Detecting Rice Fields using Radar Remote Sensing

INTRODUCTION

A majority of my previous work on Remote Sensing (as of July 2021) has involved processing a single satellite imagery dataset and interpreting its output. On a few occasions—particularly for workflows that required change detection—I have compared outputs derived from processing two or more imagery datasets.

For this post, I have processed 15 Sentinel-1 SAR satellite imagery datasets (Radar/Microwave Remote Sensing) acquired between April and December 2020 to detect the presence of rice fields (paddy farms) in the Mekong River Delta of Vietnam.

Figure 1: Area of Interest – Cross-section of the Mekong River Delta

The Mekong River Delta, also known as Vietnam’s Rice Bowl, produces more than half of the country’s rice. Vietnam was the world’s 5th largest producer and 3rd largest exporter of rice at the time of writing this post (May 2021).

Monitoring rice production—or the production of any other staple food crop—is essential. Not only does paddy cultivation employ a significant number of farmers (over 15 million in Vietnam alone), but information about crop health and growth is also of great importance to stakeholders across the supply chain. Earlier this year, due to a production shortfall, Vietnam was compelled to import rice from its trading rival, India.

---

METHODOLOGY

You may wonder: why are so many imagery datasets needed to detect rice?

Well, consider interpreting this optical satellite imagery:

Figure 2: Bing Virtual Earth Optical Imagery over Southern Vietnam (Mekong River Delta)

Can you detect the rice fields from this aerial view? Would you be able to state with certainty, for instance, that the vegetated areas are all paddy farms while the barren fields are not?

Of course, not. It is not possible to arrive at that conclusion.

What if I were to provide an additional piece of information—that this imagery was acquired during the rice harvesting season?

Would you be able to detect the rice fields now?

I do not think it would still be possible. Even if you were able to differentiate between fields that have already been harvested and those that have not, you would still lack sufficient information to conclusively delineate a rice field from a non-rice field based solely on this evidence.

Now assume you have gained additional insight into the harvesting characteristics of the rice crop from a Vietnamese farmer. If I were to show you multiple aerial photographs acquired on different days within the harvesting season, would you be able to detect the rice fields?

You would fare much better than in your previous attempts.

There are two key points I am attempting to convey here:

• Having knowledge of the crop cycle is crucial for detecting its presence

• Optical imagery—derived from reflected solar radiation—offers useful but insufficient insights for reliably identifying rice fields

Microwave/Radar Remote Sensing has unique characteristics that overcome some of the major limitations of optical remote sensing. The way microwaves interact with surface features on Earth is fundamentally different from how sunlight does—and this distinction is central to our rice field detection workflow. Microwaves are transmitted actively to acquire Synthetic Aperture Radar (SAR) imagery, whereas solar radiation is captured passively to render optical or multispectral imagery.

Figure 3: The 15 Sentinel-1 Synthetic Aperture Radar (SAR) imagery datasets (April–December 2020) used in this workflow

---

Figure 4: Paddy image by PhotographyCourse on Unsplash)

From sowing to harvesting, a rice crop typically takes about four months to mature. Paddy farms are inundated with standing water until 7-10 days prior to harvesting to promote crop growth and suppress weeds. As a result, only during the final stages—when flooding is reduced—does the crop become clearly visible to an observer (or, in our case, to a satellite). This unique crop-cycle characteristic allows rice fields to be distinguished from other farmlands and land-cover types.

Figure 5: Backscatter Depiction

When microwaves emitted by radar satellites strike a flooded rice field (with the crop submerged), the energy reflects away from the satellite due to specular reflection at the water surface.

However, when microwaves interact with the canopy of mature rice crops emerging above the water near harvest time, a significant portion of the energy is reflected back toward the satellite due to diffuse reflection. This crucial difference is illustrated in Figure 5.

When energy waves reflect in the direction of their source, the phenomenon is technically known as backscatter. Significant temporal changes in backscatter near the rice harvesting period are what I monitor to delineate rice fields.

---

Figure 6: Amplitude_VH band of SAR imagery (9 December 2020) over the Mekong River Delta

Figure 6 depicts a raw SAR imagery dataset acquired on 9 December 2020. Dark pixels represent areas from which low backscatter was received—typically flooded regions (note the East Vietnam Sea on the right, which appears jet black). Bright pixels represent areas of high backscatter, generally corresponding to vegetated or urban surfaces. Using optical imagery from Google Earth, urban areas can be ruled out in this scene, as the Mekong River Delta is predominantly agricultural.

ESA's SNAP Software was used for imagery processing and visualization.

---

Figure 7: Processed Radar Imagery Output over Mekong Delta Study Area

Several processing steps—namely Apply Orbit File → Thermal Noise Removal → Calibration → Terrain Correction → Geographic Extent Subset → Band Subset → Stack Generation → Multi-temporal Speckle Filtering → Decibel Transformation—are required to convert raw radar imagery into an analysis-ready product (technicalities elaborated in this document from Page 14 onwards).

From the processed output in Figure 7, observe that:

• The imagery appears more vivid due to decibel transformation of backscatter values

• The geographic extent has been clipped to the Mekong River Delta

• The imagery has been transformed into a map projection, rather than radar geometry

---

Figure 8: Temporal Backscatter Evolution Graph for a single pixel

Because all fifteen input datasets were batch-processed into a single imagery stack, I was able to generate a Backscatter Evolution Profile, which shows how backscatter values for a specific pixel evolved between April and December 2020.

From Figure 8, note the presence of three distinct troughs (low backscatter values), corresponding to imagery acquired on 13 April, 18 July, and 15 November 2020. Each trough is followed by a gradual increase in backscatter.

What does this imply?

In Vietnam, rice is typically sown three times a year—during spring, autumn, and winter. These sowing periods correspond to flooded conditions, hence the low backscatter values. As the rice crop grows and emerges above water, backscatter values increase. This confirms that the land underlying the selected pixel is being used for rice cultivation.

The backscatter evolution profile of rice fields is distinctive. Other land-cover types do not exhibit such pronounced temporal variation. For instance, residential areas show minimal change over time, while non-rice farmland lacks the submerged-to-visible growth cycle characteristic of rice.

Thus, pixels exhibiting large temporal variation in backscatter are likely to correspond to rice fields. Optical imagery and ground truthing can be used to further validate the radar-based results.

I applied this defining characteristic to the entire imagery stack over the Mekong River Delta. SNAP software was used to compute, for each pixel:

a) Maximum backscatter value

b) Minimum backscatter value

c) Backscatter range (maximum − minimum)

These were stored as individual bands and combined into a false-color composite, shown below-

Figure 9: Final false-color composite map of rice field detection over the Mekong River Delta. The violet pixels have the maximum variation in backscatter readings and signal Rice cultivation

The symbology ranges from blue to violet. Pixels in blue exhibit low temporal variation in backscatter, while pixels in violet show high variation and are therefore indicative of rice cultivation.

I hope you enjoyed this study on using remote sensing for rice field detection.

Can you think of other crops that could be detected using radar remote sensing? Feel free to share your thoughts.

---

ABOUT US - OPERATIONS MAPPING SOLUTIONS FOR ORGANIZATIONS

Intelloc Mapping Services, Kolkata | Mapmyops.com offers a suite of Mapping and Analytics solutions that seamlessly integrate with Operations Planning, Design, and Audit workflows. Our capabilities include — but are not limited to — Drone Services, Location Analytics & GIS Applications, Satellite Imagery Analytics, Supply Chain Network Design, Subsurface Mapping and Wastewater Treatment. Projects are executed pan-India, delivering actionable insights and operational efficiency across sectors.

My firm's services can be split into two categories - Geographic Mapping and Operations Mapping. Our range of offerings are listed in the infographic below-

Range of solutions that Intelloc Mapping Services (Mapmyops.com) offers

A majority of our Mapping for Operations-themed workflows (50+) can be accessed from this website's landing page. We respond well to documented queries/requirements. Demonstrations/PoC can be facilitated, on a paid-basis. Looking forward to being of service.

Regards,

Arpit Shah

Credits: RUS Copernicus