F is related to the presence of adjacent water bodies, which can inundate during flood events and contribute to changes on OFR boundary limits. We highlighted the impact of vegetation on the OFR E time series for two different occasions: 14 July 2020 and 16 October 2020 (Figure 10). During the first occasion, PlanetScope and Basemap indicated surface area (~9.5 ha) 95% greater than Planet Fusion (0.5 ha); on the second occasion, a contrasting scenario in which Planet Fusion surface area (12.25 ha) was 86% higher than that of PlanetScope and Basemap (~2 ha). These results shed light on the importance of assessing the OFR environmental conditions and how they impact the OFR surface area time series before employing these datasets to monitor surface area changes.

**Figure 10.** OFR E (see Table 2) PlanetScope, Basemap, and Planet Fusion false-color composites (blue: red, green: green, and red: NIR) and the surface water classification for 14 July 2020 and 16 October 2020.

#### **4. Discussion**

The surface area validation carried out using multiple SkySat imagery showed that the methodology used to classify OFR surface area performed well for PlanetScope, Basemap, and Planet Fusion, with high agreement r<sup>2</sup> ≥ 0.87 and MAPE between 7.05% and 10.08% for all image sources and all size classes (Table 4). Comparisons between Basemap and Planet Fusion with PlanetScope highlighted that most of the OFRs had good agreement with 61% of the OFRs with r2 ≥ 0.55, and small uncertainties with 70% of the OFRs with MAPE < 5% (Figure 6). Basemap and Planet Fusion presented similar monthly mean percent error (~−3–3%) and MAPE (~2.20–5.80%) throughout the year (Figure 7). In addition, percent error variability and MAPE decreased for the larger surface area observations (Figure 7). The highest monthly MAPE (5.80%) was found for Planet Fusion for observations between 0.1 and 5 ha, and the MAPE was ≤4.40% for Basemap and Planet Fusion for observations between 5 and 10 ha and between 10 and 30 ha. Furthermore, when analyzing the three Planet Fusion data categories (i.e., real, mixed, and synthetic), the greatest uncertainties were found for the synthetic images (MAPE ~ 5%), followed by mixed (MAPE ~ 4%) and real (MAPE ~ 3%) (Figure 5). These findings indicate that Basemap and Planet Fusion

images can be employed to monitor OFRs with uncertainties < 10% when the sources are compared to the validation dataset and with uncertainties < 5% when compared to PlanetScope. However, time series obtained from Basemap and Planet Fusion can be highly variable (Figure 8E,F), as surface water classification can be impacted by the size of water bodies (Table 4, Figures 2, 4 and 6) and the environment in which OFRs are located (e.g., presence of vegetation within the OFRs; Figure 8, OFRs D–F).

The number of cloud-free observations offered by Basemap and Planet Fusion enlightens the potential of these datasets to monitor OFR surface area changes (Figure 4). Both datasets pose advantages when compared to a single sensor approach—employing PlanetScope alone (Figure 4), or other sensors, for example, Landsat [23,31,56], Sentinel 1 [57], and Sentinel-2 [58–60]—or a multi-sensor approach [4,33,34]. Briefly, the use of a single sensor is limited to a few observations a month, and in some periods of the year in eastern Arkansas, there could be weeks without a cloud-free image [33]. Although the number of observations is improved when employing a multi-sensor approach, daily to sub-weekly monitoring is not attainable unless an assimilation algorithm [33] is implemented. In addition, when implementing a multi-sensor approach, it is necessary to acquire the data from different platforms (e.g., Planet Explore, Sentinel Hub, and Google Earth Engine), which can be time-consuming and a limiting factor if it is necessary to process, download, and move the satellite imagery across multiple platforms. In this study, we demonstrated that Basemap and Planet Fusion imagery processing can be done entirely in the cloud environment by leveraging the integration of Planet's Platform, Google Cloud Storage, and Google Earth Engine. This integration allows for swift analysis, and it can be used for other study regions without the need to acquire data from multiple platforms.

Daily OFR surface area time series derived from Basemap and Planet Fusion revealed important differences between the two datasets. In general, Basemap had higher surface area variability, and it was more susceptible to the presence of cloud shadows and haze when compared to Planet Fusion, which had a smoother time series with less variability and fewer abrupt changes throughout the year (Figure 8). The Planet Fusion algorithm combines data from multiple satellites to establish a baseline of OFR surface area time series by filling gaps with synthetic pixels. Nonetheless, the smoothing effect should be interpreted cautiously, as some changes in the time series due to large rainfall events or frequent irrigation activities may be smoothed out. This is especially relevant for the periods of the year when there are more synthetic observations (Figure 4) and the uncertainties in surface area tend to be higher (Figure 5). Additionally, because Planet Fusion is based on a robust algorithm that uses data from various satellites, this dataset requires more image processing steps and higher computing power when compared to Basemap, which is generated at a faster speed with lower processing costs. Meanwhile, the Basemap time series may contain a "stair-step" effect caused by repeated observations when the Basemap scene composition was kept constant due to a lack of new cloud-free scenes (e.g., Figure 8, OFR D, early March 2021). By keeping the same image composition, the Basemap algorithm avoids generating synthetic pixel values while still providing a cloud-free observation. Nonetheless, it is important to keep in mind that there could be scenarios (e.g., when there is a lack of a new cloud-free scene for weeks or more) in which the Basemap may have the same number of observations as PlanetScope, hence decreasing its monitoring capabilities.

Our findings have important implications to future hydrological studies that aim to monitor small water bodies at large scale and high temporal frequency. For the OFRs in eastern Arkansas, the Basemap and Planet Fusion surface area time series helped unravel sub-weekly changes in OFR surface area, as well as yearly seasonality (Figure 8). OFRs surface area changes are pivotal information for calculation of OFR water volume inflows and outflows. This can be achieved by combining the surface area time series with the area-volume equations (e.g., hypsometry), which are derived using the OFRs' geometric shape and depth [56,61,62]. Estimating OFRs volume change helps bridge one of the key limitations when modeling the cumulative impacts of OFRs on surface

hydrology, as OFR water volume change is commonly assumed to be equal to all OFRs located in a watershed [24,25,63]. In addition, as the number of OFRs is projected to increase globally [24,27], understanding the impact of OFRs on surface hydrology is pivotal when seeking indicators to determine the optimal spatial distribution and number of OFRs, as well as their storage capacities and water management plans aiming to mitigate downstream impacts. Beyond implications to hydrological studies, we demonstrated that Basemap and Planet Fusion can be used to monitor surface area changes for a network of OFRs (Figure 8). This information can be used by regulatory agencies to create water status reports to improve regional water management and water use efficiency. These reports would be especially relevant during the dry critical period of the year when farmers are frequently irrigating.

#### **5. Known Issues and Limitations**

We applied Basemap and Planet Fusion imagery for a one-year analysis. More research is necessary to assess the performance of these datasets for a longer study period (e.g., including periods of prolonged droughts ~3–5 years) and in other study regions—for example, in Southern India, where OFRs are common [4], and where there is a monsoon climate in which there could be weeks without a clear-sky image [64]. In addition, the validation of this study was conducted using cloud-free SkySat imagery; therefore, there is still a need to further evaluate the performance of both datasets under cloudy conditions. However, this will require extensive field work, including visiting multiple OFRs on cloudy days, which imposes several challenges, as most of the OFRs in eastern Arkansas are located on private properties. Furthermore, we assumed that OFR surface area would vary within known and limited boundaries (i.e., OFR shapefile buffered to 20 m). However, different results might be obtained if the Basemap and Planet Fusion images are used to monitor water bodies that frequently change their boundaries—water impoundments that are located close to water streams and rivers that flood frequently, impacting the edges of water bodies. Lastly, although we calculated the uncertainties introduced by Basemap and Planet Fusion, when using these datasets for monitoring purposes, it would be helpful to have an estimated uncertainty accompanying every surface area observation. For instance, whenever there are repeated observations by the Basemap or continuous synthetic observations from Planet Fusion, the uncertainties from these images will be higher; however, as of now, we cannot estimate an observation based uncertainty.

#### **6. Conclusions**

We presented a novel application of Basemap and Planet Fusion analysis ready datasets to monitor sub-weekly OFRs surface area changes. We tested both datasets to monitor 340 OFRs of different sizes, and we found that these datasets can be employed to monitor OFRs with uncertainties < 10% when compared to an independent validation dataset and with uncertainties < 5% when compared to PlanetScope imagery. While Basemap had higher surface area variability and it was more susceptible to the presence of cloud shadows and haze, Planet Fusion had a smoother time series with less variability and fewer abrupt changes throughout the year. Given that the surface area classification can be impacted by the OFR environmental conditions (e.g., presence of vegetation inside the OFR), therefore limiting the use of these datasets, we recommend assessing the OFRs' surface area time series before employing them for monitoring purposes. As the number of OFRs is expected to increase globally, the use of these datasets is of great importance to understanding OFR sub-weekly, seasonal and inter-annual surface area changes, and to improving freshwater management by allowing better assessment and management of OFRs.

**Author Contributions:** Conceptualization, V.P., S.R., J.K., T.H. and M.G.T.; methodology, V.P., S.R. and J.K.; software, V.P., S.R. and J.K.; validation, V.P.; formal analysis, V.P.; investigation, V.P.; resources, V.P., S.R. and J.K.; data curation, V.P., S.R., J.K., N.S., T.B., M.R. and M.A.Y.; writing—V.P.; writing—review and editing, V.P., S.R., J.K., T.H., N.S., T.B., M.R. and M.A.Y.; visualization, V.P.; supervision, T.H.; project administration, T.H.; funding acquisition, V.P., T.H. and M.G.T. All authors have read and agreed to the published version of the manuscript.

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

**Acknowledgments:** The first author is supported by NASA through the Future Investigators in NASA Earth and Space Science and Technology fellowship. The first author acknowledges the support provided by Planet Labs throughout his internship at the company, which included access to SkySat and PlanetScope imagery and on-demand generation of Basemap and Planet Fusion specifically for this study.

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