In September 2022, a new paper was published in Nature Sustainability that introduces a potential game-changer in food security [Krishnamurthy R et al., 2022]. The authors were able to predict every single food security crash within the record before the crash occurred, as shown in Figure 1 with a three month ahead of time prediction. Actually, in some cases, even six months before they occurred. How on Earth did they do that?
Dr. Krishna Krishnamurthy, the lead author, completed his Ph.D. at UCLA under Josh Fisher, Hydrosat’s Science Lead, after a decade at the UN World Food Programme leading climate analysis for food security. His expertise has been pivotal in advancing Hydrosat’s satellite constellation.
The economic and statistics theory, Tipping Point Theory, had been traditionally applied to relatively massive and irreversible environmental changes such as ice loss, desertification, and ocean circulation. It was identified to try the well-established approach that works well across diversified applications. The status quo food security prediction capabilities miss a lot of food crises, therefore there was plenty of room for improvement. For example, 78% of food security predictions are captured in general in Ethiopia; but, under certain conditions, this becomes a flip of a coin: >50% of crisis transitions are completely missed [Choularton and Krishnamurthy, 2019].
Outlined in the foundational framework in AGU’s Earth’s Future [Krishnamurthy R et al., 2020] and paper, Research Highlight in Nature Climate Change [Findlay, 2020], four major statistical diagnostics were assessed: [1] [Lenton, 2011]: I) autocorrelation; II) skewness; III) variance increase; and, IV) thresholds (Figure 3). These diagnostics were applied to every spaceborne terrestrial hydrology dataset and found relevance across precipitation (GPM), snowpack (MODIS), evapotranspiration (ECOSTRESS, MODIS, Landsat), soil moisture (SMAP, SMOS, Sentinel-1), and groundwater (GRACE/FO). They also took a look at carbon cycle variables in greenness indices (e.g., NDVI; MODIS, Landsat) and solar induced chlorophyll fluorescence (OCO-2, GOSAT).
[1] Lag-1 autocorrelations identify a critical diminishing of values one period apart whereby a variable becomes less sensitive to small perturbations before transitioning to a different state. Skewness is a measure of symmetry such that as the distribution of data becomes less normally distributed, one may be able to identify a shift to a different state. Increased variance identifies a system characterized by noise that could exhibit flickering whereby strong disturbances push the system past its buffer of resilience into the new state. Finally, thresholds are kind of like tipping points within tipping points such that some new state of forcing data must exist for the larger system to then also move into a new state.
Which tipping point statistic and spaceborne dataset was the favorite? The original thinking was that it would be Increased Variance X Precipitation, while not giving much thought about autocorrelation, skewness didn’t quite make sense in this context, and thresholds were just too easy. On one hand it had to be precipitation because it felt like that’s what everyone else was looking at. Although in retrospect, that was probably part of why this new discovery was made — because it undertook something that most people weren’t. Precipitation was looked at first, only being mildly successful. It was hard to figure out if the problem was precipitation as a variable, the uncertainties and nuances of the precipitation datasets, implementation of the tipping points statistics, or if the whole idea was flawed bunk.
Instead of trying to analytically diagnose those degrees of freedom, NDVI was looked at next because a lot of people were looking at it and it’s a basic dataset to work with. Again, results were lackluster to say the least. But, at least some signals were showing that the implementation was being done correctly.
Next was ET, given Josh’s aptitude and passion for the subject as well as having launched a whole mission focused on ET (ECOSTRESS) as Science Lead and has taken this expertise into his role as our Science Lead, which has also radically transformed the field of remotely sensed ET [Fisher et al., 2021]. His expertise has enabled ET to tell us exactly what plants are seeing/feeling and it can be done at high spatial resolution. Despite this, it brought sub-optimal results. ET is extremely valuable for within-season crop stress; enabling us to see what the plants are feeling, and therefore it’s important for Hydrosat to capture that at high spatiotemporal resolution, despite not necessarily what the plants are going to feel 6 months in advance.
Following a non mind-blowing look at OCO-2 SIF and GRACE-based groundwater, soil moisture was next.
Why not start with soil moisture? The reason being that these other non-soil moisture datasets had longer records with SMAP having launched in 2015.
The first soil moisture analysis showcased that the signal was there, it was clear, and it was loud. The plan worked, and that was exciting. Through the Autocorrelation Metric, the real work was ahead: 1) elucidating the mechanistic explanation for the model; 2) testing its universality; and, 3) figuring out why some food crashes did not behave like most of the others.
There is a lot of complexity beyond just simple autocorrelation for the model, establishing a baseline upon which the anomaly deviates, and at what point a state transition truly did occur. Crucially, the model had to understand when a transition was bad (food crash) or good (recovery). To help organize, distill, and mentally assimilate all this complexity and different types of information, a focus was placed on visualization; specifically, creating a dashboard wherein all the different types of data, information, and statistics that could go into the model could be seen.
After creating the visualization dashboard, the team could fully see what was going on and understand the mechanics of the bigger picture. They had in-depth discussions of how the processes worked both in the real world and the statistical world, figuring out how to integrate food prices into the mix t as a non-linear log function. Additionally, the team also wanted to identify universal trigger thresholds across different regions, and so they started examining the detailed mathematics and mechanisms of how those may be represented in the model. In doing all this, their results became even more refined.
Meanwhile, they also took a look at SMOS soil moisture, which was quite good, but notably not as good as SMAP primarily due to SMAP’s increased resolution and accuracy. A couple of food security crashes were caused by conflict — one could have plenty of soil moisture and food but if a crisis occurred and burned it all or people were prevented from accessing it; then this counts as a food security crash.
Following all this work the team had incredible results, coming up with the model’s name, SMART. The name stands for Soil Moisture Auto-Regressive Threshold. Their paper went in for review at both Nature and Science, being criticized for using Tipping Point Theory for non-traditional tipping points, and that their sample size was too small. Ultimately, their paper found a home in Nature Sustainability.
Different institutions are starting to look at SMART, especially from humanitarian organizations including the World Bank. The team are hoping to continue exploring the analysis as well as helping to implement SMART in famine prediction systems, ultimately saving lives.
While the statistics and model are now established, SMART does rely on SMAP, which is well past its prime mission and beyond optimal for modern science. The US National Academies of Sciences, Engineering, and Medicine did not prioritize measurements to NASA for a follow-on mission to SMAP. While there have been some gap-fill solutions and great data coming out of ESA, it is important to ensure that the future of soil moisture does not dry up.
At Hydrosat, Josh is driving how we revolutionize spaceborne measurement of thermal infrared with immediate applications for agriculture, and utilizing AIML to forecast crop yield. We’ve figured out how to create daily field-scale measurements of TIR and VNIR beyond what institutional space agencies are able to offer. Hydrosat is initially focusing on within-season crop stress and water use, however our data can also be used to create unprecedented spatiotemporal resolution soil moisture. Hydrosat’s first launch of a constellation of 17+ satellites took place in 2023, it could be the right moment to detect these signals in time to mitigate increasing famines in the years to come (Figure 6). Hydrosat is driving its mission to be a positive tipping point into an exciting new state of knowledge and capabilities for food security prediction.
This article, originally written by Hydrosat’s Science Lead, Josh Fisher, can be found at the link here.
Hydrosat’s constellation will provide unprecedented field-scale (<50 m) TIR and VNIR data, every day.
References
Choularton, R. J., and P. K. Krishnamurthy (2019), How accurate is food security early warning? Evaluation of FEWS NET accuracy in Ethiopia, Food Security, 11(2), 333–344.
Findlay, A. (2020), Detecting famine from space, Nature Climate Change, 10(4), 273–273.
Fisher, J. B., S. Soenen, A. Abello, F. Werner, K. Lalli, R. Dalby, and P. Fossel (2021), Towards daily, field-scale, global thermal infrared measurements from space, paper presented at AGU Fall Meeting 2021, AGU.
Fisher, J. B., et al. (2020), ECOSTRESS: NASA’s Next Generation Mission to Measure Evapotranspiration From the International Space Station, Water Resources Research, 56(4), e2019WR026058.
Krishnamurthy, P. K., J. B. Fisher, and C. Johnson (2011), Mainstreaming local perceptions of hurricane risk into policymaking: A case study of community-based vulnerability GIS in Mexico, Global Environmental Change, 21, 143–153.
Krishnamurthy R, P. K., J. B. Fisher, D. S. Schimel, and P. M. Kareiva (2020), Applying tipping point theory to remote sensing science to improve early warning drought signals for food security, Earth’s Future, 8(3), e2019EF001456.
Krishnamurthy R, P. K., J. B. Fisher, R. J. Choularton, and P. M. Kareiva (2022), Anticipating drought-related food security changes, Nature Sustainability, 1–9.
Lenton, T. M. (2011), Early warning of climate tipping points, Nature Climate Change, 1(4), 201–209.
