Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Using sentinel-2 data to quantify the impacts of drought on crop yields at local and regional scales in Sweden

Müller, Mitro LU orcid ; Thapa, Shangharsha LU orcid ; Bouras, El houssaine LU orcid ; Olsson, Per Ola LU ; Jamali, Sadegh LU orcid ; Eklundh, Lars LU orcid and Ardö, Jonas LU orcid (2025) In Agricultural and Forest Meteorology 373.
Abstract

A causal inference framework was developed to investigate crop responses to agricultural drought by integrating meteorological data, Sentinel-2-derived data, and soil property maps. To account for crop rotation, soil, and topographical variables, propensity score matching was employed to estimate drought-induced yield losses at the field level for selected periods. The Plant Phenology Index (PPI) and the derived Total Productivity (TPROD) parameter enabled monitoring of crop development and productivity. TPROD showed high regional accuracy (R² = 0.93) and field-level accuracy for estimating crop yields (R² = 0.42–0.73, varying by crop type). The monitoring of common production crops in Sweden during the 2018 drought revealed that all... (More)

A causal inference framework was developed to investigate crop responses to agricultural drought by integrating meteorological data, Sentinel-2-derived data, and soil property maps. To account for crop rotation, soil, and topographical variables, propensity score matching was employed to estimate drought-induced yield losses at the field level for selected periods. The Plant Phenology Index (PPI) and the derived Total Productivity (TPROD) parameter enabled monitoring of crop development and productivity. TPROD showed high regional accuracy (R² = 0.93) and field-level accuracy for estimating crop yields (R² = 0.42–0.73, varying by crop type). The monitoring of common production crops in Sweden during the 2018 drought revealed that all crops had a shortened growing season, with spring-sown crops experiencing greater yield losses. The influence of soil texture variables, which act as indicators of water holding capacity, on the variability of drought-induced yield losses was assessed, and seasonal dynamics were examined, thereby improving the comprehension of the interactions among soil-plant-atmosphere dynamics at a local scale. We conclude that applying propensity score matching combined with satellite remote sensing can provide site-specific information on crop selection and timing and facilitate economically efficient irrigation planning. Nevertheless, further improvements are recommended, such as incorporating more detailed field-level data on yields and management practices, to enhance the approach's robustness and applicability for drought preparedness and adaptive agricultural management.

(Less)
Please use this url to cite or link to this publication:
author
; ; ; ; ; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Agricultural drought, Causal inference, Machine learning, Matching, Propensity score, Satellite remote sensing, Yield modelling
in
Agricultural and Forest Meteorology
volume
373
article number
110789
publisher
Elsevier
external identifiers
  • scopus:105013599856
ISSN
0168-1923
DOI
10.1016/j.agrformet.2025.110789
language
English
LU publication?
yes
id
06d58f5b-24b0-4e64-b2df-2078cd6fe77e
date added to LUP
2025-10-09 13:35:55
date last changed
2025-10-10 08:35:04
@article{06d58f5b-24b0-4e64-b2df-2078cd6fe77e,
  abstract     = {{<p>A causal inference framework was developed to investigate crop responses to agricultural drought by integrating meteorological data, Sentinel-2-derived data, and soil property maps. To account for crop rotation, soil, and topographical variables, propensity score matching was employed to estimate drought-induced yield losses at the field level for selected periods. The Plant Phenology Index (PPI) and the derived Total Productivity (TPROD) parameter enabled monitoring of crop development and productivity. TPROD showed high regional accuracy (R² = 0.93) and field-level accuracy for estimating crop yields (R² = 0.42–0.73, varying by crop type). The monitoring of common production crops in Sweden during the 2018 drought revealed that all crops had a shortened growing season, with spring-sown crops experiencing greater yield losses. The influence of soil texture variables, which act as indicators of water holding capacity, on the variability of drought-induced yield losses was assessed, and seasonal dynamics were examined, thereby improving the comprehension of the interactions among soil-plant-atmosphere dynamics at a local scale. We conclude that applying propensity score matching combined with satellite remote sensing can provide site-specific information on crop selection and timing and facilitate economically efficient irrigation planning. Nevertheless, further improvements are recommended, such as incorporating more detailed field-level data on yields and management practices, to enhance the approach's robustness and applicability for drought preparedness and adaptive agricultural management.</p>}},
  author       = {{Müller, Mitro and Thapa, Shangharsha and Bouras, El houssaine and Olsson, Per Ola and Jamali, Sadegh and Eklundh, Lars and Ardö, Jonas}},
  issn         = {{0168-1923}},
  keywords     = {{Agricultural drought; Causal inference; Machine learning; Matching; Propensity score; Satellite remote sensing; Yield modelling}},
  language     = {{eng}},
  publisher    = {{Elsevier}},
  series       = {{Agricultural and Forest Meteorology}},
  title        = {{Using sentinel-2 data to quantify the impacts of drought on crop yields at local and regional scales in Sweden}},
  url          = {{http://dx.doi.org/10.1016/j.agrformet.2025.110789}},
  doi          = {{10.1016/j.agrformet.2025.110789}},
  volume       = {{373}},
  year         = {{2025}},
}