Coastal farmlands across US mid-Atlantic turning too salty for traditional farming

In this study, we found that hundreds of hectares of farmland have visible salt patches, and thousands of additional hectares have turned into marsh. The annual loss in profit might be in millions of US dollars, especially in a corn-based economy.
Coastal farmlands across US mid-Atlantic turning too salty for traditional farming

The problem: Saltwater intrusion

Many of the world's productive farmlands are experiencing water scarcity issues due to inadequate rainfall or groundwater depletion. But what about a region where there’s too much water but not the right kind? Let me introduce you to the low-lying coastal farmlands of the Delmarva Peninsula covering coastal counties of the US states of Delaware, Maryland and Virginia.

Delmarva farmlands along low turbulence brackish tidal creeks. Saltwater intrusion is increasingly resulting from frequent far-reaching seasonal high tides. Photo by Jarrod Miller.

Below you see a drone photo of a farm in Maryland. I would like to draw your attention to the white rim along the field edge – the salt patch. These salt patches are appearing on many farms across Delmarva, even in Georgia and the Carolinas. There are several processes, often happening at the same time, which might result in such a visual manifestation of salty soil. These include sea-level rise, off-shore storms, excessive groundwater pumping, and drought among others.

Bright white salt patch along a field edge in Maryland. Photo by Jarrod Miller.

Saltwater intrusion into coastal ground- and surface water is leading to a suite of environmental issues such as ghost forests, crop yield decline, and a rise in invasive species that might be more salt tolerant. The salt patches are of particular concern in a corn-based economy, such as Delmarva, since corn has a lower salinity tolerance compared to other crops including soybeans or wheat. Salinity data at a regular interval can help farm-owners in short- and long-term decision making before it is too late and they have no option other than abandoning their farms altogether. Yet, measuring soil salinity through field- and laboratory-based methods is labor-intensive, time-consuming and expensive.

How we used remote sensing

Salt patches reflect sunlight differently than healthy crop or water. We leveraged this behavior and captured these salt patch signatures using machine learning and remotely sensed aerial and satellite images between 2011 and 2017. While remote sensing has been used before to document salinity, that effort has mostly been focused on capturing reduced crop health resulting from saline soil. The issue with that approach is that it is very challenging to identify the exact reason behind that reduced crop health. Identifying these bright white salt patches is a direct way of mapping salty soil. However, these salt patches are harder to monitor due to their elusive nature and their fine-scale appearances, often ranging from a few to tens of meters.

We used high-resolution (1 m) aerial images from the National Agriculture Imagery Program (NAIP) for 14 coastal counties in 3 US states from 2011-2013 and 2016-2017. It was necessary to use multiple years for any given time-step since NAIP images are not uniformly available for all US states. Since NAIP images are only captured during summer or early fall, initially our machine learning model was missing the seasonal information that was critical to differentiate between different vegetation types, e.g. crops, residential lawn, forest. To fill in this gap, we used satellite data from the Landsat mission for the same years as the NAIP images, but covering all seasons. We visually assessed NAIP images to identify 94,240 locations (known as reference points) for 8 land covers: forest, marsh, salt patch, built, water, farmland, bare soil, and other vegetation. This visual assessment was facilitated by field data collected during 2019 and 2022 when we identified the same land covers on the ground and recorded their locations. These 94,240 points were then split 70:30 for training and testing a machine learning based model, called Random Forest, on the cloud computing platform Google Earth Engine. Utilizing the training points, these models examined each grid cell of the input NAIP and Landsat images, and predicted a land cover for each grid cell. The results are two gridded datasets (Mondal et al. 2022) that we used to calculate the area under salt patches and marsh – another important land cover in the context of saltwater intrusion.

Salt patches and marsh conversion on the rise

We found that the visible salt patches have almost doubled, from 472 to 905 ha between 2011 and 2017. These patches represent a complete loss of productive farmland as almost no crop grows there. These numbers might look small at a first glance. But doesn't the mere presence of these patches along the farm fringes denote that the entire field might be too salty for traditional crops? In fact, we recorded high soil salinity values hundreds of meters away from these visible salt patches.

Between 2011 and 2017, an additional 8,000 ha of farmland converted to marsh. Moreover, in some instances, salinity in the salt-impacted farmlands might reduce crop yield, but may not be high enough to leave a visible patch or a land cover conversion (agriculture to marsh). To grasp the true extent of this problem, we calculated farmland area within 50, 100 and 200 m of the salt patches. In 2016-2017, about 167,000 ha of farmland were within 200 m of a visible salt patch.

Over 8,000 ha of farmlands converted to marsh between 2011 and 2017.

What does it mean for farm productivity?

An increasing level of soil salinity does not bode well for a corn-soy economy. Corn has a low tolerance for salinity (up to 1.7 mS/cm of electrical conductivity (EC) value; a measure of soil salinity), while soybean's tolerance is a bit higher at 5 mS/cm. We recorded >5 mS/cm of EC values hundreds of meters away from visible salt patches. Since the same level of salinity would affect corn and soybeans differently, it is challenging to predict the exact productivity loss. To address this challenge, we provided profit loss estimates for a range of scenarios. These scenarios assume a complete loss in profit, rather than a complete loss in crop yield. In a business-as-usual scenario where profits are derived from both corn and soybeans, we estimated annual profit loss of US$70.7 million for all at-risk farmlands within 200 m of salt patches in 2016-2017. For the same extent of farmlands in 2016-2017, but considering 100% soybeans and 100% corn scenarios, annual profit losses are estimated at US$39.4 million and US$107.5 million, respectively. Since corn is more profitable, the loss in profit is also greater than those for business-as-usual or 100% soybeans scenarios.

What can we do?

The good news is that we have options. But we need to act now. Reducing farm inputs, adding gypsum to the salt-impacted lands, or using crop insurance might work in the short term, until the landowners are forced to abandon the farms. Strategically increasing the share of farmlands under more salt-tolerant crops, such as soybeans, sorghum, barley, might be a more long-term solution. Controlled conversion of these landscapes into marsh is another proactive step to protect the coastal lands against increasing soil salinization. Our high-resolution datasets are better suited, compared to global datasets with coarser resolution, for designing such farm-level interventions. We shared our datasets through a public repository and a web-based/mobile app so that those can be used for crafting new incentive programs for the landowners who need them the most in this rapidly changing coastal landscape.


Mondal, P., Walter, M., Miller, J., Epanchin-Niell, R., Yawatkar, V., Nguyen, E., Gedan, K., & Tully, K. High-resolution remotely sensed datasets for saltwater intrusion across the Delmarva Peninsula [Data set]. Zenodo. (2022).

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in