SOON after the invention of the telescope in 1608, astronomers in Europe began observing the sun and noticed dark spots on its surface in varying numbers, and started tracking the daily and monthly number of these sunspots. In 1749, the record-keeping of sunspots was institutionalised. Earlier, non-continuous records from various sources were collected, compiled and prepended, giving us what is known today as the International Sunspot Number dataset. This dataset has proved essential to understanding solar physics, the relationship between the solar cycle and Earth's climate, space weather, solar flares, and the risks to modern technology like communication systems, GPS, power grids, etc. And it all began, not for a practical use, not out of profit motives, but simply because it could be done as an exercise in data collection, driven by scientific curiosity, to know the world we inhabit.
For the last few weeks, news outlets have reported on the floods and the damage they have caused. I have been struck by the seeming lack of official data about flooding and resulting agricultural losses. Instead, the vacuum left the door open to just about anyone to inject their own estimates. While the waters were still rising and making their way downstream, questionable estimates of agricultural losses were being thrown around, gaining traction.
One that was widely cited was an estimate by the Pakistan Business Forum. Those estimates were picked up and repeated by successive news reports, including some ministers. In the absence of alternatives, they even made their way into reports by foreign news outlets. While I am sure the PBF is capable of and resourced to do a great many things, I am certain that generating running estimates of countywide agricultural losses is not among them. Its estimates are not accompanied by any meaningful description of the used methodology and are likely projections derived from incomplete reports of the Provincial Disaster Management Authority and other governmental bodies.
Even without floods, accurate figures about crop production are a constant challenge. Estimates, which inform decisions about the export of agricultural products, vary widely and have led to bad calls. There is no reliable, continuous data collection mechanism in place.
Estimates, which inform decisions about the export of agricultural products, vary widely and have led to bad calls.
On April 16, 2024, Dubai, which usually receives 140 to 200 mm of rain annually, was hit by 250 mm of rain in less than 24 hours and suffered flooding in some areas. It took only one day for the European Space Agency's Sentinel-2 satellite to release images of the affected areas. Nasa's Landsat-9 satellite was able to survey and release imagery of the same area after three days, on April 19. The data from these satellites is open and free to use by all. Depending on technical specifications, such Earth observation satellites are capable of surveying large areas, for many different applications, quickly and irrespective of accessibility challenges on the ground.
Pakistan's flood-prone areas are, of course, much larger than Dubai. I had the opportunity to speak to Dr Murtaza Khan, one of the few Pakistanis who did his PhD work in the area of remote sensing and hyperspectral imaging. A cautious estimate for the time needed to capture Pakistan's flood-affected areas should be in the range of one to a few weeks.
The accuracy of estimates built on top of these images, such as crop classification or flood damage estimates, depends greatly on the spatial and spectral resolution with which sensors capture images. Spatial resolution is the distance on the ground that is captured by a single pixel of the image -- the smaller, the better. For Sentinel-2 and Landsat-9, the spatial resolutions for the visible spectrum are 10 metres and 30m, respectively. For Sentinel-2's near-infrared and shortwave-infrared, used to detect the presence of water, the resolutions are approximately 10m and 20m, respectively. Both Sentinel-2 and Landsat-9 have more than 10 spectral bands. Generally, the higher the number of spectral bands, the better the ability of the sensor to capture images that can be used to differentiate between ground cover types, ie, types of crops.
If that does not impress you, the good news is that Suparco's own Pakistan Remote Sensing Satellite (PRSS-1) has even higher resolutions in the range of 0.98m to 2.89m for its panchromatic and multi-spectral bands, with a revisit time of four days. The bad news is that the data from PRSS-1 appears to be tightly held and is not available to the public nor shared with more than a very few government departments. Neither is the spectral information of the satellite available, nor is it clear how the data may be accessed, even when it is mentioned that data may be made available for academic purposes.
Satellite images by themselves are not sufficient. To derive crop and damage estimates from them, they are passed through image-processing and machine-learning algorithms. Different parameters of those algorithms will produce different estimates. Finding the optimal parameters requires additional data, which people in the business call the 'ground truth', of the actual situation on the ground to tune and train algorithms. And that is our dilemma: The best satellite imagery is under lock and key. Ground truth data that could be used to develop better image-processing algorithms is scattered or not readily available. The result is multiple estimates that can vary widely, sometimes by as much as 100 per cent, making it impossible to base informed decisions on.
The answer will not come from a one-off committee tasked to produce estimates of crop damage, as has been done again this time. What is needed is a monitoring mechanism that is made routine and that gives a credible and continuous picture. If government departments lack the expertise to develop a capability approaching it, then at least a limited dataset of satellite imagery accompanied by ground truth data (possibly collected by a one-time, painstaking effort) should be prepared and released to the public; maybe a Kaggle-style national agri-hackathon can be organised so that individuals and teams with technical expertise can develop the algorithms and expertise that are needed.
The writer has a PhD in education.
Published in Dawn, October 3rd, 2025