Data In Focus: A New Way of Looking at Water

As the world’s population approaches 10 billion people and chronic droughts and water shortages proliferate, water has become our most critical and contested resource. Although technological advances have helped alleviate the threat of shortages in oil and other commodities, a global water crisis could be disastrous. Already, 80 countries face water shortages, two billion people lack access to clean drinking water, and another billion cannot meet their most basic water needs, according to the World Bank. In developing countries, population growth is projected to drive demand 40 percent above sustainable levels by 2030.

Figure1

Beyond praying for rain, what can we do? Can technology help us measure, plan, and more effectively allocate precious water resources where they are needed most — especially for fresh drinking water and food production — before it’s too late? The short answer, we have discovered, is yes. Applying cutting-edge data science and deep learning technology to analyze the increasing amount of satellite imagery, we have developed potentially transformative technology to monitor surface water levels and fluctuations around the world.

Such advances can help communities, planners, and policymakers make more informed and sustainable decisions about how to use and save water resources, especially as population grows and water supplies shrink. Consider what’s at stake when it comes to food. Agriculture, which already accounts for 70 percent of global water consumption, will only intensify water demands as it strives to feed several billion more people. Water shortages could spell food shortages — a loss of up to 350 million metric tons per year of crops (more than the entire US output), the International Food Policy Research Institute (IFPRI) estimates.

Water Intelligence: Using Landsat 8 to Monitor Supply

To prepare for water crises, we need rich, accurate data. Using an innovative blend of satellite imagery, deep learning, and data science, Orbital Insight has developed a method for monitoring fresh-water supplies at local and global levels. Our goal is to produce a fully automated water monitor that yields weekly updates and provides statistically significant measures of changes in water surface area. We use freely available USGS Landsat data and leverage the latest developments in deep learning and cloud computing to deliver our signal with minimal overhead. With this technology, we can observe surface water fluctuations at county, state, national and global levels — providing planners and policymakers with more timely, comprehensive information.

California, which has suffered from an intense drought since 2011, offers a potent local case in point. To enrich public understanding about these changing water levels, we applied our technology to the state’s reservoirs.

Figure 1 San Luis Reservoir. Left: Raw image. Middle: Landsat 8 water mask. Right: Orbital insight water mask.

Figure 1 San Luis Reservoir. Left: Raw image. Middle: Landsat 8 water mask. Right: Orbital insight water mask.

In Figure 1, you can see the raw image of the San Luis Reservoir with the Landsat 8 and Orbital Insight’s water masks. To produce the middle image with the Landsat 8 water mask, we utilize information captured in the quality assurance (QA) band. The image on the right shows Orbital Insight’s water mask, calculated using a convolutional neural network approach; this technique effectively filters out the shadows that are mistaken for water in the QA band of Landsat 8.

Figure 2 Comparison of water amount in San Luis reservoir using different methods. Storage and height data courtesy: http://cdec.water.ca.gov/

Figure 2 Comparison of water amount in San Luis reservoir using different methods. Storage and height data courtesy: http://cdec.water.ca.gov/

The simplest next step is to calculate the water surface area by adding up the number of water pixels and scaling them by the area per pixel. For Landsat 8’s panchromatic band, the pixel size is 15 m, the same as the ground sample distance (GSD). Using this approach, we can compare the normalized value of the water surface area in the reservoir, determined through remote sensing, with the water level measured by local in-situ sensors.

As one might expect, we found an annual pattern in which the amount of water dips to its minimum at the end of summer and through the fall, and increases again through the winter due to precipitation. These levels continue to rise in spring as the snowpack thaws. By analyzing the depth of those dips over the years, we can illustrate the drought’s intensity since 2011, at least for that one reservoir. To produce a more comprehensive analysis of the drought, we must measure and aggregate water levels in many reservoirs throughout the state.

Figure 3 Processed data from Landsat QA band showing blue water mask for Lake Haditha in Iraq which upon visual inspection is inaccurate. Middle: Right: Landsat 8 data processed using Orbital Insight algorithms showing accurate water mask.

Figure 3 Processed data from Landsat QA band showing blue water mask for Lake Haditha in Iraq which upon visual inspection is inaccurate. Middle: Right: Landsat 8 data processed using Orbital Insight algorithms showing accurate water mask.

Moving our lens abroad, Figure 3 demonstrates our capabilities on Lake Haditha in Iraq. Most of the rice and wheat grown in northern Iraq is irrigated using water from Lakes Haditha, Mosul, and Habbaniyah. Water shortages in these reservoirs directly affect the country’s grain production through the amount of cropland that is planted. As Figures 3 and 4 illustrate, Orbital Insight’s approach produces more accurate data on water levels, enabling better planning and management of resources. For instance, the water masks derived from Landsat 8’s QA band can significantly underestimate the amount of water available (see figure 4). Such inaccuracies result in a very noisy, high-variance time series, impeding the ability to track trends. By comparison, Orbital Insight’s more accurate and comprehensive water masks lead to much smoother time series.

Expanding on our water monitoring innovations, Orbital Insight plans to leverage the availability of multispectral and hyperspectral satellite and drone imagery to examine pollutants and contaminants in lakes, rivers, and other bodies of water. By combining this imagery with deep learning methods such as convolutional neural networks, we can quantify threats to water quality. Using our expertise in data science, we can detect significant changes as they occur, and signal decision makers when quality threatens to drop below agreed-upon thresholds.

Whether in California or Iraq, this technique has the power to globally impact issues of fresh water availability, water security, and geopolitical risk through persistent coverage and wide reach. It can therefore play a defining role in evaluating and monitoring Integrated Water Resources Management practices across the world. By leveraging remote sensing capabilities, Orbital Insight’s technology enables a standardized global approach that can access hard to reach locations around the world that frequently endure more water stress than the developed world.

Figure 4 Time series of water surface area of Lake Haditha using Landsat 8 water mask (green), and Orbital Insight’s water mask (Blue)

Figure 4 Time series of water surface area of Lake Haditha using Landsat 8 water mask (green), and Orbital Insight’s water mask (Blue)

A Geopolitical Tool

As water shortages threaten to exacerbate conflicts in water-stressed areas, these new data-gathering tools could even play a useful role in geopolitics. Research tells us that shortages will be most poignant in regions that are already battling over water — and states will increasingly come into conflict over water access. Representing 60 percent of the world’s surface water, 263 rivers around the globe cross international borders, and 158 have no laws regulating how these resources get shared. Many of these potentially contentious rivers run through regions that suffer water scarcity and have either a history of conflict or a powerful military, such as Egypt (Nile), China (Mekong), India/Pakistan (Indus), Iraq/Syria/Turkey (Tigris and Euphrates) and Israel/Jordan (Jordan). Although widespread water wars have not broken out in most of these regions, some analysts believe that water issues have played a major role in the Syrian Civil War and, as water security worsens, will drive future conflict. With improved water data, policy makers can engage in more informed dialogue and negotiations that may help prevent future water-driven conflicts.

Meeting our Challenges with Better Data

To create a sustainable water future, we must understand the current water situation around the world. Remote sensing combined with advances in machine learning enables us to develop this understanding in a global context. Current efforts to monitor fresh-water levels rely on individual measurements or monitoring campaigns for specific lakes, reservoirs or rivers by different groups at different times, with varying metrics and accuracies. Likewise, satellite campaigns to evaluate water levels currently suffer from poor accuracy (e.g. Landsat 8’s own Quality Assurance (QA) band). Orbital Insight’s innovative approach provides both superior accuracy and a more comprehensive analysis, in terms of time and space. By measuring bodies of water consistently over long periods of time, we can produce standardized and data-rich knowledge about the state of water in the U.S. and other countries. As water shortages and contentions intensify around the world, our technology can help planners, policymakers, and the public to better understand and address these challenges.

Back to Blog