Divergent crop mapping accuracies across different field types in smallholder farming regions
Huang, Xin ; Vrieling, Anton ; Dou, Yue ; Li, Xueying LU
and Nelson, Andrew
(2025)
In International Journal of Applied Earth Observation and Geoinformation
139.
- Abstract
- Accurately mapping crop types in smallholder farming regions is crucial for monitoring crop dynamics and estimating production but remains challenging, especially over large extents. Remote sensing based crop mapping studies in smallholder farming regions often focus on major crops and the challenge of mapping small fields. However, minor but possibly emerging crops are often overlooked, nor is the impact of environmental factors, such as water stress, on mapping accuracy considered. This study addresses these gaps by categorizing crop fields into different types and assessing mapping accuracy for both major and minor crops within each field type. We first categorized crop fields into four field types (big/small fields with/without water... (More)
- Accurately mapping crop types in smallholder farming regions is crucial for monitoring crop dynamics and estimating production but remains challenging, especially over large extents. Remote sensing based crop mapping studies in smallholder farming regions often focus on major crops and the challenge of mapping small fields. However, minor but possibly emerging crops are often overlooked, nor is the impact of environmental factors, such as water stress, on mapping accuracy considered. This study addresses these gaps by categorizing crop fields into different types and assessing mapping accuracy for both major and minor crops within each field type. We first categorized crop fields into four field types (big/small fields with/without water stress) based on field size and the shortwave infrared water stress index (SIWSI) derived from Sentinel-2 (S2). Crop mapping accuracies for different field types and crops (maize as a major crop and soybean as a minor crop) were compared at pixel-based (PB) and object-based (OB) levels using random forest classification applied to S2 and two additional publicly accessible multispectral datasets (PlanetScope with four bands (PS4) and eight bands (PS8)). The season-averaged SIWSI (SIWSImean) provided a useful categorization of field types, as it is sensitive to mapping accuracy and is independent from field size. Based on S2 data, big fields without water stress can be most accurately mapped (F1-score = 0.89 for maize and 0.85 for soybean), followed by small fields without water stress (0.85 and 0.68) and big fields with water stress (0.82 and 0.59), while small fields with water stress are the most challenging type (0.77 and 0.37). Despite that the use of PS8 data with higher spatial resolution and OB classification improved mapping accuracy for small soybean fields with water stress, limitations to map such fields remain (F1-score < 0.50). This study provides a new perspective on crop type mapping in smallholder farming regions by using a simple and relevant categorization of field types and offers valuable insights on potentials and limitations for large-scale crop type mapping using machine learning algorithms. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/4ddaf5f1-1ee8-4a5f-a203-9ec518e58895
- author
- Huang, Xin
; Vrieling, Anton
; Dou, Yue
; Li, Xueying
LU
and Nelson, Andrew
- organization
- publishing date
- 2025-05
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Crop classification, Smallholder farming, Balanced random forest, Object-based classification, PlanetScope
- in
- International Journal of Applied Earth Observation and Geoinformation
- volume
- 139
- article number
- 104559
- publisher
- Elsevier
- external identifiers
-
- scopus:105003738659
- ISSN
- 1569-8432
- DOI
- 10.1016/j.jag.2025.104559
- language
- English
- LU publication?
- yes
- id
- 4ddaf5f1-1ee8-4a5f-a203-9ec518e58895
- date added to LUP
- 2025-07-13 17:29:18
- date last changed
- 2025-08-12 12:59:43
@article{4ddaf5f1-1ee8-4a5f-a203-9ec518e58895,
abstract = {{Accurately mapping crop types in smallholder farming regions is crucial for monitoring crop dynamics and estimating production but remains challenging, especially over large extents. Remote sensing based crop mapping studies in smallholder farming regions often focus on major crops and the challenge of mapping small fields. However, minor but possibly emerging crops are often overlooked, nor is the impact of environmental factors, such as water stress, on mapping accuracy considered. This study addresses these gaps by categorizing crop fields into different types and assessing mapping accuracy for both major and minor crops within each field type. We first categorized crop fields into four field types (big/small fields with/without water stress) based on field size and the shortwave infrared water stress index (SIWSI) derived from Sentinel-2 (S2). Crop mapping accuracies for different field types and crops (maize as a major crop and soybean as a minor crop) were compared at pixel-based (PB) and object-based (OB) levels using random forest classification applied to S2 and two additional publicly accessible multispectral datasets (PlanetScope with four bands (PS4) and eight bands (PS8)). The season-averaged SIWSI (SIWSImean) provided a useful categorization of field types, as it is sensitive to mapping accuracy and is independent from field size. Based on S2 data, big fields without water stress can be most accurately mapped (F1-score = 0.89 for maize and 0.85 for soybean), followed by small fields without water stress (0.85 and 0.68) and big fields with water stress (0.82 and 0.59), while small fields with water stress are the most challenging type (0.77 and 0.37). Despite that the use of PS8 data with higher spatial resolution and OB classification improved mapping accuracy for small soybean fields with water stress, limitations to map such fields remain (F1-score < 0.50). This study provides a new perspective on crop type mapping in smallholder farming regions by using a simple and relevant categorization of field types and offers valuable insights on potentials and limitations for large-scale crop type mapping using machine learning algorithms.}},
author = {{Huang, Xin and Vrieling, Anton and Dou, Yue and Li, Xueying and Nelson, Andrew}},
issn = {{1569-8432}},
keywords = {{Crop classification; Smallholder farming; Balanced random forest; Object-based classification; PlanetScope}},
language = {{eng}},
publisher = {{Elsevier}},
series = {{International Journal of Applied Earth Observation and Geoinformation}},
title = {{Divergent crop mapping accuracies across different field types in smallholder farming regions}},
url = {{http://dx.doi.org/10.1016/j.jag.2025.104559}},
doi = {{10.1016/j.jag.2025.104559}},
volume = {{139}},
year = {{2025}},
}