A single satellite image can show where a crop looks different. A sequence of images can reveal when that difference appeared, whether it persists, and how it relates to the growing season. That distinction is central to useful agricultural remote sensing. Looking back at 2023, the important question is how a monitoring service can turn repeated observations into a dependable account of change that an agronomist, plantation manager or researcher can investigate.
Cloud computing makes large image collections easier to process, but processing capacity alone does not produce agricultural understanding. Field boundaries, crop calendars, quality checks and observations from the ground remain essential. A practical satellite time-series workflow treats each measurement as evidence with a date and a degree of uncertainty. This article explains how to design that workflow, from choosing observations to delivering an inspection list, without confusing a visually convincing chart with a validated diagnosis.
Why 2023 matters for harmonised satellite observations
NASA reports that version 2.0 of the Harmonized Landsat and Sentinel-2 dataset was completed in summer 2023. The project is an example of making observations from different missions more consistent for analysis. It provides a useful reference point for discussing agricultural time series, while the later NASA account should not be mistaken for an article originally published in 2023. [1]
The lesson for an agricultural service is that combining images requires an explicit comparison method. Two sensors may observe similar wavelengths yet differ in their response, acquisition geometry and pixel placement. A chart that alternates between them can inherit those differences. Before interpreting a sudden change as a crop event, an analyst should establish whether the change also coincides with a sensor transition or a change in the processing method.
Start with a field registry and a clear decision
Useful monitoring begins with a stable description of what is being monitored. Each field or management block needs an identifier, a boundary, a crop record and a history of relevant changes. Where boundaries overlap or shift between seasons, a comparison can accidentally mix different areas. A field registry should therefore preserve previous versions instead of quietly replacing them. The resulting history helps explain whether an apparent improvement reflects the crop or a newly drawn boundary.
The intended decision should be equally specific. A regional crop analyst might need to compare seasonal development across districts. A plantation team might need a short list of blocks to inspect after unusually dry weather. These tasks can use some of the same imagery, but their reporting units and response times differ. Defining the action first prevents the service from collecting attractive indicators that do not change anyone's work.
Preserve quality information alongside every observation
NASA's description of the HLS processing approach includes atmospheric correction, cloud and shadow screening, geometric alignment and adjustments intended to improve consistency between sensors. These are steps in producing comparable measurements, rather than proof that every remaining pixel is suitable for every agricultural task. An application still needs its own quality policy. [2]
For a field-level chart, retain the acquisition date, source product, valid area and reason for rejection when an image is excluded. A small clear patch should not automatically represent an entire field. Setting a minimum valid-area requirement can make the series more stable, but that requirement should be visible and consistent. If a threshold changes, the processing record should explain when it changed and which results were recomputed.
Handle cloud gaps without inventing observations
Optical crop monitoring often produces an irregular sequence. Some dates contain a useful observation; others contain cloud, shadow or an unsuitable view. A smooth line can hide those gaps. A more informative chart distinguishes measured points from interpolated values and shows how long the field has gone without a usable observation. That presentation lets the reader judge whether a change is well supported or largely dependent on the chosen interpolation method.
Interpolation can be helpful for exploratory analysis, but it should serve a defined purpose. An estimated value between two observations is different from a measurement of the crop on that day. Around harvest, flooding or another abrupt event, a smooth estimate may be misleading. Where evidence is insufficient, the operational result can simply be that the field requires another observation or a ground visit. An honest gap is useful information.
Read crop development in the context of time
A seasonal crop curve becomes more meaningful when its horizontal axis reflects crop development. Comparing two fields on the same calendar date may compare different growth stages if planting dates differ. Where reliable records exist, aligning observations by time since planting can improve interpretation. Where planting information is uncertain, the chart should make that uncertainty clear rather than assigning a precise growth stage from a vegetation indicator alone.
Historical comparisons also need a relevant reference group. A young plantation block should not be judged against mature trees without accounting for age and canopy structure. An irrigated field should not automatically be grouped with rain-fed fields. Organising comparisons around crop type, management and local conditions helps turn a broad anomaly into a focused question. The aim is to identify an unexpected trajectory within a meaningful peer group.
Build a cloud workflow that can be explained
A practical cloud pipeline can separate observation selection, preprocessing, indicator calculation, field aggregation and reporting. This structure makes errors easier to trace. If a field statistic changes unexpectedly, the team can check the source scene before reviewing later steps. Keeping intermediate summaries and processing versions also helps explain why a previous report differs from a newly generated one, which is important when reports support operational or research decisions.
Efficiency comes from processing the relevant area and dates, reusing verified results, and avoiding repeated work. Cost monitoring belongs in the design as well. A service should understand the resources needed to add a new farm, rerun a season or update an entire portfolio. Technical scale is valuable when it supports predictable delivery. It becomes less useful when the team cannot explain its own processing costs or reproduce a result.
Validate patterns before automating recommendations
Field evidence helps determine what an unusual curve actually represents. A decline might coincide with harvest, a change in management, standing water or an imaging artefact. Those explanations lead to different responses. During a pilot, record the alert, the inspection finding and the action taken. Reviewing that sequence can show whether the indicator is useful, whether a threshold is too sensitive, and whether the information reaches the person able to act.
Validation should include ordinary fields as well as suspicious ones. Inspecting only the most dramatic examples can exaggerate success and conceal missed problems. A planned sample across crop stages and conditions gives the team a more balanced picture. For predictive models, evaluation on a different season or location can be more informative than testing neighbouring pixels from the same image. The test should resemble the conditions under which the service will be used.
Deliver an observation record that supports action
A useful agricultural time-series report does not need to expose every processing detail on its main screen. It should show the field, the latest reliable observation, the comparison being made and the reason for attention. Supporting details can include the recent image sequence, data coverage and notes from previous inspections. The reader should be able to distinguish an observed change from its possible explanations without needing to understand the entire data pipeline.
For bhumisat's wider Earth observation vision, time-series thinking connects satellite design with needs on the ground. A future observation mission should be evaluated partly by whether its measurements can fit into a repeatable, validated workflow. The value lies in a growing body of usable evidence about each landscape. When that evidence is connected to field knowledge and clear decisions, satellite remote sensing becomes a practical component of agricultural intelligence.
Sources & further reading
Explore the bhumisat spacecraft concept and our vision towards 2030, or return to the journal.



