Crop growth and yield monitoring in smallholder agricultural systems: a multi-sensor data fusion approach

Research output: ThesisPhD Thesis - Research UT, graduation UT

33 Downloads (Pure)

Abstract

Smallholder agricultural systems are highly vulnerable to production risks posed by the intensification of extreme weather events such as drought and flooding, soil degradation, pests, lack of access to agricultural inputs, and political instability. Monitoring the spatial and temporal variability of crop growth and yield is crucial for farm management, national-level food security assessments, and famine early warning. However, agricultural monitoring is difficult in fragmented agricultural landscapes because of scarcity and uncertainty of data to capture small crop fields. Traditional pre- and post-harvest crop monitoring and yield estimation based on fieldwork is costly, slow, and can be unrepresentative of heterogeneous agricultural landscapes as found in smallholder systems in sub-Saharan Africa. Devising accurate and timely crop phenology detection and yield estimation methods can improve our understanding of the status of crop production and food security in these regions.
Satellite-based Earth observation (EO) data plays a key role in monitoring the spatial and temporal variability of crop growth and yield over large areas. The small field sizes and variability in management practices in fragmented landscapes requires high spatial and high temporal resolution EO data. This thesis develops and demonstrates methods to investigate the spatiotemporal variability of crop phenology detection and yield estimation using Landsat and MODIS data fusion in smallholder agricultural systems in the Lake Tana sub-basin of Ethiopia. The overall aim is to further broaden the application of multi-sensor EO data for crop growth monitoring in smallholder agricultural systems.
The thesis addressed two important aspects of crop monitoring applications of EO data: phenology detection and yield estimation. First, the ESTARFM data fusion workflow was modified based on local knowledge of crop calendars and land cover to improve crop phenology monitoring in fragmented agricultural landscapes. The approach minimized data fusion uncertainties in predicting temporal reflectance change of crops during the growing season and the reflectance value of fused data was comparable to the original Landsat image reserved for validation. The main sources of uncertainty in data fusion are the small field size and abrupt crop growth changes between the base and
vii
prediction dates due to flooding, weeding, fertiliser application, and harvesting. The improved data fusion approach allowed us to determine crop phenology and estimate LAI more accurately than both the standard ESTARFM data fusion method and when using MODIS data without fusion. We also calibrated and validated a dynamic threshold phenology detection method using maize and rice crop sowing and harvest date information. Crop-specific phenology determined from data fusion minimized the mismatch between EO-derived phenometrics and the actual crop calendar. The study concluded that accurate phenology detection and LAI estimation from Landsat–MODIS data fusion demonstrates the feasibility of crop growth monitoring using multi-sensor data fusion in fragmented and persistently cloudy agricultural landscapes.
Subsequently, the validated data fusion and phenology detection methods were implemented to understand crop phenology trends from 2000 to 2020. These trends are often less understood in smallholder agricultural systems due to the lack of high spatial resolution data to distinguish crops from the surrounding natural vegetation. Trends based on Landsat–MODIS fusion were compared with those detected using MODIS alone to assess the contribution of data fusion to discern crop phenometric change. Landsat and MODIS fusion discerned crop and environment-specific trends in the magnitude and direction of crop phenology change. The results underlined the importance of high spatial and temporal resolution EO data to capture environment-specific crop phenology change, which has implications in designing adaptation and crop management practices in these regions.
The second important aspect of the crop monitoring problem addressed in this thesis is improving crop yield estimation in smallholder agricultural systems. The large input requirements of crop models and lack of spatial information about the heterogeneous crop-growing environment and agronomic management practices are major challenges to the accurate estimation of crop yield. We assimilated leaf area index (LAI) and phenology information from Landsat–MODIS fusion in a crop model (simple algorithm for yield estimation: SAFY) to obtain reasonably reliable crop yield estimates. The SAFY model is sensitive to the spatial and temporal resolution of the calibration input LAI, phenology information, and the effective light use efficiency (ELUE) parameter, which needs accurate field level inputs during model
viii
optimization. Assimilating fused EO-based phenology information minimized model uncertainty and captured the large management and environmental variation in smallholder agricultural systems.
In the final research chapter of the thesis, we analysed the contribution of assimilating LAI at different phenological stages. The frequency and timing of LAI observations influences the retrieval accuracy of the assimilating LAI in crop growth simulation models. The use of (optical) EO data to estimate LAI is constrained by limited repeat frequency and cloud cover, which can reduce yield estimation accuracy. We evaluated the relative contribution of EO observations at different crop growth stages for accurate calibration of crop model parameters. We found that LAI between jointing and grain filling has the highest contribution to SAFY yield estimation and that the distribution of LAI during the key development stages was more useful than the frequency of LAI to improve yield estimation. This information on the optimal timing of EO data assimilation is important to develop better in-season crop yield forecasting in smallholder systems.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • University of Twente
  • Faculty of Geo-Information Science and Earth Observation
Supervisors/Advisors
  • Nelson, Andy, Supervisor
  • Marshall, Michael Thomas, Supervisor
  • Mengistu, D.A., Supervisor, External person
Award date22 May 2024
Place of PublicationEnschede
Publisher
Print ISBNs978-90-365-6080-1
Electronic ISBNs978-90-365-6080-8
DOIs
Publication statusPublished - 22 May 2024

Fingerprint

Dive into the research topics of 'Crop growth and yield monitoring in smallholder agricultural systems: a multi-sensor data fusion approach'. Together they form a unique fingerprint.

Cite this