Satellite Records Reveal Rising Rainfall in One of Earth’s Driest Corners of Iraq

In one of the most water-stressed corners of the Middle East, a team of Iraqi researchers has pieced together a thirty-five-year portrait of rainfall behaviour along the Arar–Karbala dryland transect in southwestern Iraq, and the results challenge some assumptions about how hyper-arid landscapes respond to a changing climate. Drawing on monthly precipitation data from the NASA POWER satellite-derived database covering 1990 to 2025, the study, published in Theoretical and Applied Climatology, documents a statistically significant upward trend in annual rainfall of just over two millimetres per year, alongside a striking concentration of the region’s scarce precipitation into a small number of relatively wet months. In a landscape where mean annual rainfall barely reaches 63 millimetres, even modest shifts in the timing and clustering of rain carry profound consequences for water resources, agriculture, and the people who depend on both.

The research team, led by Ruqaya Ahmed M. Amin of AL-Iraqia University in Baghdad, analysed data from three climatic grid cells straddling the transect, a corridor that stretches across some of the driest terrain in southwestern Iraq. Rather than relying on sparse ground stations, which are notoriously uneven in this part of the world, the researchers turned to NASA POWER, a publicly available reanalysis product that blends satellite observations with modelled atmospheric data. This choice reflects a growing recognition that in data-scarce dryland environments, satellite-based datasets may be the only practical way to build a continuous, long-term climatic record. It also places the study within a broader movement in Middle Eastern climate science, where reanalysis products have increasingly been used to fill observational gaps across the Arabian Peninsula, the Levant, and North Africa.

To quantify how rainfall varied from year to year, the team applied Z-score standardisation, a statistical technique that expresses each year’s precipitation as a deviation from the long-term mean, allowing unusually dry and unusually wet years to be identified on a common scale. The analysis singled out 2018, 2020, and 2023 as very wet anomalies, years in which rainfall climbed well above the historical average. Perhaps more revealing was the behaviour of what the authors call the Wet-Month Contribution Ratio, or WMCR, a threshold-based metric that measures how much of the annual rainfall total is delivered by months receiving at least 20 millimetres. In a hyper-arid setting, where most months pass with little or no rain, this ratio effectively captures how dependent the water budget is on a handful of productive storms.

The WMCR results were among the most striking findings in the study. The contribution of relatively wet months to annual rainfall varied strongly between years, but high values became noticeably more frequent after 2015, peaking at 0.91 in 2020. In practical terms, that means nearly all of the year’s rain in 2020 fell in just a few months, a pattern that hydrologists describe as increased precipitation concentration. Concentrated rainfall is a double-edged sword in drylands: the same total volume of water arriving in fewer, heavier events is harder for soils and vegetation to absorb, raises flash-flood risk, and reduces the slow recharge of soil moisture and shallow aquifers that sustained ecosystems and traditional agriculture. The finding echoes similar concentration trends documented in Spain, the Mediterranean, and other semi-arid regions, suggesting that the intensification of the hydrological cycle may be reshaping dryland rainfall regimes well beyond Iraq’s borders.

Trend detection relied on two of the most widely used tools in climatology: the Mann–Kendall test, a nonparametric method that identifies monotonic trends without assuming any particular data distribution, and Sen’s slope estimator, which provides a robust measure of the trend’s magnitude. Together they revealed a significant increasing trend in annual rainfall, with a Sen’s slope of +2.03 millimetres per year. The wet-month signal strengthened in parallel: the frequency of wet months increased at a rate of 0.050 months per year, and the rainfall accumulated within those wet months grew by 1.76 millimetres per year. In other words, the region is not simply getting marginally wetter overall; its wet episodes are becoming both more frequent and more productive, while the arid baseline remains essentially unchanged.

That baseline remains formidable. Mean annual reference evapotranspiration, a measure of the atmospheric demand for water driven by temperature, sunshine, and wind, reached 2519.92 millimetres per year, dwarfing the 62.87-millimetre mean annual rainfall by a factor of roughly forty. The resulting aridity index, the ratio of precipitation to potential evapotranspiration, stood at approximately 0.025, placing the region firmly in the hyper-arid class under standard climate classifications. This means that despite the encouraging upward rainfall trend, the fundamental water balance of the region remains overwhelmingly negative: the atmosphere demands vastly more water than the sky delivers. Any increase in rainfall, however welcome, is absorbed into an enormous evaporative deficit, which is why the authors emphasise that hydroclimatic conditions remained highly arid throughout the study period even as wet anomalies multiplied.

The second half of the study turned from diagnosis to prediction, pitting two very different forecasting approaches against each other. The first, SARIMA, or Seasonal AutoRegressive Integrated Moving Average, is a classical statistical model that explicitly encodes the seasonal rhythm and temporal dependence structure of a time series, making it a natural fit for rainfall data with strong winter-wet and summer-dry cycles. The second, XGBoost, is a machine learning algorithm based on gradient-boosted decision trees that has earned a reputation for outperforming traditional models across many hydrological prediction tasks. The authors evaluated both models on an independent testing period using three standard metrics: root mean square error, mean absolute error, and the coefficient of determination.

The verdict was sobering. XGBoost achieved an RMSE of 19.24 millimetres, an MAE of 9.13 millimetres, and an R² of just 0.01, while SARIMA posted an RMSE of 19.43 millimetres, an MAE of 9.55 millimetres, and an R² of −0.01. An R² near zero means the models explained essentially none of the variance in the test data, barely improving on a naive prediction. The machine learning approach was only marginally better than the statistical one, and neither demonstrated meaningful predictive skill. This result carries an important lesson for the increasingly fashionable field of AI-driven environmental forecasting: in hyper-arid environments, where rainfall is sparse, episodic, and dominated by rare synoptic events, the signal available to any model, however sophisticated, may simply be too weak and noisy to extract reliable month-to-month predictions. Data-hungry algorithms cannot conjure information that the climate itself does not provide.

For the exploratory projections covering 2026 to 2035, the team chose SARIMA, a decision grounded not in superior accuracy but in the model’s structural advantages: it explicitly represents seasonal temporal dependence and provides model-based prediction intervals, allowing uncertainty to be quantified and communicated. The authors are careful to note that this choice does not imply SARIMA outperformed XGBoost. The projections suggest that the region’s seasonal rainfall architecture will persist, with winter remaining the wettest season and summer the driest, while forecast uncertainty widens steadily with lead time, a familiar and honest feature of statistical extrapolation. Such exploratory outlooks are not crystal balls, but they offer planners a plausible envelope of future conditions against which drought contingencies and water-allocation strategies can be stress-tested.

The broader significance of the study lies less in any single number than in its integrated approach. By combining anomaly detection, concentration analysis, trend testing, hydroclimatic classification, and comparative forecasting within a single framework, the researchers have assembled exactly the kind of evidence base that drought monitoring, rainfall-risk assessment, and adaptive water-resource planning require in regions where ground observations are thin and climate pressures are mounting. As competition for water intensifies across the Middle East, and as global assessments warn of deepening aridity in the Mediterranean-Middle East hotspot, studies of this kind demonstrate that even the driest landscapes can be monitored, understood, and, to a degree, anticipated, provided the analytical tools are matched honestly to the limits of the data. For southwestern Iraq, the message is nuanced: more rain may be falling, but it arrives in fewer, sharper bursts, on land where the atmosphere’s thirst remains nearly insatiable.

Subject of Research: Rainfall variability, concentration, and exploratory forecasting in a hyper-arid dryland region of southwestern Iraq

Article Title: Rainfall concentration variability and exploratory forecasting in a dryland region of Southwestern Iraq

Article References: Amin, R. A. M., Shinichel, B. S., Abbas, A. M., & Suhail, M. A. (2026). Rainfall concentration variability and exploratory forecasting in a dryland region of Southwestern Iraq. Theoretical and Applied Climatology, 157(11), Article 696. https://doi.org/10.1007/s00704-026-06624-x

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06624-x

Keywords: rainfall variability, drylands, hyper-arid climate, southwestern Iraq, NASA POWER, SARIMA, XGBoost, Mann-Kendall test, Sen's slope, aridity index, drought monitoring, precipitation concentration