Decoupling Crop Stressors in the Bărăgan Plain: A Multi-Sensor Remote Sensing Framework

Abstract

Abstract With agriculture moving to a more performant and autonomous-driven sustainability, advanced digital monitoring is becoming the key to resilient land management. This study investigates hydro-climatic conditions and urban air pollution in 2024 in the Bărăgan Plain, Romania’s primary grain-producing region. Our approach uses the Google Earth Engine (GEE) platform, high-resolution multispectral imagery (Sentinel-2), and atmospheric trace gas data from Sentinel-5P TROPOMI to map spatiotemporal interactions between agricultural health and urban air pollution. To find the most robust approach we compared three machine learning algorithms: Multiple Linear Regression, Random Forest (RF), and Extreme Gradient Boosting (XGBoost), the Enhanced Vegetation Index (EVI) was used as a predictor of the Normalized Difference Vegetation Index (NDVI). Although the models that included the EVI achieved higher accuracy (R2 of 93%), the non-EVI Random Forest model performed better (R 2 of 86%) and revealed moisture availability (NDWI) as the primary regulator of crop vigor, with an importance of 81.8%. To isolate the effect of atmospheric chemistry alone, spatial residuals from the optimized RF model were extracted and plotted. Negative residuals that produced anomalies pointed to a unique type of crop stress in which the plants were underperforming even when water was adequate. These anomalies spatially align the urban pollution plume trajectories. Even if the NO2 has a rapid distance-decay effect, the secondary pollutants (O3) reach over the agricultural land through the photochemical titration effect. Our results show that while water determines regional agricultural baselines, atmospheric chemistry from urban sources can independently cause significant crop stress even in the absence of drought. This study provides a robust proof of concept on the separation of climatic and anthropogenic stressors and lays a basic pathway for a multi-sensor diagnostic framework in the domain of remote sensing.

Plant phenotyping relevance

衛星マルチセンサー画像と機械学習を統合し、作物の活力・ストレスを推定する診断フレームワークが研究の中心であり、単なる農業実験のルーチン測定ではない。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.