Soil moisture prediction is a key parameter for effective irrigation scheduling and water use efficiency. However, accurate long-term prediction remains challenging, as most existing models excel in short- to medium-term prediction but struggle to capture the complex temporal dependencies and non-linear interactions of soil moisture variables over extended horizons. This study proposes a hybrid soil moisture prediction method, integrating a long short-term memory (LSTM) network and extreme gradient boosting (XGBoost) model for multistep soil moisture prediction at 24 h, 72 h, and 168 h horizons. The LSTM captures temporal dependencies and extracts high-level features from the dataset, which are then used by XGBoost for final predictions. The study uses real-world data from the D.A.T.A (Demonstrating Applied Technology in Agriculture) research farm at ABAC (Abraham Baldwin Agricultural College) Tifton, GA, USA, utilizing watermark soil moisture sensors and weather station’s data installed on the farm. Results show that the proposed method outperforms other hybrid models, achieving R² values of 98.67%, 98.54%, and 98.56% for 24, 72, and 168 h predictions, respectively. The study findings highlight that LSTM-XGBoost offers a precise long-term soil moisture prediction, making it a practical tool for real-time irrigation scheduling, enhancing water use efficiency in precision agriculture.

Soil moisture is a key parameter in agriculture. Its accurate prediction is essential for effective irrigation scheduling and water use efficiency. This study introduces a hybrid approach integrating Long Short-Term Memory (LSTM) network and Extreme Gradient Boosting (XGBoost) model for multistep soil moisture prediction for (24 hours, 72 hours, and 168 hours) ahead. The LSTM captures temporal dependencies and extracts high-level features from the dataset, while XGBoost uses these features to make final predictions. The proposed method was trained and evaluated on a real-world data from the D.A.T.A (Demonstrating Applied Technology in Agriculture) research farm at ABAC (Abraham Baldwin Agricultural College) Tifton, GA, USA, utilizing watermark soil moisture sensors and weather station’s data installed in the farm. Experimental results show that, the pro-posed method outperforms standalone models like LSTM, XGBoost, Gradient Boosting (GB), Extra Trees (ET) and others. It achieved R² values of 0.9867, 0.9854, and 0.9856 for 24, 72 and 168-hour predictions respectively. These results demonstrate the effectiveness of LSTM in extracting high-level temporal features, and XGBoost in making final predictions. The proposed hybrid model offers precise soil moisture prediction, making it a practical tool for real-time irrigation scheduling and enhancing water use efficiency in precision agriculture.

Localized temperature (T) and relative humidity (RH) in bulk peanut warehouses can differ from warehouse-average conditions, limiting single-point monitoring. This study retrospectively evaluated a temporal-leakage-controlled sensor-network workflow for forecasting warehouse microclimate and reproducing rule-defined advisory states. Environmental records from 18 locations in a commercial-scale in-shell peanut warehouse were harmonized to 15 min intervals from 22 October 2025 to 16 March 2026. Persistence, a leakage-controlled 24 h seasonal-naive baseline, a development-selected persistence-seasonal blend, and an attention-based long short-term memory (LSTM) benchmark were compared using blocked train-core, internal-development, and final-test partitions separated by 48 h embargoes. Final-test metrics used strict raw-observed targets and aggregated errors across all 192 forecast leads over the full 48 h horizon. Seasonal naive had the lowest RH mean absolute error (MAE; 4.15 percentage points), the blend had the lowest T MAE (1.92 °C), and persistence had the lowest dew-point MAE (1.58 °C); attention-LSTM had higher MAE than all temporal baselines. Across 13 non-overlapping 48 h backtests, the blend achieved 80.2% advisory-state agreement, 83.4% Danger recall, and an 8.65% false-safe rate. Final-test raw support covered 11 of 18 locations. The workflow is a forecasting and advisory-screening benchmark, not validated automated control or direct prediction of kernel moisture, fungal activity, aflatoxin, or peanut quality. The findings provide a basis for testing forecast-assisted inspection prioritization, with particular attention to the less predictable upper air void.

Effective irrigation advisory requires accurate prediction of how soil water tension (SWT) changes across the soil profile and over multiple forecast horizons. This study developed a depth-specific, multi-horizon machine-learning (ML) framework for forecasting SWT at 8-, 16-, and 24-inch depths from one to seven days ahead using field-deployed IoT Watermark sensors, sensor-quality indicators, soil-profile features, historical observed weather, and future realized-weather information. A progressive ablation design compared four model versions: V1A used current and historical SWT with sensor-quality variables; V1B added soil-profile dynamics; V2A added historical observed RealmFive weather; and V2B added future realized-weather summaries derived from the Open-Meteo historical archive. Random Forest, Extra Trees, Gradient Boosting, and XGBoost were evaluated independently across 21 depth–horizon forecasting tasks using chronological training, validation, and test periods. Weather information produced the largest improvement in forecasting accuracy. Mean RMSE decreased from 10.30 kPa for V1A to 9.56 kPa for V2A and 9.05 kPa for V2B, representing a 12.1% reduction relative to the sensor-only baseline. Mean MAE decreased from 6.58 to 5.88 kPa, while mean R² increased from 0.385 to 0.518. Prediction error increased with forecast horizon, and performance varied across sensor depths, supporting independent depth–horizon model selection. The forecasts were subsequently integrated into a prototype irrigation-advisory application using days after planting (DAP)-dependent depth weights and threshold-based decision rules. The framework demonstrates the technical integration of IoT sensing, weather-informed forecasting, and irrigation-advisory delivery, while field trials remain necessary to evaluate agronomic outcomes.