📄 earth_science

Climate Change Impacts on Reference Evapotranspiration in Arid and Semi-Arid Regions under CMIP6 Scenarios using ERA5-Land and Machine Learning

This study projects that climate change in Iran's arid Semnan Province will significantly increase reference evapotranspiration by up to 16% by the end of the century, particularly in agricultural zones, under CMIP6 scenarios, necessitating urgent adaptive water management strategies.

Original authors: Ali Zolfaghari, Maryam Raeesi, Giuseppe Longo-Minnolo, Azade Soltani, Davoud Kartoulinezhad

Published 2026-07-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ali Zolfaghari, Maryam Raeesi, Giuseppe Longo-Minnolo, Azade Soltani, Davoud Kartoulinezhad

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Thirsty Landscape

Imagine the province of Semnan in Iran as a giant, thirsty sponge sitting between a mountain range and a scorching desert. This sponge relies on rain and underground water to stay alive. The researchers wanted to know: How much thirstier will this sponge get over the next 80 years as the planet warms up?

They didn't just guess; they built a high-tech "weather crystal ball" to predict exactly how much water will evaporate from the ground and plants in the future. This process is called Reference Evapotranspiration (ET0). Think of ET0 as the "thirst meter" of the land. A higher number means the land is sucking up water faster.

The Tools: A Digital Weather Lab

To make these predictions, the team used three main tools, which they combined like ingredients in a recipe:

The Global Forecasters (CMIP6): Imagine five different super-computers, each run by a different country (Australia, China, Japan, etc.). Each computer runs a simulation of the future climate under two different "storylines":
- Storyline A (SSP2-4.5): The "Middle Ground." We try to reduce emissions a bit, but the world still warms up moderately.
- Storyline B (SSP5-8.5): The "Worst Case." We keep burning fossil fuels at a high rate, and the world gets very hot.
The High-Res Map (ERA5-Land): The global computers are a bit blurry (like a low-resolution photo). To get a clear picture of Semnan, the researchers used a detailed European weather map (ERA5-Land) that covers the whole region.
The AI Corrector (Machine Learning): Even the detailed map had some errors (it was slightly too hot compared to real thermometers on the ground). The researchers used a Random Forest machine learning model. Think of this AI as a smart editor that looks at the blurry map, compares it to real ground data, and "fixes" the errors to create a perfect, high-definition map of the past.

The Process: Fixing the Lens and Looking Forward

The researchers followed a specific workflow:

Step 1: Calibration. They took the "blurry" global maps and the "detailed but slightly wrong" European maps. They used the AI to fix the European map so it matched the real weather stations perfectly.
Step 2: Projection. Once the map was fixed, they used it to correct the five global super-computers. This ensured that the future predictions started from a realistic baseline.
Step 3: The Thirst Calculation. They fed these corrected future temperatures into a famous formula (FAO Penman-Monteith) to calculate how much water the land would lose to the air.

The Key Discovery: Heat is the Main Driver

The researchers ran a "sensitivity test" to see which weather factor made the land thirstiest.

The Result: Temperature was the boss. It accounted for the vast majority of the change.
The Analogy: Imagine the land is a car engine. Humidity and wind are like the oil and air filters, but temperature is the gas pedal. If you press the gas pedal (heat) harder, the engine (thirst) revs up immediately. The study found that for every 10% increase in temperature, the land's thirst (ET0) increases by about 6%.

The Future Predictions: Getting Hotter and Thirstier

The crystal ball showed a clear trend of warming, but the speed depends on the storyline:

The "Middle Ground" Storyline (SSP2-4.5):
- The region warms up by about 1°C every 30 years.
- The land's thirst increases by about 2.85 mm per year.
- By the end of the century, the land will be about 6.5% thirstier than it is today.
The "Worst Case" Storyline (SSP5-8.5):
- The region heats up much faster, about 2°C every 30 years.
- The land's thirst skyrockets by 5.2 mm per year.
- By the end of the century, the land will be about 16% thirstier than today.

The Surprise: The Mountains are in Trouble

You might think the desert areas would get the hottest and thirstiest. However, the study found a twist:

The Desert (South): It is already so hot and dry that adding more heat doesn't change its thirst that much. It's like a sponge that is already bone-dry; it can't get much drier.
The Mountains and Farms (North/Central): These areas have more vegetation and are currently cooler. As they warm up, their "thirst meter" spikes dramatically.
The Metaphor: Imagine a dry desert rock and a lush green lawn. If you turn up the heat, the rock doesn't change much, but the green lawn starts wilting and sucking up water frantically. The study predicts that the agricultural zones in the north and center will face the biggest shock in water demand.

The Bottom Line

The paper concludes that climate change is turning up the "heat dial" on Semnan Province. Because the land is getting hotter, it is evaporating water much faster.

The Risk: The areas where people grow food (the northern and central parts) are projected to need significantly more water to survive.
The Solution: The researchers suggest that farmers and water managers need to adapt immediately. They can't just rely on current water plans; they need better irrigation and smarter water use to handle a future where the land is constantly "sweating" more water into the air.

In short: The land is getting hotter, and because of that, it is getting much thirstier, especially in the places where we grow our food.

Technical Summary: Climate Change Impacts on Reference Evapotranspiration in Arid and Semi-Arid Regions under CMIP6 Scenarios

Problem Statement
Climate change is significantly disrupting the hydrological cycle, particularly in arid and semi-arid regions like Semnan Province, Iran, which faces severe heat and water stress. Reference evapotranspiration ( $ET_0$ ) is a critical indicator for estimating crop water requirements and irrigation scheduling. However, existing studies in Iran often rely on coarse-resolution climate datasets, station-based observations with sparse spatial coverage, or earlier-generation General Circulation Models (GCMs). These limitations hinder the ability to capture local climatic gradients and spatial variability necessary for effective water-resource planning. Furthermore, future projections are subject to substantial uncertainty arising from inter-model variability, emission pathways, and downscaling techniques. This study addresses the need for high-resolution, reliable projections of $ET_0$ under future climate scenarios to support adaptive water management in data-scarce, topographically complex regions.

Methodology
The study employs a hybrid framework integrating physical climate modeling, reanalysis data, and machine learning to project $ET_0$ across Semnan Province at a 1 km resolution.

Study Area and Data: The research focuses on Semnan Province, characterized by a transition from the Alborz Mountains to central deserts. The study utilizes five CMIP6 General Circulation Models (ACCESS-CM2, BCC-CSM2-MR, CAMS-CSM1-0, MIROC6, Tai_ESM1) under two Shared Socioeconomic Pathways (SSP2-4.5 and SSP5-8.5) across three time horizons: near (2025–2050), mid (2051–2075), and far future (2076–2099).
Baseline and Reanalysis Processing:
- FAO-56 Penman-Monteith: Daily $ET_0$ was calculated at 10 synoptic stations using observed meteorological data (1990–2014) to establish a baseline.
- Downscaling and Bias Correction: ERA5-Land reanalysis data (originally 9 km) were downscaled to 1 km using ordinary kriging with elevation as an external drift. These downscaled data were then bias-corrected against station observations using a Random Forest (RF) machine learning model. This corrected dataset served as a high-resolution gridded reference.
GCM Downscaling and Bias Correction: CMIP6 outputs were resampled to 1 km and bias-corrected using the Delta Change Factor (DCF) method, utilizing the bias-corrected ERA5-Land data as the reference for the historical period (1990–2014).
Sensitivity Analysis and Projection: A sensitivity analysis of the FAO Penman-Monteith equation was conducted to determine the influence of climatic variables. Given the high uncertainty in wind and humidity projections, the study focused on temperature as the dominant driver. Future $ET_0$ was projected by applying the derived temperature sensitivity coefficients to the bias-corrected temperature projections.
Spatialization: An RF model was trained to spatialize $ET_0$ at 1 km resolution, using station-based $ET_0$ as the target and gridded temperature, DEM, and ERA5-Land variables as predictors.

Key Results

Model Performance: The bias correction significantly improved the accuracy of ERA5-Land temperature data, reducing the Root Mean Square Error (RMSE) from 2.35°C to 1.00°C and the bias from 1.70°C to -0.01°C. The bias-corrected CMIP6 ensemble showed strong agreement with the reference data (correlation coefficient $r$ > 0.97).
Temperature Projections: A steady warming trend is projected across all horizons. Under the intermediate scenario (SSP2-4.5), temperatures are expected to rise by approximately 1.0°C every 30 years. Under the high-emission scenario (SSP5-8.5), the rate accelerates to approximately 2.0°C every 30 years. By the far future (2076–2099), mean temperatures could increase by up to 5°C relative to the baseline under SSP5-8.5.
Sensitivity: Sensitivity analysis confirmed that mean temperature is the dominant driver of $ET_0$ in this arid environment, with an annual sensitivity coefficient of 0.6. This implies that a 10% increase in air temperature leads to an approximate 6% increase in $ET_0$ .
$ET_0$ Projections:
- SSP2-4.5: $ET_0$ is projected to increase by an average of 2.85 mm/year, resulting in a total rise of approximately 6.5% by the end of the century.
- SSP5-8.5: $ET_0$ is projected to increase by 5.2 mm/year, culminating in a total rise of up to 16% relative to the baseline.
Spatial Heterogeneity: The analysis reveals that northern mountainous areas with denser vegetation cover are projected to experience more intense absolute increases in $ET_0$ compared to the southern desert regions. This suggests that agricultural zones, rather than the most arid desert areas, face heightened vulnerability to future moisture deficits.

Significance and Claims
The paper claims that its primary contribution lies in the methodological integration of bias-corrected reanalysis data with machine learning to produce high-resolution ($1$ km) $ET_0$ projections in a region with sparse observational networks. By using bias-corrected ERA5-Land as a gridded observational reference, the study overcomes the limitations of relying solely on station data or coarse GCM outputs.

The authors assert that their findings provide a robust basis for climate-adaptive water management. The projected increases in evaporative demand, particularly in the northern agricultural zones, necessitate urgent adjustments in irrigation scheduling and water resource planning. The study emphasizes that while the approach focuses on temperature-driven changes (a first-order approximation), it offers a pragmatic and reliable assessment for regions where humidity and wind speed projections carry high uncertainty. Ultimately, the research underscores the necessity for optimized irrigation strategies and localized cropping pattern adjustments to mitigate accelerating drought stress and maintain agricultural productivity in the 21st century.