Integrating stream gage data and Landsat imagery to complete surface water records

Release Date:

This article is part of the Spring 2020 issue of the Earth Science Matters Newsletter

Map of Central Valley California

Figure 1. Map of the expanded Central Valley study area showing the 50 8-digit hydrologic unit codes (HUC) that form the spatial analysis units; major waterways; and the number of stream gages in each HUC with sufficient temporal overlap with the imagery time series. Figure from Figure 1 in (Walker et al. 2020).

(Public domain.)

Rainstorms, snowmelt, and river overflows can cause excess water to pool across a landscape. Knowing where water collects and how long it persists is important for managing flood risk and conserving habitat. Unfortunately, direct knowledge of current and historical surface water dynamics is incomplete due to inherent limitations in overflight timing and satellite image collection.   

To create a more complete record of water extents, USGS scientists explored the integration of stream gage data with surface water measurements derived from Landsat imagery. Stream gages complement the spatial views of satellite imagery by recording volumetric information at discrete locations. Unlike Landsat satellites, which only reimage a location every 16 days, stream gages do not require clear weather to operate and they take more frequent measurements. This study investigated whether a quantitative relationship could be established between data collected simultaneously from Landsat satellites (i.e., water area) and stream gages (i.e., water volume). Such a relationship could help estimate water coverage area when satellites are unable to view land surfaces due to cloud obstruction. This complete record of surface water extents could be connected to changes in land cover and storm events, allowing researchers to study how land development and climate change affected water extents in the past. 

USGS scientists tested this hypothesis in California’s Central Valley (Figure 1), which faces complex water management challenges due to a growing human population, agricultural intensification, and projections of increased flooding. First, researchers extracted monthly water surface area maps, spanning from 1984 to 2015, from two independent Landsat-based datasets: the European Commission’s Joint Research Center (JRC) Monthly Water History and the USGS Dynamic Surface Water Extent (DSWE). These two datasets classify the landscape into different categories of water presence (Figure 2). Researchers then downloaded monthly average discharge data from the USGS National Water Information System for 548 stream gages throughout the study area. After examining the relationship between stream gage and satellite data for 50 Central Valley watersheds, researchers found strong correlations when imagery extent was paired with discharge from any gage. This discovery provided a solid basis for reconstructing water extent values and drove the generation of continuous time-series for 30+ years in 35 watersheds. By demonstrating that this technique provides quantitative estimates of historical surface water extents, scientists can further elucidate phenomena such as flooding or drought events occurring from 1984 onward. The resulting complete water surface record furthers our understanding of how land cover changes and altered climate regimes may affect future water dynamics.

Landsat 8 image classified by two different models

Figure 2. Example classification of surface water by the DSWE and JRC algorithms in a Landsat 8 scene acquired near Stockton, California. The Landsat image is a false-color representation of the landscape; here red depicts growing vegetation and green shades largely depict barren agricultural areas. Water bodies are dark blue. The reduction of the scene to multiple classes of water presence (DSWE) and binary water/not-water presence (JRC) illustrates how sequences of such maps make it easier to track the effects of development or agricultural conversion on patterns of surface water across the landscape. These insights can be applied to anticipated land changes to estimate the future flooding potential. Figure from Figure 2. in (Walker et al. 2020).

(Public domain.)

The paper “Integrating stream gage data and Landsat imagery to complete time-series of surface water extents in Central Valley, California” appeared in the International Journal of Applied Earth Observation and Geoinformation and is available here: https://www.sciencedirect.com/science/article/pii/S0303243419308049.

<< Back to Spring 2020 Newsletter

Related Content