Monitoring Drought at Indirasagar Reservoir in India using Multispectral Remote Sensing
- Arpit Shah
- Nov 18, 2022
- 10 min read
Updated: 3 days ago
INTRODUCTION
As someone who has lived most of my life in the tropical wetland that is Kolkata, I’ve always found the city to be remarkably water-positive. The south-westerlies bring generous rain, the murky Hooghly river rarely looks anything but swollen, vegetation is lush throughout the year, and the soft soil always seems to carry a distinct watery scent. Yet beneath this apparent abundance lies a worrying reality: Kolkata’s groundwater levels are receding.
HYPERLINKS TO SECTIONS
Credits: RUS Copernicus for the tutorial and ArcGIS Pro for the software trial.
BACKGROUND
For all the devastation it brings—drought is the second-costliest weather disaster after cyclones—the U.S. Geological Survey defines it rather plainly: a period of drier-than-normal conditions that results in water-related problems. Much of it is driven by natural causes, such as inadequate precipitation, but human-induced climate change is undeniably making it worse.
Drought, much like volcanic deformation (another topic I studied earlier), evolves gradually. To understand it better, droughts are commonly categorized as:
Meteorological Drought – when precipitation stays below average for an extended period.
Hydrological Drought – when reservoirs, lakes, rivers, or aquifers fall below critical thresholds.
Agricultural/Ecological Drought – when low precipitation or excessive heat reduces soil moisture available for crops.
The impacts are wide-ranging: crop failure, famine, migration, soil erosion, wildfires, biodiversity loss, water pollution, scarcity for urban and agricultural consumption, and reduced hydropower generation. There is even a Socio-economic Drought, which occurs when water scarcity disrupts supply chains for water-dependent products such as grains and vegetables.
Does a drought end when it rains?
Not quite. The rains must persist long enough to replenish soil moisture and groundwater, allowing vegetation to recover.
Developing and underdeveloped economies face the highest vulnerability because of their dependence on agriculture (UNCCD 2022 Drought Report). India has witnessed alarming levels of drought: by 2019, nearly 50% of its land area was affected (30% as of 2023). Between 2016–17 and 2021–22, droughts damaged 35 million hectares of farmland, roughly 22% of the country’s arable land (2011 data).
Yet there are reasons for optimism. For instance, the western state of Gujarat has implemented extensive drought-monitoring and mitigation strategies—a potential model for other states. Globally, early warning systems, responsible irrigation, rainwater harvesting, heat-resistant crops, sustainable farming practices, and even cloud-seeding are gaining attention.
DROUGHT MONITORING USING REMOTE SENSING
In this study, I use Sentinel-2 multispectral satellite imagery to monitor hydrological drought conditions at Indirasagar Reservoir, Madhya Pradesh, India, over four years (2019–2022). Indirasagar is India’s largest reservoir by area and storage capacity, supplying irrigation water to a state experiencing rising mean annual temperatures and growing drought risk.
The following video walkthrough demonstrates the entire imagery-processing workflow—from downloading datasets to analysing them, using multiple geospatial platforms-
Before analysing Indirasagar, I also studied the Theewaterskloof Dam in South Africa for the same period. This serves as an extension to an online tutorial which monitored drought at that site until 2020. Theewaterskloof, Cape Town’s largest water supplier, suffered severe drought between 2016–18, with water levels dropping by up to 89%.
a. TIMELINE OF IMAGERY USED
For Indirasagar, I selected two Sentinel-2 images per year—one in summer and one during the monsoon.
For Theewaterskloof (southern hemisphere), I selected three images per year—two summer, one winter (drought can persist despite seasonal rains).
To maintain consistency, I chose imagery from similar date ranges across years to avoid seasonal biases. For example, May 28 is typically much drier than May 2 in India.
Equally critical is cloud cover—even small cloud patches directly over the reservoir can disrupt reflectance analysis. For Indirasagar, most selected images had <1% cloud cover (I had to be mindful of this and had to move to selecting another date if there were no Cloud-free images in the desired date-range).
Note: One must also consider that Sentinel-2 images have large spatial coverage. If a study area overlaps multiple tiles, they must be mosaicked; if it fits within one tile, clipping (subsetting) helps reduce dataset size (Sentinel-2 L2A dataset is ~1 GB per image) and quickens processing.
The polygon on the left image in Figure 6 above depicts the geographic extent of a Sentinel-2 multispectral dataset. The water body - Indirasagar Reservoir - lies completely within the extent. Also, notice the clouds to the south-west of the image on the right. While the cloud cover percentage of this product is miniscule (<1 % as mentioned in the product page), had those chunk of clouds been directly over the reservoir, I would have been hesitant to select this imagery dataset for the study.
By contrast, this product (Figure 7 below) is not useful at all due to excessive cloud cover-
I was fortunate to have been able to utilize Sentinel-2 Level-2A (L2A) products for this study - these are available since December 2018 for this region, and contain atmospherically corrected Bottom-of-Atmosphere reflectance which make it ideal for drought studies. Otherwise, one must convert L1C products which carry Top of Atmosphere (ToA) radiance to L2A using Sen2Cor, a computationally heavy process.
b. PROCESSING TOOLS UTILIZED
I have used three geospatial applications for this study:
SNAP for radiometric analysis and water-pixel extraction
QGIS for vectorization and layer cleaning
ArcGIS Pro for refinement, mapping, analysis, and visualization
c. PROCESSING THE IMAGERY ON SNAP
SNAP (Sentinel Application Platform), is a powerful open-source tool for satellite-imagery analytics. Let me begin by highlighting its processing steps-
At first, I will visualize the Sentinel-2 L2A product in RGB mode (natural color)-
L2A products capture data across 12 spectral bands (ranges of wavelength) as can be seen in Figure 11. (L1C product has an additional band - B10).
L2A product has atmospherical corrections data stored in separate masks - such as the cloud pixels in the bottom two images of Figure 12 below-
The raw L2A dataset undergoes three processing phases—Refining, Analysing, and Extracting—for the purposes of this Drought Mapping study.
In Phase 1 (Refining), the Sentinel-2 L2A imagery is first resampled so that all spectral bands share the same 10 m spatial resolution. Several SNAP tools require uniform resolution across bands; a multi-sized product cannot be processed consistently. Thereafter, I subset the imagery—removing unnecessary geographic extent—making the dataset lighter and faster to process.
Phase 2 (Analysis) involves manipulating the resampled spectral bands to isolate water pixels. Since the objective is to monitor hydrological drought, I focus on how the reservoir’s area changes across time. A significant contraction in water extent is indicative of drought conditions.
I used four widely known water radiometric indices:
Each index has strengths and weaknesses depending on the water body and its surrounding environment (urban vs rural, vegetated vs sparsely vegetated, etc.).
Inputting these indices in SNAP is straightforward—they function like arithmetic formulas.For example:
NDWI = (B3 − B8) ÷ (B3 + B8) where:
B3 (Green) = central wavelength of 560 nm
B8 (Near Infrared) = central wavelength of 842 nm
"What is the logic involved?", you may wonder.
As you'll infer from Figure 14 above, Water reflects very little visible light (~10%) and absorbs all NIR wavelengths, whereas soil and vegetation exhibit strong NIR reflectance. NDWI magnifies this contrast, making water pixels stand out clearly.
Earth-observation satellites acquire multispectral images passively, recording solar radiation reflected from the surface. Since sunlight spans Visible, Near Infrared (NIR), Short Wave Infrared (SWIR), and Ultraviolet (UV) wavelengths of the electromagnetic spectrum, indices such as NDWI, MNDWI, and AWEI can effectively distinguish water from land.
In Phase 3 (Extraction), I will have SNAP isolate only those pixels classified as water across all four indices. This conservative approach ensures high confidence in the retained water pixels. However, it may also create false negatives—pixels that genuinely contain water but are removed because one or more indices failed to classify them correctly.
Next, I generate a Mask layer that removes all NaN (non-water) pixels, retaining only the isolated water pixels. I repeat this workflow for all eight datasets across the four study years and export each output as a GeoTIFF to use in QGIS.
SNAP also supports basic measurements, such as calculating the area of detected water:
Next, I'll use the SNAP’s Time Series tool, which allows comparison of NDWI values across all eight images. Hovering over the same pixel across images reveals seasonal patterns. For example, during summer, the edges of the reservoir dry up and NDWI becomes negative. During monsoon, these pixels revert to water and NDWI becomes positive. The cyclical behaviour is clearly visible:
d. SUBSEQUENT PROCESSING ON QGIS
Although I would have preferred to complete the workflow entirely within SNAP, some steps are more convenient in QGIS. It is a powerful, open-source alternative capable of handling extensive geospatial processing tasks.
In QGIS, I perform data transformation and water-pixel extraction across all eight images.This begins with converting the water-mask GeoTIFFs (Raster format) into Vector polygons (Vector vs Raster explained). Raster pixels do not naturally form grouped containers; converting them into vector polygons helps produce distinct shapes that represent water bodies. These shapes allow me to isolate the Indirasagar Reservoir more easily.
The final extraction and refinement steps could have been done on QGIS itself, however, I have carried it out in ArcGIS Pro which I find to be more user-friendly.
f. SUBSEQUENT PROCESSING ON ARCGIS Pro
ArcGIS Pro is a premium, highly user-friendly GIS application capable of advanced mapping, analytics, and imagery workflows A trial version is also available (I have utilized the same).
The software offers several dedicated extensions for performing specialized workflows specific to certain sectors and industries such as solar, airports, military, urban planning among others. The optional Image Analyst extension can replicate many SNAP operations as well.
My detailed video demonstration explains the ArcGIS Pro steps more clearly than text alone.
Explore the slider below. What have I done here?
Slider 1: Before and After — Refining water polygons
Using the vector layer from QGIS, I have removed small peripheral water bodies surrounding the reservoir. This ensures that area calculations reflect only the reservoir, improving the accuracy of drought assessment.
Although I could have used a pre-existing reservoir shapefile from (using ArcGIS Online, Living Atlas, or other sources), that approach is unsuitable for a temporal (year-to-year) study. Manual extraction ensures each year’s boundary accurately reflects that specific dataset.
This workflow uses:
Besides helping to clean up, organize, and refine the data, ArcGIS Pro is particularly convenient when it comes to symbolizing a layer. I can visualize the output in myriad of ways - using colors, shapes and effects.
In many ways, this is the essence of map-making which is to present the information in a highly impactful way to the viewer.
The output displayed in Figure 20 helps me to ascertain the presence of drought at Theewaterskloof and those edges of the reservoir which receded the most. The output indicates that the reservoir was recovering from the 2016-18 drought - as the water area is increasing almost every year between 2019–2022.
The application also enables statistical charts and dynamic visual analytics:
Tt becomes evident from Figure 21 that there was a sharp drop in water area between the 2021 monsoon and 2022 summer at Indirasagar.
The charts below have been prepared using Microsoft Excel-
Finally, ArcGIS Pro helps me to combine all components together into a printable map:
From this study, while I cannot technically validate the presence of Drought - a hydrological expert with access to historical data and benchmarks would be the right person to make that assessment - the dip in the water area at Indirasagar Reservoir is very evident nonetheless.
From this study, although I cannot formally confirm drought conditions—this requires hydrological benchmarks and long-term time-series data—the decline in Indirasagar’s water area is clearly visible.
I hope you found this post informative and gained a renewed appreciation for water as a precious resource.Feel free to email your thoughts or post it in the video's comments section.
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-
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,
