The study, published in Theoretical and Applied Climatology, targets the IMERG product, the Integrated Multi-satellitE Retrievals for GPM, which is generated by NASA’s Global Precipitation Measurement mission. IMERG stitches together observations from a constellation of satellites to produce near-global rainfall maps at high spatial and temporal resolution. Yet over the interior of the Tibetan Plateau, the product suffers from well-documented biases. It tends to overestimate both the intensity and the spatial extent of precipitation in some areas while underestimating rainfall locally in others, and its errors grow worse as rainfall intensifies. Because the plateau feeds the headwaters of major Asian rivers and drives regional hydrological cycles, these errors ripple directly into flood simulation, drought monitoring, and water resource management.

Lead author Jiaxin Zhao and colleagues, including Aili Liu, Chong Wang, Lin Zhao, and Tingting Dong, tackled the problem with a two-stage machine learning pipeline that pairs a convolutional neural network, or CNN, with Extreme Gradient Boosting, better known as XGBoost. The key innovation is that the framework is explicitly aware of precipitation intensity. Rather than treating all rainy days as a single undifferentiated class, the system first classifies each satellite estimate into rainfall categories and then uses that classification to guide the correction of the rainfall amounts themselves.

The technical architecture works as follows. The researchers trained a CNN-based classification model using IMERG estimates together with nine auxiliary topographic and meteorological variables, such as terrain characteristics and atmospheric conditions, with ground-based gauge observations serving as the reference truth. The CNN learns to recognize the spatial patterns and contextual signatures that distinguish light rain from moderate, heavy, and torrential rainfall in satellite data. Crucially, the model outputs not just a single category label but a set of probabilities across all rainfall classes. These probability vectors are then fed into the XGBoost regressor as additional features, alongside the original IMERG values and environmental variables, allowing the gradient-boosted trees to correct the magnitude of rainfall differently depending on how likely an event is to be a deluge rather than a drizzle.

This division of labor exploits the complementary strengths of the two algorithms. CNNs excel at extracting spatial features from gridded data, making them well suited to interpreting the two-dimensional structure of satellite precipitation fields and their relationship to the plateau’s complex topography. XGBoost, a scalable tree-boosting system renowned for its performance on tabular regression problems, handles the numerical correction task efficiently and robustly. By passing the CNN’s rainfall-category probabilities into the regression stage, the framework ensures that the intensity information directly shapes the bias correction, rather than being lost in a one-size-fits-all adjustment.

To evaluate the approach, the team benchmarked the corrected IMERG estimates against two independent, long-term, high-resolution meteorological datasets: the Third Pole long-term high-resolution meteorological forcing dataset, known as TPMFD, and the China Meteorological Forcing Dataset, or CMFD. Both are widely trusted reference products for the region, and comparing against two separate benchmarks helps guard against the possibility that improvements simply reflect fitting to one particular dataset’s quirks.

The results are striking. On a daily timescale, the corrected IMERG substantially outperformed the original product, with the correlation coefficient increasing by 0.15 and the root mean square error and mean absolute error decreasing by 0.46 and 0.04 millimeters per day, respectively. The framework also proved seasonally robust, maintaining satisfactory performance in both the dry and wet seasons, with correlation coefficients of 0.35 and 0.65. In a region where monsoon dynamics produce sharp seasonal contrasts in precipitation regime, that consistency matters: many correction methods tuned for the wet season degrade badly during dry months, or vice versa.

The spatial improvements are equally significant. The original IMERG product over the plateau’s interior contains spurious high-value regions where it paints unrealistic pockets of intense rainfall, alongside localized zones of underestimation. The corrected product suppressed these artifacts, producing rainfall maps whose spatial patterns align much more closely with gauge observations. When compared with the CMFD reference, the corrected IMERG reduced the overestimation of both precipitation intensity and spatial coverage, two of the most persistent failure modes of satellite retrieval over high terrain, where orographic effects and complex cloud microphysics confound passive microwave and infrared retrieval algorithms.

Perhaps the most consequential gains came in the extreme rainfall categories. For heavy and torrential rainfall events, the corrected IMERG achieved correlation coefficient increases of 0.14 to 0.31 and reductions in root mean square error of 1.47 to 2.01 millimeters per day relative to the original product, all while maintaining stable performance for light and moderate rainfall. That balance is rare. Conventional bias correction methods often improve average conditions at the cost of smearing out extremes, or they sharpen extremes at the cost of distorting the drizzle that dominates the record. An intensity-aware framework that improves both ends of the spectrum simultaneously represents a meaningful advance for hydrological extreme-event simulation, where the difference between a corrected and uncorrected heavy rainfall estimate can determine whether a flood model captures a real hazard or misses it entirely.

The implications extend well beyond the Tibetan Plateau. Roughly a third of humanity depends on water that originates in High Mountain Asia, and as climate change intensifies both the hydrological cycle and the frequency of extreme precipitation, the demand for accurate, gauge-independent rainfall estimates in ungauged mountains will only grow. The framework’s ingredients are all publicly available: the IMERG product is distributed freely by NASA, and the TPMFD and CMFD datasets are hosted by the National Tibetan Plateau Data Center. That openness, combined with the modular design of the CNN-XGBoost pipeline, suggests the approach could be adapted to other satellite precipitation products and other data-sparse mountain regions worldwide, from the Andes to the Hindu Kush. For a region where every rain gauge covers hundreds of square kilometers of some of the most inaccessible terrain on Earth, teaching satellites to see rainfall more clearly is not just a technical refinement. It is a step toward understanding, and preparing for, the water future of an entire continent.

Subject of Research: Machine learning-based bias correction of GPM IMERG satellite precipitation estimates over the interior of the Tibetan Plateau

Article Title: A precipitation-intensity-aware CNN-XGBoost framework for IMERG bias correction over the interior of the Tibetan Plateau

Article References: A precipitation-intensity-aware CNN-XGBoost framework for IMERG bias correction over the interior of the Tibetan Plateau. (n.d.). https://doi.org/10.1007/s00704-026-06606-z

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06606-z

Keywords: IMERG, Tibetan Plateau, satellite precipitation, bias correction, convolutional neural network, XGBoost, machine learning, extreme rainfall, hydrology, GPM mission, climate data, remote sensing

The research, published in Environmental Challenges, stands out for the rigor of its methodology. Rather than taking raw climate model output at face value, the team led by Kassa Tesfaye Erenso, Abdella Kemal Mohammed and Tarun Kumar Lohani first evaluated five CMIP6 models against quality-controlled daily observations from eleven ENACTS meteorological stations spanning 1981 to 2022. The stations cover the watershed’s three physiographic zones, from the 2,350-meter highland station at Bonga down to the lowland plains below 1,000 meters. Each model was scored using the Taylor skill score, the modified Kling-Gupta efficiency and root mean square error, producing a composite ranking in which Japan’s MRI-ESM2-0 performed best overall, with temperature efficiency values above 0.81.

A crucial innovation lies in how the ensemble was assembled. Climate models are not statistically independent; shared code, parameterizations and institutional histories mean some simulations carry similar errors. Treating all models equally can therefore give disproportionate weight to closely related model families. The researchers quantified inter-model similarity using pairwise correlations, hierarchical clustering and multidimensional scaling, then combined performance scores with independence factors to produce final ensemble weights ranging from 0.14 to 0.24. Effective ensemble sizes of 3.8 to 4.4 confirmed that the five-model set retains a substantial amount of genuinely independent information. A leave-one-out analysis showed the conclusions did not hinge on any single model.

Bias correction was handled through empirical quantile mapping, a distribution-based technique that aligns the full statistical shape of simulated temperature and precipitation with observed baselines over 1985 to 2004. The improvements were striking. For the watershed-averaged MRI-ESM2-0 simulation, maximum temperature bias fell from minus 2.86 degrees Celsius to essentially zero, and precipitation bias dropped from minus 15.2 percent to minus 2.2 percent. Root mean square error for maximum temperature collapsed from 3.12 to 0.26 degrees Celsius. Critically, an independent validation on the held-out 2005 to 2014 period showed nearly identical skill, with Kling-Gupta efficiency values of 0.78 to 0.94, demonstrating that the correction generalizes beyond its calibration window.

The projections themselves reveal persistent, accelerating warming. By the 2090s, ensemble-mean maximum temperature rises by 2.21 degrees Celsius under the moderate SSP2-4.5 scenario and 3.85 degrees Celsius under the high-emission SSP5-8.5 pathway, while minimum temperatures climb even further, to 2.30 and 4.02 degrees Celsius respectively. The divergence between scenarios grows from a mere 0.17 degrees Celsius in the 2020s to 1.64 degrees Celsius by century’s end, eventually exceeding the inter-model spread itself. Warming is not distributed evenly: the analysis uncovered a pronounced elevation-dependent gradient, with maximum temperature warming increasing by 1.20 degrees Celsius per 1,000 meters of elevation under SSP2-4.5 and by 2.04 degrees Celsius per 1,000 meters under SSP5-8.5, meaning the western highlands emerge as the watershed’s primary warming hotspot.

Precipitation tells a subtler and more troubling story. Annual totals increase only modestly, by roughly 3.5 percent under SSP2-4.5 and 5.8 percent under SSP5-8.5 by the 2090s, concentrated in the September-to-November season and the northeastern headwaters. But the extremes behave very differently. Maximum one-day rainfall rises by 38 percent, precipitation from extremely wet days by 67 percent, and the number of days exceeding 25 millimeters by 125 percent. Meanwhile the September-to-November coefficient of variation surges by about 50 percent under high emissions, signaling far wilder year-to-year swings. The physical explanation follows the Clausius-Clapeyron relationship: a warmer atmosphere holds roughly six to seven percent more moisture per degree of warming, supercharging individual storms even as average rainfall barely moves.

Perhaps the most eye-catching numbers concern how rarely extreme events will remain rare. Fitting generalized extreme value distributions to annual maxima, the team found that the historical 10-year maximum one-day rainfall event of 55 millimeters grows to 78 millimeters and would recur roughly every four years by late century under SSP5-8.5. The historical 100-year rainfall event, at 85 millimeters, intensifies by 59 percent to 135 millimeters and returns approximately every 25 years. The 100-year five-day rainfall total jumps by 61 percent. In contrast, the historical 100-year consecutive dry spell shortens so much that its future equivalent return period stretches to around 200 years, suggesting exceptionally long droughts of past magnitude become far less likely even as rainfall grows more erratic.

The uncertainty analysis adds a strategic dimension. Before 2050, differences among climate models dominate the projection spread, accounting for 52 to 65 percent of total variance, while internal variability contributes a steady 15 to 20 percent. After about 2065, the choice of emissions scenario becomes the largest source of uncertainty at roughly 48 percent. Signal-to-noise analysis shows the anthropogenic warming signal crossing the detection threshold by 2037 under SSP5-8.5 and emerging clearly from natural variability by 2045; under the moderate pathway those milestones slip to 2048 and 2062, a 17-year delay. The comparison between scenarios quantifies the tangible payoff of mitigation: the intermediate pathway limits end-of-century warming by about 1.6 to 1.8 degrees Celsius and buys the basin nearly two additional decades before the climate signal becomes unmistakable.

For water managers, the implications are concrete. The Omo-Kuraz Sugar Development Project and associated irrigation schemes will face higher evaporative demand, more variable reservoir inflows, and design storms that existing infrastructure standards never anticipated. The southeastern lowlands, including the irrigation command area, are flagged as the zone most vulnerable to rainfall variability and associated flood and drought risk, while the northwestern highlands face the steepest warming. The authors recommend climate-adjusted design standards, adaptive reservoir operation, improved flood forecasting and climate-informed irrigation scheduling, tailored to each zone rather than applied uniformly. The projections also align with Ethiopia’s Climate-Resilient Green Economy strategy and support Sustainable Development Goals on clean water and climate action.

The study acknowledges its limits: five models cannot span the full structural uncertainty of the CMIP6 archive, quantile mapping assumes a stationary bias relationship, daily resolution misses sub-daily cloudbursts relevant to flash flooding, and large-scale drivers such as El Nino, the Indian Ocean Dipole and Congo Basin moisture transport were not explicitly diagnosed. Still, by integrating station-validated bias correction, weighted ensembling, extreme-index analysis, return-period statistics and signal-emergence diagnostics into one watershed-scale framework, the work offers a template for climate risk assessment in data-scarce regions. Its central message is stark: in the Omo-Kuraz Watershed, the future will be shaped less by how much rain falls in a year than by how violently it falls, and how often the mercury climbs past thresholds once considered exceptional.

Subject of Research: Bias-corrected CMIP6 projections of temperature and precipitation extremes in the Omo-Kuraz Watershed, Ethiopia

Article Title: Bias-corrected CMIP6 ensemble projections of temperature extremes and precipitation regimes in the Omo-Kuraz Watershed, Ethiopia: Implications for water security

Article References: Erenso, K. T., Mohammed, A. K., & Lohani, T. K. (2026). Bias-corrected CMIP6 ensemble projections of temperature extremes and precipitation regimes in the Omo-Kuraz Watershed, Ethiopia: Implications for water security. Environmental Challenges, 25, Article 101649. https://doi.org/10.1016/j.envc.2026.101649

Image Credits: AI Generated

DOI: 10.1016/j.envc.2026.101649

Keywords: CMIP6, climate extremes, Ethiopia, Omo-Kuraz Watershed, bias correction, water security, elevation-dependent warming, precipitation extremes, SSP scenarios, Lake Turkana, return periods, East Africa

The study, conducted by Hüdaverdi Gürkan of the University of Florida, the Turkish State Meteorological Service, and Adana Alparslan Turkes Science and Technology University, together with Gerrit Hoogenboom of the University of Florida’s Global Food Systems Institute, set out to answer a deceptively simple question: when NASA POWER says it rained, or reports yesterday’s maximum temperature, how closely does that match what actually happened on the ground? The answer, the researchers found, depends heavily on which variable you care about and where the point of interest sits on the map. For temperature and solar radiation, the news is overwhelmingly positive. For humidity, wind, and precipitation, geography matters enormously, and the dataset’s reliability can degrade sharply along coastlines and in mountainous terrain.

The statistical evidence for temperature is striking. Maximum daily temperature records from NASA POWER explained between 87 and 98 percent of the variance in ground observations, with root mean square errors ranging from 1.50 to 6.89 degrees Celsius. Minimum temperatures performed nearly as well, with R-squared values between 0.78 and 0.97 and errors from 1.23 to 8.37 degrees Celsius. Solar radiation estimates tracked ground measurements with R-squared values from 0.76 to 0.97 and root mean square errors between 1.73 and 3.84 megajoules per square meter. Across all temperature and radiation variables, the study reports overall R-squared values between 0.76 and 0.98, a level of agreement that the authors say establishes NASA POWER as a dependable foundation for climate modeling and agricultural decision support in most settings.

The implications for crop modeling are significant. Crop simulation models, which are used worldwide to forecast yields, evaluate irrigation strategies, and assess climate change impacts, are notoriously sensitive to input weather data. Small biases in daily minimum temperature can alter calculations of growing degree days and chilling requirements, while errors in solar radiation propagate directly into photosynthesis and evapotranspiration estimates. Previous work by Hoogenboom and colleagues, including earlier evaluations of NASA satellite-derived temperature and radiation data over the continental United States, laid the groundwork for using such datasets in agricultural applications. The new study extends that legacy by testing the modern POWER product across a far broader range of topographic and climatic settings, from the humid subtropics of Florida to the complex mountain terrain of Türkiye.

That topographic diversity is precisely where the study’s cautionary findings emerge. The researchers found that the dataset’s performance for relative humidity, wind speed, and precipitation is strongly controlled by local geography. In coastal areas, land-sea breeze circulations and marine atmospheric layers introduce discrepancies that the gridded satellite and reanalysis data struggle to capture. NASA POWER produces values on a grid, essentially averaging conditions over cells of finite size, and these spatial averages smooth over the sharp gradients that occur where land meets ocean. A weather station a few hundred meters from the shoreline can experience humidity and wind regimes that differ substantially from the grid-cell average, producing errors that would surprise a user relying on the dataset without checking it first.

High-relief terrain presents a parallel problem that atmospheric scientists call topographic smoothing. When elevation changes dramatically within a single grid cell, as it does across the mountain ranges of Anatolia or the slopes of Washington State’s Cascades, a single gridded value cannot represent the true spread of conditions from valley floor to ridge crest. The dataset’s estimates regress toward the mean of the cell, underestimating the extremes that stations actually record. This phenomenon, well documented in studies of gridded precipitation in complex terrain, means that hydrologists and agronomists working in mountainous regions should treat NASA POWER outputs for precipitation, humidity, and wind with particular skepticism, or apply bias correction before feeding them into models.

The practical value of the study lies in the regional performance thresholds it establishes. Rather than issuing a blanket verdict on the dataset, the authors provide what amounts to a decision guide: researchers can look up their variable of interest and their terrain type, and determine in advance whether raw NASA POWER data is adequate or whether bias correction is necessary. This distinction matters because bias correction is not free. It requires access to local observations, adds processing steps, and introduces its own uncertainties. Knowing that temperature and radiation can typically be used as-is, while coastal precipitation may demand correction, saves both computational effort and scientific risk. In an era when open-access geospatial datasets are increasingly substituted for sparse station networks, such interoperability benchmarks become essential infrastructure for environmental monitoring.

The validation design itself reflects a growing awareness that dataset quality assessments must span decades, not just single seasons. By drawing on records of 15 to 43 years from networks including the Florida Automated Weather Network, the Georgia Automated Environmental Monitoring Network, Washington State University’s AgWeatherNet, and the Turkish State Meteorological Service, the study captures the full range of interannual variability, El Niño and La Niña cycles, heat waves, and drought years that any operational dataset must handle. A dataset that performs well only in average years is of limited use for the extreme events that matter most to farmers and water managers. The multi-decadal approach also guards against the possibility of agreement arising by coincidence in a short window of unusually stable weather.

The research also situates NASA POWER within a crowded field of gridded climate products, including ERA5, MERRA-2, CRU TS, and CHIRPS, each with its own strengths and known weaknesses. What distinguishes POWER is its accessibility: it is free, requires no registration, and delivers daily agroclimatic variables in formats that plug directly into widely used crop models. The authors’ benchmark gives the global community of agricultural modelers, many of whom work in data-sparse regions of Africa, Central Asia, and South America, a defensible basis for choosing POWER, and a clear-eyed understanding of where it will need help. As climate change intensifies the demand for reliable weather information in agriculture, studies like this one transform a free dataset from a convenient substitute into a rigorously characterized scientific instrument.

Subject of Research: Multi-decadal validation of the NASA POWER gridded meteorological dataset against ground weather stations in the United States and Türkiye for climatology and agricultural modeling applications

Article Title: Validating NASA POWER meteorological data for applied climatology and agricultural modeling: a multi-decadal assessment across diverse topographies in the US and Türkiye

Article References: Gürkan, H., & Hoogenboom, G. (2026). Validating NASA POWER meteorological data for applied climatology and agricultural modeling: a multi-decadal assessment across diverse topographies in the US and Türkiye. Theoretical and Applied Climatology, 157(10), Article 621. https://doi.org/10.1007/s00704-026-06532-0

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06532-0

Keywords: NASA POWER, gridded meteorological data, weather station validation, applied climatology, agricultural modeling, crop simulation, solar radiation, temperature data, precipitation accuracy, topographic effects, bias correction, Türkiye