The research, led by Satti J. R. Kamenan of the National Polytechnic Institute Félix Houphouët-Boigny in Yamoussoukro, is the first attempt at near real-time rainfall prediction in any of Ivory Coast’s river basins. The team built a forecasting framework around Long Short-Term Memory neural networks, a deep learning architecture specifically designed to remember patterns across long sequences of data. LSTM networks, first proposed by Sepp Hochreiter and Jürgen Schmidhuber in 1997, use internal gates—forget, input and output gates—that regulate what information is stored, discarded or passed forward through the network’s memory cells. This structure allows them to capture the long-term temporal dependencies and nonlinear behavior that characterize daily rainfall, something conventional statistical models have long struggled with.

Feeding the neural network required an ambitious data pipeline. The researchers combined ground-based rainfall records from SODEXAM, the Ivorian meteorological agency, covering ten stations in the Marahoué watershed from 1961 to 2022, with two NASA satellite products: GPM IMERG, a high-resolution precipitation dataset with a spatial resolution of 0.1 degrees and roughly 30-minute temporal resolution, and MERRA-2, a reanalysis that blends satellite observations, ground measurements and numerical weather models into consistent atmospheric fields including temperature, wind, pressure and humidity. All of this was processed on Google Earth Engine, the cloud-based platform that has become a workhorse for large-scale environmental data analysis. Before the satellite data could be trusted, the team validated it against rain gauge observations from 2000 to 2017, finding correlation coefficients of 0.80 at Bouaflé and 0.83 at Zuénoula—meaning the satellite product captured more than 80 percent of the observed rainfall variability.

Quality control was a critical and methodical step. The historical gauge records contained gaps of 7.29 to 8.04 percent, and raw observations can harbor errors that would corrupt any machine learning model. The team applied a two-stage screening protocol: first, the interquartile range method flagged values that deviated unusually far from each station’s typical distribution; second, a spatial consistency check using the network-wide median and median absolute deviation determined whether a flagged value was physically plausible or simply wrong. Crucially, extreme rainfall events that were spatially coherent across the network were retained even when they exceeded statistical thresholds, preventing the accidental deletion of genuine meteorological extremes. Only 689 of 29,510 observations—2.34 percent—were ultimately identified as outliers, and erroneous values were corrected using the validated GPM IMERG estimates as an auxiliary source.

Selecting the right predictors was equally deliberate. From twenty-one candidate atmospheric variables, the researchers used a Pearson correlation matrix to eliminate multicollinearity, keeping only variables with pairwise correlations below 0.60. They then ran a Random Forest model to rank the remaining candidates by importance, retaining twelve predictors—including seven-day antecedent rainfall, vertical atmospheric motion at 500 hectopascals, specific humidity at multiple pressure levels, and wind components—that each contributed meaningfully to predictive accuracy. The final dataset spanned 15,094 daily time steps from January 1984 to April 2024, with data through January 2020 used for training and the final four years reserved entirely for independent validation, a rigorous test of whether the models had truly learned generalizable patterns rather than memorizing the past.

Three separate LSTM models were trained, one for each forecast horizon: one day ahead, three days ahead and seven days ahead. The architecture settled on three layers of 100 neurons each, a learning rate of 0.001 with automatic reduction, a batch size of 32, and an early stopping mechanism that halted training when validation loss stopped improving. The team then pitted the LSTM against three popular machine learning benchmarks—Random Forest, Extra Trees and XGBoost—using standard metrics including the coefficient of determination, Nash-Sutcliffe Efficiency, normalized root mean square error and mean absolute error. The results were striking. At the one-day horizon, the LSTM achieved R-squared and NSE values exceeding 95 percent at both stations, with a normalized RMSE below 10 percent and a mean absolute error under 1 millimeter, placing it firmly in the ‘excellent’ category by conventional hydrological standards.

The gap between the deep learning model and its competitors widened as the forecast horizon stretched. At three days ahead, the LSTM still maintained R-squared values between 88 and 90 percent with a normalized RMSE below 10 percent, while the tree-based models saw their accuracy slip, with R-squared often dropping below 65 percent. By seven days ahead, the LSTM was the only model still producing statistically reliable forecasts, with R-squared values between 70 and 81 percent and errors remaining controlled, whereas Random Forest and Extra Trees deteriorated dramatically—mean absolute errors exceeding 7 millimeters and NSE values sometimes falling below 40 percent—and XGBoost performed worst of all. The explanation lies in fundamental architecture: tree-based ensembles treat observations largely as static feature sets, while the LSTM’s recurrent memory cells explicitly model the sequential structure and long-range dependencies inherent in daily precipitation series.

Honest scrutiny of the model’s errors revealed both strengths and limits. Residual analysis showed that predictions were largely unbiased for light to moderate rainfall, which constitutes the majority of events in the Marahoué basin, with mean residuals near zero at both stations. But the Kolmogorov-Smirnov test confirmed the residuals were not normally distributed, and a class-by-class examination exposed an intensity-dependent bias: some low-intensity events were slightly underestimated while moderate and heavy rainfall tended to be overestimated, with error dispersion growing sharply for extreme events. This is a well-documented weakness of deep learning applied to precipitation—intense rainfall is rare, irregular and heavily skewed, so models see too few examples to learn its behavior precisely, and recent studies have shown LSTM networks tend to smooth high-intensity peaks rather than reproduce them faithfully.

For the people of Bouaflé and Zuénoula, however, the practical implications are significant. Early warning systems depend less on perfectly reproducing once-a-decade extremes than on reliably tracking the ordinary rainfall regime and flagging major events in time for communities to act—and the LSTM delivers exactly that at lead times of one to three days, with usable guidance even at seven. The authors point to clear paths forward: enriching the training data with additional atmospheric predictors such as convective available potential energy, total column water vapor and sea surface temperatures over the Gulf of Guinea and ENSO regions, applying post-prediction bias correction, and exploring hybrid models that couple neural networks with physical approaches for extreme events. As climate change intensifies rainfall variability across West Africa, this study demonstrates that a developing region can build world-class forecasting capability from open satellite data, cloud computing and a well-designed neural network—no expensive radar infrastructure required.

Subject of Research: Deep learning-based daily rainfall forecasting using satellite and reanalysis data for flood early warning in the Marahoué River basin, Ivory Coast

Article Title: Intelligent daily rainfall prediction for early warning using deep learning and satellite data: application to Bouaflé and Zuénoula stations, Ivory coast

Article References: Kamenan, S. J. R., Youan, T. M., Adja, M. G., Soro, S. I., & Kouassi, A. M. (2026). Intelligent daily rainfall prediction for early warning using deep learning and satellite data: application to Bouaflé and Zuénoula stations, Ivory coast. Advances in Statistical Climatology, Meteorology and Oceanography, 12(1), 173-193. https://doi.org/10.5194/ascmo-12-173-2026

Image Credits: AI Generated

DOI: 10.5194/ascmo-12-173-2026

Keywords: deep learning, LSTM neural networks, rainfall prediction, flood early warning, GPM IMERG, satellite data, Ivory Coast, Marahoué River basin, machine learning, hydrology, MERRA-2, extreme rainfall

The research, published in Theoretical and Applied Climatology by Eskinder Gidey and Paidamwoyo Mhangara of the University of the Witwatersrand, stitched together two very different strands of climate evidence. For the historical record, spanning 1984 to 2023, the authors drew on NASA POWER data built on the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2), a satellite-assimilating reanalysis product that reconstructs consistent daily climate variables across the globe. For the future, they turned to the Coupled Model Intercomparison Project Phase 6 (CMIP6), using the medium-emissions SSP2-4.5 scenario to project conditions in Gauteng from 2024 to 2050.

What makes the analysis distinctive is its statistical machinery. Rather than relying on simple averages, the researchers applied an outlier-resistant, median-based robust approach to tame erratic values in the datasets. They then ran the Modified Mann-Kendall test, a non-parametric trend detector that corrects for the autocorrelation that plagues climate time series, and estimated the magnitude of change with Sen’s slope estimator, which computes the median of all pairwise slopes and is therefore largely immune to extreme outliers. All of the processing was carried out in RStudio, and the combination allowed the team to distinguish genuine long-term signals from the year-to-year noise that dominates semi-arid climates.

The historical results are unambiguous on temperature. Gauteng warmed at a rate of about 0.02 degrees Celsius per year between 1984 and 2023, a statistically significant trend with a Kendall’s tau of 0.28 and a p-value of 0.01. That warming translated directly into more frequent heat extremes: heatwave events climbed by roughly 0.04 events per year, a highly significant increase with tau of 0.30 and p below 0.001. Intriguingly, cold waves also ticked upward slightly, by about 0.01 events per year, but that trend failed the significance test with a p-value of 0.26, suggesting it may be a statistical artefact of natural variability rather than a real climatic shift.

The water-side variables tell a more troubling story. Precipitation across the province declined by 3.63 millimetres per year over the four-decade observation window. Cloud cover fell by 0.33 percent per year, and both soil moisture and relative humidity showed declining trends as well. For a landlocked province perched on continental high ground, far from major rivers, these coupled declines matter enormously: less rain, drier soils, drier air and thinner cloud all conspire to reduce the water that reaches reservoirs, recharges groundwater and sustains the wetlands and streams of the Vaal and Crocodile river systems that Gauteng depends on.

The projections sharpen the warning. Two CMIP6 models, the Australian ACCESS-CM2 and the United Kingdom’s HadGEM3-GC31-LL, both foresee significant warming in Gauteng between 2024 and 2050. ACCESS-CM2 projects a temperature rise of 1.68 degrees Celsius, equivalent to a trend of 0.35 percent per year, while HadGEM3-GC31-LL projects 1.60 degrees Celsius, or 0.33 percent per year. On the precipitation side, the models diverge in magnitude but agree in direction: HadGEM3-GC31-LL projects rainfall declining by 7.26 millimetres per year, while ACCESS-CM2 projects a gentler decline of 1.95 millimetres per year. Either way, the province faces a hotter future with less water falling from the sky.

The implications ripple far beyond climatology. Gauteng is the engine of the South African economy, home to Johannesburg, Pretoria and a population that continues to grow through migration, all of it concentrated on a water supply already stretched by periodic drought. The province’s water security has long depended on inter-basin transfers, most famously the scheme that pipes water from the Lesotho highlands into the Vaal system. A simultaneous increase in evaporative demand, driven by rising temperatures and falling humidity, together with declining rainfall, means that reservoir inflows could shrink even as household and industrial demand climbs. Previous research on urban water demand in Gauteng has already flagged the compounding effect of climate change and population growth on the province’s supply systems.

The study’s authors frame their findings as a contribution to South Africa’s National Development Plan 2030, which calls for climate-resilient water management and robust early warning systems. By combining multi-decadal reanalysis observations with CMIP6 projections and rigorous trend-detection statistics, the work offers water managers a defensible quantitative baseline: not just the assertion that conditions are changing, but measured rates of change for temperature, heatwave frequency, rainfall, soil moisture, humidity and cloud cover, together with model-based expectations for the coming quarter century. That kind of evidence is what allows planners to size new infrastructure, set demand-management targets and design drought triggers with some confidence about the trajectory they are planning against.

There are also broader scientific lessons in the methodology. Reanalysis products like NASA POWER (MERRA-2) provide consistent, freely available daily climate data even in regions with sparse ground-based station networks, a chronic problem across much of Africa. Pairing such observations with the CMIP6 archive, and filtering both through robust, autocorrelation-aware statistics, offers a template that can be replicated for other provinces and catchments. The researchers acknowledge the inherent uncertainties of model projections, and the divergence between the two models’ rainfall estimates is a reminder that scenario-based futures are ranges rather than forecasts. But the direction of travel, warmer, drier and more extreme, is consistent across both the observed record and the projections.

For the residents of Gauteng, the message distilled from forty years of satellite-era data and twenty-six years of model projections is stark but actionable. The province has already lost rainfall at more than three and a half millimetres per year while gaining heatwaves at a statistically robust clip, and the coming decades promise roughly another degree and a half of warming on top of what has already occurred. Whether that future translates into crisis or into a managed transition will depend on how quickly the province’s water institutions, municipalities and households absorb what the data is telling them, and on whether the policies built for the climate of the twentieth century are retooled for the one now arriving.

Subject of Research: Historical and projected climate variability and trends in Gauteng, South Africa, and their implications for water resources

Article Title: Climate variability and its implications for water resources in Gauteng, South Africa (1984–2050) based on NASA POWER (MERRA-2) observations and CMIP6 SSP2-4.5 projections

Article References: Gidey, E., & Mhangara, P. (2026). Climate variability and its implications for water resources in Gauteng, South Africa (1984–2050) based on NASA POWER (MERRA-2) observations and CMIP6 SSP2-4.5 projections. Theoretical and Applied Climatology, 157(10), Article 676. https://doi.org/10.1007/s00704-026-06600-5

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06600-5

Keywords: Gauteng, climate variability, water resources, NASA POWER, MERRA-2, CMIP6, SSP2-4.5, heatwaves, precipitation trends, Sen's slope estimator, Modified Mann-Kendall test, South Africa

Ciguatera fish poisoning is caused by ciguatoxins, potent neurotoxins produced by benthic dinoflagellates of the genus Gambierdiscus and related organisms that grow on dead coral and macroalgae in tropical reef ecosystems. The toxins accumulate in herbivorous reef fish and pass up the food web to large predatory species that are a staple of island diets. Disturbances that create fresh surfaces for dinoflagellate colonisation, such as cyclone damage to coral or sediment and nutrient pulses from land, can therefore increase the abundance of toxigenic organisms and raise the risk of poisoning. Because there is no routine test for ciguatoxin in fish sold or caught for local consumption, island communities depend heavily on their own environmental knowledge to judge when and where fish are safe to eat.

To capture that knowledge, the researchers conducted structured interviews with 37 long-term residents of Ambae, asking them to describe changes in rainfall, tropical cyclones, land use, runoff and reef conditions over their lifetimes. The interview responses were then compared with MERRA-2 reanalysis rainfall data, the Modern-Era Retrospective Analysis for Research and Applications, version 2, produced by NASA, along with tropical cyclone records for the southwest Pacific and provincial-level ciguatera case data in which Ambae cases dominate. This dual approach, pairing perceived change with measured change, is a recognised method for validating local observations and for identifying where memory, experience and instrumentation diverge.

The comparison produced a striking pattern of convergence on intensity and divergence on frequency. Residents overwhelmingly perceived that rainfall had become more intense and that cyclones had grown stronger, and these perceptions broadly matched the observed record, which shows increasing rainfall variability and rising cyclone intensity across the study period. When it came to how often cyclones strike, however, perceptions diverged from the data: community members tended to believe cyclones were becoming more frequent, while the observed records did not show a matching upward trend in frequency. The authors attribute this gap to the psychology of high-impact storms, in which memories of devastating events such as Cyclone Funa, which struck Ambae in January 2008, and Cyclone Pam in 2015, anchor perceptions of frequency around the most traumatic experiences rather than the full statistical record.

This memory effect has practical consequences. Risk perception research has repeatedly shown that experience of extreme events shapes how communities prepare, adapt and respond, and the Ambae findings reinforce the point that perceived frequency can be inflated by salient disasters even when the underlying hazard statistics remain more stable. For disaster risk managers in Vanuatu and other small island developing states, the message is that community early-warning knowledge is reliable for intensity and impact but should be complemented by objective monitoring when it comes to frequency-based planning, so that adaptation resources are allocated according to actual hazard profiles rather than recollection alone.

The second strand of the study turned to health. Using exploratory statistical analysis, the researchers examined whether annual ciguatera case counts were more closely associated with rainfall anomalies or with tropical cyclone activity. The results pointed clearly toward cyclones: years with greater cyclone activity were more strongly linked to elevated case counts than years with unusual rainfall. This is mechanistically plausible, since cyclones deliver a double disturbance, physically shattering coral structures that dinoflagellates readily colonise and dumping large pulses of terrestrial sediment, nutrients and organic material onto nearshore reefs through intensified runoff. Studies from other Pacific islands, including Rarotonga in the Cook Islands and reef systems in American Samoa and Palau, have documented similar ridge-to-reef pathways in which catchment disturbance and storm runoff degrade benthic habitat and alter the ecological conditions that favour toxigenic dinoflagellates.

The authors are careful to stress the limits of causal inference. The ciguatera case series is short, recorded only at annual resolution, and compiled at a coarse provincial scale dominated by Ambae cases, which means the analysis cannot isolate the exact timing and location of exposures or control for changes in reporting practice, health-seeking behaviour and island population movements. Ambae’s own recent history complicates the picture further: the Ambae volcano erupted in a Plinian episode during 2017 and 2018, forcing the evacuation of the entire island population and disrupting both land use and fishing patterns. Volcanic ashfall, deforestation and displacement all interact with climatic stressors in ways that a short annual case record cannot disentangle, and the researchers explicitly frame their ciguatera findings as indicative associations rather than proof of causation.

Even so, the study’s central contribution is methodological and quietly transformative. It demonstrates that community knowledge and satellite-based environmental monitoring can be systematically combined to identify plausible ridge-to-reef pathways linking climatic disturbance, land-use change, reef degradation and climate-sensitive disease risk. Earlier work by the same lead author, published in Communications Earth & Environment in 2024, showed that Indigenous knowledge in Vanuatu, combined with scientific observation, can function as an early warning system for ciguatera outbreaks. The new research strengthens that foundation by quantifying which components of community observation are most trustworthy, and by anchoring them in a 43-year environmental record that no single resident could hold in memory.

For the people of Ambae, where reef fish remain central to nutrition, livelihoods and cultural practice, the stakes of this research are immediate. Ciguatera is the most widely reported marine food-borne illness in the Pacific islands, with tens of thousands of cases recorded regionally over recent decades, and its burden falls disproportionately on communities with the least access to laboratory diagnostics. Understanding that cyclone years may seed conditions for later poisoning outbreaks gives health authorities a possible window for targeted public health messaging, and gives fishing communities a scientific echo of the traditional knowledge they already apply when deciding whether a reef is safe. As climate change continues to intensify extreme rainfall and cyclone behaviour across the southwest Pacific, integrating local observation with environmental records offers small island communities one of the few practical tools available for anticipating a hazard that hides inside an otherwise valuable food source.

The research also carries a broader message for climate adaptation science. Perception studies conducted across Africa, Asia and the Pacific have often found mismatches between what communities observe and what meteorological records show, and those mismatches are frequently used to question local knowledge. The Ambae study flips that framing, showing that local observation can be highly accurate where it matters most, for intensity and impact, and that apparent errors are themselves informative about how humans experience and remember environmental risk. In an era when monitoring infrastructure is sparse across much of the Pacific, the residents of small islands are, in effect, a distributed sensing network, and this study shows how their observations can be calibrated against, and combined with, the satellite record to protect both ecosystems and human health.

Subject of Research: Community perceptions of climate variability, land-use change and ciguatera fish poisoning risk on Ambae Island, Vanuatu, compared with observed rainfall and cyclone records

Article Title: Community perceptions and observed environmental records of climate variability, land-use change, and ciguatera fish poisoning risk on Ambae Island, Vanuatu

Article References: Rarai, A., Weber, E., Sagero, P. O., & Parsons, M. (2026). Community perceptions and observed environmental records of climate variability, land-use change, and ciguatera fish poisoning risk on Ambae Island, Vanuatu. Regional Environmental Change, 26(4), Article 193. https://doi.org/10.1007/s10113-026-02686-9

Image Credits: AI Generated

DOI: 10.1007/s10113-026-02686-9

Keywords: ciguatera fish poisoning, Vanuatu, Ambae Island, climate variability, tropical cyclones, rainfall, land-use change, coral reefs, community perception, MERRA-2, ridge-to-reef, Pacific Islands