研究目的
To present a multispectral temporal-based remote sensing technique using a modified Kalman filter for clear-air detection from geostationary visible-infrared radiometric passive measurements.
研究成果
The modified Kalman temporal filter effectively detects clear-air scenarios with high accuracy (around 90% matching with EUMETSAT cloud mask), demonstrating its applicability for global environmental monitoring and anomaly detection in multispectral satellite data.
研究不足
Potential limitations include rough characterizations of background pixel models, geographic mismatches in false alarms, reliance on thresholds that may not be universally optimal, and the absence of ground truth for validation.
1：Experimental Design and Method Selection:
A modified Kalman filter algorithm is designed to detect anomalies in spectral radiance by modeling the daily measurement cycle in clear-sky conditions. The filter uses a time-discrete approach with constant time sampling and a daily cycle, updating states based on measurements and predefined thresholds.
2：Sample Selection and Data Sources:
Data from the Meteosat Second Generation satellite's SEVIRI instrument is used, specifically five spectral channels (VIS 0.6 μm, IR 3.9 μm, 10.2 μm, 12.0 μm, 13.4 μm) over West Africa (ECOWAS countries plus Mauritania and Chad) from December 2015 to February 2016, sampled every 15 minutes.
3：6 μm, IR 9 μm, 2 μm, 0 μm, 4 μm) over West Africa (ECOWAS countries plus Mauritania and Chad) from December 2015 to February 2016, sampled every 15 minutes.
List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: Meteosat Second Generation satellite with SEVIRI radiometer, calibrated spectral measurements in specified units.
4：Experimental Procedures and Operational Workflow:
Apply the Kalman filter algorithm to the dataset, compute a-priori and a-posteriori states, detect anomalies based on residual thresholds, and validate with EUMETSAT cloud mask product.
5：Data Analysis Methods:
Inter-comparison with EUMETSAT cloud mask product using defined classes (matching, missed detection, false alarm), calculation of percentages, and temporal trend analysis.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
