Joseph Kagotho Muriithi

This study presents a novel multitask deep learning framework that integrates models designed for simultaneous, multi‐horizon forecasting of Hp60 and Disturbance storm time (Dst) indices with lead times up to 3 hr. Using a Long Short‐Term Memory architecture, the models capture both mid‐latitude variability and the ring current dynamics. Input features include interplanetary magnetic field (IMF) parameters, solar wind plasma data, and historical geomagnetic indices. Model interpretability is addressed through permutation‐based feature importance analysis, providing physical context to the machine learning predictions. The models demonstrate strong predictive performance across all forecast horizons. For the Dst index, they achieve a Root Mean Square Error (RMSE) of 3.81 nT and Pearson correlation coefficient of 0.98 at 1‐hr horizon ( t + 1 hr), remaining robust (RMSE = 9.01 nT, = 0.91) at 3‐hr horizon ( t + 3 hr). More importantly, the models perform well under both quiet and disturbed geomagnetic conditions. However, while they accurately capture storm onset and recovery phases, they underestimate extreme storm intensities. Hp60 predictability is primarily associated with related mid‐latitude indices, such as planetary K (Kp) and ap, that are derived from similar observatory networks and respond to the same underlying geomagnetic disturbances. Dst forecasting is largely influenced by its own historical values (persistence). Southward IMF is also significantly influential for forecasting both indices during disturbed periods, showing that the models learn physically consistent storm‐time behavior. The multitask approach offers a scalable and efficient pathway for future space weather models by leveraging shared representations of common solar wind drivers.

We investigate the ionospheric response to the intense St. Patrick's Day geomagnetic storm of 17 March 2015 (minimum Dst ≈ −223 nT) using Global Navigation Satellite System (GNSS)–derived Total Electron Content (TEC) observations from 11 East African stations: Addis Ababa‐ADIS (9.04°N, 38.77°E), Arusha‐ARSH (3.38°S, 36.69°E), Dodoma‐DODM (6.19°S, 35.75°E), Eldoret‐MOIU (0.29°N, 35.29°E), Malindi‐MAL2 (2.99°S, 40.19°E), Mbarara‐MBAR (0.61°S, 30.65°E), and Mbeya‐MBEY (9.03°N, 38.74°E, Asab‐ASAB (13.06°N. 43.65°E), Ginr‐GINR (7.15°N, 40.71°E), Bahir Dar‐BDMT (11.60°N, 37.36°E), Debarek‐DEBK (13.15°N, 37.89°E), Assosa‐ASOS (10.05°N, 34.55°E). Instrumental biases were removed, data were screened for satellite elevation angles ≥40°, and analyzed using detrended TEC (dTEC) and Morlet continuous wavelet transform to identify storm‐time ionospheric disturbances across multiple temporal scales. The storm main phase was characterized by prolonged southward z‐component of interplanetary magnetic field (IMF‐Bz) (<−20 nT), enhanced solar wind speeds (>600 kms ⁻¹ ), and elevated y‐component of the interplanetary electric field (IEF‐Ey), indicating strong solar wind–magnetosphere coupling. These conditions drove significant ionospheric variability, with the largest dTEC amplitudes (>±4 TECU) observed near the Equatorial Ionization Anomaly (EIA), particularly at ADIS and MAL2. Wavelet analysis revealed the coexistence of medium‐scale traveling ionospheric disturbances (MSTIDs; ∼0.25–1 hr), large‐scale travelling ionospheric disturbances (LSTIDs; ∼1–4 hr), and short‐period fluctuations (<0.25 hr). Disturbances were most pronounced during the storm main and early recovery phases and exhibited clear latitudinal dependence, with enhanced activity near the EIA crest. Post‐sunset ionospheric irregularities, interpreted as equatorial plasma bubbles (EPBs), were identified through sharp dTEC depletions and enhanced short‐period wavelet power. EPBs occurred on most days but showed increased frequency, earlier onset, and wider latitudinal extent during the storm, consistent with intensified pre‐reversal enhancement and storm‐time electric field penetration. The temporal association between MSTIDs and EPB onset suggests a potential seeding role of atmospheric wave activity. These results demonstrate that ionospheric disturbances over East Africa are strongly controlled by the interplay between storm‐time electrodynamics and equatorial processes. The study highlights the effectiveness of combined dTEC and wavelet analysis for resolving multiscale ionospheric variability and provides important insights into space weather impacts on GNSS performance in equatorial regions.

Rice ( Oryza sativa L.) crop monitoring in Africa requires efficient retrieval of biophysical parameters such as leaf area index (LAI) and leaf chlorophyll content (LCC), which are essential for crop stress detection and nutrient management. This study aimed at developing a parsimonious modeling approach that overcomes the limitations of existing single‐ and multi‐target models. We developed a novel multi‐output framework that integrates Pareto optimization with adaptive stopping rules to simultaneously retrieve LAI and LCC from Sentinel‐2 imagery. Model performance was evaluated using in situ data ( n = 177) collected from rice farms in Kenya's Ahero irrigation scheme using coefficient of determination ( R ² ) and root mean square error (RMSE) metrics. The parsimonious XGBoost (eXtreme gradient boosting) model trained on 11 vegetation indices achieved high accuracies ( R ² = 0.76 and RMSE = 0.766 m ² m ^{− 2} for LAI; R ² = 0.74 and RMSE = 3.590 µg cm ^{− 2} for LCC). These results represent gains of up to 32% for LAI and 47% for LCC compared to prior Sentinel‐2‐based multi‐target retrieval approaches. Notably, the model achieved these improvements with a 54% reduction in vegetation indices and a 28% smaller model size relative to a 24‐vegetation indices baseline model. Therefore, our innovative design results in improved stability, generalization, and computational efficiency. By reducing complexity while improving accuracy, the framework offers a practical pathway for scalable crop monitoring and productivity improvement using freely available satellite data.