Application of Multi ML-Model for Multi-Scale Meteorological Drought Monitoring under CMIP6-Climate Scenario in Data-Scarce Semi-Arid Regions

Imagine the Earth's weather as a giant, chaotic orchestra. Sometimes the instruments play in perfect harmony, bringing gentle rain and steady winds. Other times, the conductor gets lost, and the music turns into a screeching storm or a deafening silence. This silence is what scientists call "drought"—a period where the rain just doesn't show up, threatening the food we eat and the water we drink. To understand this silence before it gets too loud, scientists use special tools called "indices." Think of these like a weather thermometer, but instead of measuring heat, they measure how dry or wet the air is compared to the usual average. One famous tool is the SPI (Standardized Precipitation Index), which acts like a scorecard: a high score means a wet party, a low score means a dry desert, and a zero means things are just normal.

But here's the tricky part: the Earth is huge, and our weather models are like giant, blurry maps. They can tell us the general vibe of the climate, but they often miss the tiny, local details that matter most to a specific river valley. To fix this blurry map, scientists use "downscaling," which is like taking a low-resolution photo and using a super-smart computer to sharpen the pixels until you can see individual leaves on the trees. Once they have a sharp picture, they need to predict the future. Since the weather is a messy, non-linear puzzle (where a small change can cause a huge effect), they turn to "Machine Learning." These are computer programs that learn from history, kind of like a student who studies thousands of past exams to guess the answers on the next one. The big question is: which computer student is the smartest at guessing when the rain will stop?

This paper is a deep dive into the Karkheh Hydrosystem, a massive river basin in Iran that is crucial for feeding millions of people but is also a semi-arid region prone to drying out. The researchers wanted to know: if we look at the future using the latest climate models (CMIP6), how bad will the droughts get? And more importantly, which machine learning "student" is the best at predicting these dry spells so we can prepare?

The team set up a digital laboratory. They took data from four different super-computer climate models and ran them through two different "future scenarios." One scenario, called SSP126, is the "optimistic" path where the world tries hard to be green and sustainable. The other, SSP585, is the "pessimistic" path where we keep burning fossil fuels and the climate goes wild. They fed this data into five different machine learning models: GPR, XGBoost, SVM, MLP, and Random Forest. Think of these five models as five different detectives trying to solve the mystery of the missing rain. They tested these detectives on historical data from 1966 to 2018 to see who could guess the past droughts most accurately.

The results were clear. The detective named GPR (Gaussian Process Regression) was the star of the show. It didn't just win; it dominated. When the researchers looked at how well each model guessed the drought levels, GPR achieved a score (called R²) of 0.944 at the 6-month scale. That is an incredibly high score, meaning it got the pattern almost perfectly right. The other detectives, like XGBoost and SVM, did a good job, but they couldn't match GPR's precision. The study found that looking at a 6-month window was the "sweet spot" for prediction. It was long enough to smooth out the daily noise but short enough to catch the real drought trends.

When the team used the best detective (GPR) to look into the future (2026–2039), the story split depending on which scenario you believed. Under the optimistic SSP126 scenario, the future looks relatively calm. The river basin will mostly stay in "normal" or slightly wet conditions. Droughts will happen, but they will be mild, like a light drizzle of dryness that is easy to manage. However, under the pessimistic SSP585 scenario, the weather gets much more chaotic. While the average might still look okay, the swings become wilder. The simulations suggest that under this "worst-case" path, the region could face much more intense wet periods (floods) and, crucially, much more severe droughts.

Here is the twist that the paper highlights: as you look further into the future with longer time windows (like 12 months), the difference between the two scenarios grows. In the 12-month view, the optimistic scenario stays safe with only mild droughts. But the pessimistic scenario? It starts showing signs of severe drought (SPI values dropping below -1.5) and even moderate drought (SPI between -1 and -1.5). The paper explicitly notes that under the pessimistic path, the Karkheh basin could experience a severe drought event in 2037, a risk that simply doesn't appear in the optimistic timeline.

The researchers also discovered something interesting about time scales. The longer you wait to measure the rain, the less often droughts seem to happen. If you look at just 3 months, about 42% of the time is dry. But if you look at a full 12 months, that drops to about 13–14%. This is because long-term averages smooth out the short dry spells. However, the paper warns that even if droughts are less frequent in the long run, when they do happen in the pessimistic scenario, they hit harder and last longer.

So, what's the takeaway? The paper suggests that for planning purposes, we should trust the GPR model and focus on the 6-month timeframe for medium-term decisions. It's the most reliable tool for the job. For the long term (12 months), the data suggests that if we choose the pessimistic path, we need to prepare for extreme swings—both floods and severe droughts. The optimistic path offers a much more stable future, but the paper makes it clear that the risk of severe drought is real if we don't change our course. The authors conclude that while the future isn't hopeless, we need to be smart about storing water when it's wet and managing it carefully when it's dry, especially if the world keeps heading toward that wilder, more unpredictable climate scenario.