Can Satellites Predict Yield? Ensemble Machine Learning and Statistical Analysis of Sentinel-2 Imagery for Processing Tomato Yield Prediction

In this paper, we propose an innovative approach for robust prediction of processing tomato yield using open-source AutoML techniques and statistical analysis. Sentinel-2 satellite imagery was deployed to obtain values of five (5) selected vegetation indices (VIs) during the growing season of 2021 (April to September) at 5-day intervals. Actual recorded yields were collected across 108 fields, corresponding to a total area of 410.10 ha of processing tomato in central Greece, to assess the performance of Vis at different temporal scales. In addition, VIs were connected with the crop phenology to establish the annual dynamics of the crop. The highest Pearson coefficient (r) values occurred during a period of 80 to 90 days, indicating the strong relationship between the VIs and the yield. Specifically, RVI presented the highest correlation values of the growing season at 80 (r = 0.72) and 90 days (r = 0.75), while NDVI performed better at 85 days (r = 0.72). This output was confirmed by the AutoML technique, which also indicated the highest performance of the VIs during the same period, with the values of the adjusted R² ranging from 0.60 to 0.72. The most precise results were obtained with the combination of ARD regression and SVR, which was the most successful combination for building an ensemble (adj. R² = 0.67 ± 0.02).

Fire severity and soil erosion susceptibility mapping using multi-temporal Earth Observation data: The case of Mati fatal wildfire in Eastern Attica, Greece

In recent years, forest fires have increased in terms of frequency, extent and intensity, especially in Mediterranean countries. Climate characteristics and anthropogenic disturbances lead forest environments to display high vulnerability to wildfires, with their sustainability being threatened by the loss of vegetation, changes on soil properties, and increased soil loss rates. Moreover, wildfires are a great threat to property and human life, especially in Wildland-Urban Interface (WUI) areas. In light of the impacts and trends mentioned above, this study aims to assess the impact of the Mati, Attika wildfire on soil erosion. The event caused 102 fatalities, inducing severe consequences to the local infrastructure network; economy; and natural resources. As such, the Revised Universal Soil Loss Equation (RUSLE) was implemented (pre-; post-fire) at the Rafina, Attika watershed encompassing the Mati WUI. Fire severity was evaluated based on the Normalized Burn Ratio (NBR). This index was developed utilizing innovative remotely sensed Earth Observation data (Sentinel-2). The high post-fire values indicate the fire's devastating effects on vegetation loss and soil erosion. A critical "update" was also made to the CORINE Land Cover (CLC) v. 2018, by introducing a new land use class namely "Urban Forest", in order to distinguish the WUI configuration. Post-fire erosion rates are notably higher throughout the study area (4.53-5.98 t ha^-1 y^-1), and especially within the WUI zone (3.75-18.58 t ha^-1 y^-1), while newly developed and highly vulnerable cites now occupy the greater Mati area. Furthermore, archive satellite data (Landsat-5) revealed how the repeated (historical) wildfires have ultimately impacted vegetation recovery and erosional processes. To our knowledge this is the first time that RUSLE is used to simulate soil erosion at a WUI after a fire event, at least at a Mediterranean basin. The realistic results attest that the model can perform well at such diverse conditions, providing a solid basis for soil loss estimation and identification of high-risk erosion areas.

The Significance of Land Cover Delineation on Soil Erosion Assessment

The study aims to evaluate the significance of land cover delineation on soil erosion assessment. To that end, RUSLE (Revised Universal Soil Loss Equation) was implemented at the Upper Acheloos River catchment, Western Central Greece, annually and multi-annually for the period 1965-92. The model estimates soil erosion as the linear product of six factors (R, K, LS, C, and P) considering the catchment's climatic, pedological, topographic, land cover, and anthropogenic characteristics, respectively. The C factor was estimated using six alternative land use delineations of different resolution, namely the CORINE Land Cover (CLC) project (2000, 2012 versions) (1:100,000), a land use map conducted by the Greek National Agricultural Research Foundation (NAGREF) (1:20,000), a land use map conducted by the Greek Payment and Control Agency for Guidance and Guarantee Community Aid (PCAGGCA) (1:5,000), and the Landsat 8 16-day Normalized Difference Vegetation Index (NDVI) dataset (30 m/pixel) (two approximations) based on remote sensing data (satellite image acquired on 07/09/2016) (1:40,000). Since all other factors remain unchanged per each RUSLE application, the differences among the yielded results are attributed to the C factor (thus the land cover pattern) variations. Validation was made considering the convergence between simulated (modeled) and observed sediment yield. The latter was estimated based on field measurements conducted by the Greek PPC (Public Power Corporation). The model performed best at both time scales using the Landsat 8 (Eq. 13) dataset, characterized by a detailed resolution and a satisfactory categorization, allowing the identification of the most susceptible to erosion areas.