Potential and limitation of Landsat, Sentinel-2 and Planet datasets in monitoring the intermittency regime of non-perennial rivers

One of the major difficulties in studying and protecting non-perennial rivers is the lack of knowledge about the occurrence of dry periods and their duration. Traditional flow measurement systems are not reliable in measuring zero or near-zero flows and do not provide any information on the presence of isolated ponds during periods of no flow. In this context, satellite observations can make a crucial contribution thanks to their global coverage and high observation frequency, especially if freely available data can be exploited. In the field of multispectral satellite data, the free datasets provided by Landsat (USGS/NASA) and Sentinel-2 (ESA) are particularly useful. Another dataset with interesting features is that of PlanetScope, but unfortunately this data is not free, although available on request for research purposes. In this study, we present an analysis of the potential and limitations of these three datasets in observing the intermittency regime of non-perennial rivers. The differences in their spatial, temporal and spectral resolution make them more or less suitable for monitoring specific rivers with given characteristics and observation requirements. It emerged that thanks to a long archive of observations (more than 40 years), Landsat is particularly useful for analyzing changes in the intermittent flow regime over time, enabling the detection of climatic trends over the standard climatological period of 30 years but due to coarse spatial resolution (30 m) it only allows observation of rivers with sufficiently wide active riverbeds (around 90 m or more). Thanks to the finer spatial resolution (10 or 20 m depending on the band), Sentinel-2 allows observation of water features greater than 6-15 m in rivers larger than approximately 30 m for an observation period that currently stands at 9 years. Thanks to the short revisit time of 5 days or less and the free availability of data, this dataset is particularly useful for continuous observations and obtaining the annual intermittency regime in a larger set of rivers of adequate size. PlanetScope provides data with spatial resolution of around 3 m and a revisit time up to 1 day. Although the spatial resolution is significantly higher than that of Sentinel-2, the ability to observe small water surfaces is not improved proportionally. In fact, we have found that this data allows for the observation of water features greater than 4-10 m in rivers larger than approximately 20 m. This is likely due to the different spectral characteristics of the acquired data. Another factor affecting performance is related to the acquisition time, which for Sentinel-2 is the same for all acquisitions of the same scene, while for PlanetScope images it is variable. This leads to inconsistency in the dataset, making it more challenging to identify water surfaces.

For all considered datasets, the “flowing,” “ponding,” and “dry” phases can be distinguished in a supervised manner using false-color images or automatically by exploiting the reflectance characteristics of water. The performances of both supervised and unsupervised classification are analyzed for different datasets and in various case studies.