BEGIN:VCALENDAR VERSION:2.0 PRODID:-//pretalx//talks.staging.osgeo.org//foss4g-europe-2024-academic-tra ck//talk//WLUUSK BEGIN:VTIMEZONE TZID:EET BEGIN:STANDARD DTSTART:20000101T000000 RRULE:FREQ=YEARLY;BYMONTH=1;UNTIL=20001231T220000Z TZNAME:EET TZOFFSETFROM:+0200 TZOFFSETTO:+0200 END:STANDARD BEGIN:STANDARD DTSTART:20021027T050000 RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10 TZNAME:EET TZOFFSETFROM:+0300 TZOFFSETTO:+0200 END:STANDARD BEGIN:DAYLIGHT DTSTART:20020331T040000 RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3 TZNAME:EEST TZOFFSETFROM:+0200 TZOFFSETTO:+0300 END:DAYLIGHT END:VTIMEZONE BEGIN:VEVENT UID:pretalx-foss4g-europe-2024-academic-track-WLUUSK@talks.staging.osgeo.or g DTSTART;TZID=EET:20240703T160000 DTEND;TZID=EET:20240703T163000 DESCRIPTION:Soil Erosion\, the displacement of topsoil by water and wind\, poses a significant threat to global land health\, impacting food security \, water quality\, climate change\, and ecosystem stability. Earth Observa tion (EO) and remote sensing technologies play a crucial role in monitori ng and assessing soil erosion\, offering valuable spatial and temporal dat a for informed decision-making. This paper applied three (3) Machine Learn ing (ML) models\, namely the XGBoost classifier\, LightGBM classifier\, an d CatBoost classifier to perform soil erosion classification in the Europe an Union (EU) region. The data used in this study were sourced from Kaggle \, a huge repository of community-published machine learning models and da ta\, and it includes several EO data namely the Landsat 7 seasonal Analysi s Ready Data (ARD)\, BioClim v1.2 historical (1981-2010) average climate d ata using the CHLSA classification system\, annual MODIS EVI data\, climat ic variables (water vapour\, monthly snow probability\, annual MODIS LST i n daytime or night time\, annual CHELSA rainfall V2.1)\, Human footprint ( Hengl et al.\, 2023)\, Land cover\, Landform and landscape parameters (Hen gl\, 2018)\, Lithology (Hengl\, 2018). The dataset has a total of 3754 sam ple points and 139 features. A detailed description of the dataset feature s can be found here.\n\nDuring the Exploratory Data Analysis (EDA) process \, the visual relationship between the Landsat bands and the target variab le (erosion category)\, revealed that the Near Infrared (NIR) \, Short-Wav e Infrared I (SWIR1)\, Short-Wave Infrared II (SWIR2)\, and Thermal bands were effective in differentiating between the various erosion categories\, compared to other bands. This insight gave direction in the feature engin eering process. As suggested by Puente et al. (2019)\, vegetation indices could prove effective in predicting soil erosion. Consequently\, we comput ed various vegetation indices such as the Normalised Difference Water Inde x (NDWI)\, Normalised Difference Infrared Index (NDII)\, and Shortwave Inf rared Water Stress Index (SIWSI) as well as applied the Tasseled Cap Tran sformation which includes Brightness\, Wetness and Greenness\, to augment the features. To capture textural variations of each pixel location\, elev ation\, and slope-based measures were computed. The Topographic Position I ndex (TPI) was computed for each position using a 100\,000-metre radius\, calculating the mean elevation of points within the radius and subtracting it from each point elevation within the radius. Other features computed w ere the Topographic Wetness Index (TWI)\, Aspect\, LS-Factor\, and Stream Power Index (SPI) which reflects the erosive power of streams. Leveraging the thermal band\, Land Surface Temperature (LST) was derived. As noted by Ghosal (2021)\, combining LST with temporal data can identify regions vul nerable to soil erosion.\n\nThe development of these models incorporated S cikit-Learn Recursive Feature Elimination (RFE) in the preliminary feature selection process using the XGBoost model as the estimator. The goal of R FE is to return “n” features by training the model on all features\, r ank all features by importance\, and remove the least important features u ntil “n” features remain. The RFE “n” features were set to 200. Af terward\, an XGBoost model was trained with the 200 features\, and Scikit- Learn’s Randomised Search CV was employed to optimise its hyperparameter s\, leading to an improved F-1 score for the XGBoost classifier. Using the XGBoost’s classifier feature importance ranking\, the top 155 features were selected for use in the final ensemble model for predictions. To prov ide a more reliable estimate of the performance of the training model\, Sc ikit-Learn's Stratified KFold was implemented with n_splits set to 5 and t he erosion category as the stratification variable. By using stratified KF old\, a balanced class representation in each fold during training was ach ieved. For modelling of erosion categories\, an ensemble voting classifier combined predictions from three optimised gradient boosting models (XGBoo st\, LightGBM\, CatBoost) using a "soft" voting scheme. This approach aime d to improve accuracy and reduce overfitting compared to individual models . The confusion matrix was used to evaluate the ensemble's performance\, c onsidering precision\, recall\, and F1-score metrics. These metrics assess the model's ability to correctly identify positive and negative cases\, w ith a higher F1 score indicating better overall performance.\n\nThe weight ed F-1 score reached 0.86\, and the weighted precision and recall were 0.8 6 and 0.86 respectively\, indicating that the proposed method using variou s EO data to predict soil erosion categories (No Gully/badland\, Gully\, B adland\, Landslides) displayed good performance. Specifically\, No Gully/b adland (0.89\, 0.91) and Landslides(1.00\, 1.00) had higher precision and recall values\, which means that the model can correctly identify areas th at fall within these erosion categories with low false positives and false negatives. The Badland(0.49) had the least recall value indicating that t he model could not identify a substantial amount of this category. \n\nAc cording to the Feature Importance analysis\; Year\, Latitude Coordinates\, Topographic Wetness Index (TWI)\, Longitude Coordinates\, Maximum Fractio n of Absorbed Photosynthetically Active Radiation (FAPAR)\, Minimum Annual Water Vapour\, Mean of Slope\, Weighted Difference Vegetation Index (WDV I)\, Normalised Difference Snow Index (NDSI) and Standard Deviation of Slo pe emerged as the top ten (10) factors influencing soil erosion. Indicatin g that Topographic factors and vegetation indices were important for predi cting soil erosion. The year was the most important feature\, which shows that temporal trends have a huge impact in predicting soil erosion.\n\nIn conclusion\, this project successfully explored the potential of ensemble learning and EO data for classifying soil erosion\, highlighting its promi sing role in addressing this crucial environmental issue. The proposed fra mework indicates that Topographic indices like the TWI and vegetation indi ces like the WDVI hold valuable information for predicting soil erosion. F urthermore\, band combinations using near-infrared (NIR)\, SWIR1\, SWIR2\, and thermal bands can significantly improve the classification of soil er osion categories. Crucially\, EO data like digital elevation models (DEMs) and Analysis Ready Landsat data serve as the foundation for accurate soil erosion prediction. The proposed approach to incorporate multi-temporal E O data offers exciting prospects for even more accurate soil erosion class ification. DTSTAMP:20260509T090313Z LOCATION:Omicum SUMMARY:Mapping Soil Erosion Classes using Remote Sensing Data and Ensemble Models - Ayomide Oraegbu\, Emmanuel Jolaiya URL:https://talks.staging.osgeo.org/foss4g-europe-2024-academic-track/talk/ WLUUSK/ END:VEVENT END:VCALENDAR