The most expensive mistakes in a remote-sensing study are the ones you cannot undo once the fieldwork is done. A handful of decisions, made at the design stage, quietly settle whether the data will answer your research question or merely describe the scene — and most of them cannot be retrofitted after the season has passed. If a project is going to lean on satellite or drone data, these are the choices to get right before anyone collects a single point.
Match the sensor to the phenomenon, not the budget
Resolution and spectral range have to suit what you are actually trying to measure, and that is a decision with no do-over. A 10 metre pixel cannot resolve an individual plant; a broadband index like NDVI cannot separate two species with similar canopies; a sensor without a red-edge or shortwave-infrared band cannot see the stress signal a study might depend on. These limits are physical, and the window to choose differently closes with the season — you cannot re-fly last summer at a finer resolution. Choosing the data to fit the question, rather than the question to fit whatever data is free, is the first thing a reviewer will probe.
Design the ground-truth so it can actually validate
A classification or a map is a hypothesis until it is tested against independent reference data — and that reference data has to be planned into the fieldwork, not scraped together afterwards. A defensible accuracy assessment needs enough reference points per class, collected independently of the data used to train the model, and spread across the study area rather than clustered where access was easy. Points gathered as an afterthought are usually too few, non-independent, or spatially biased — and no amount of later processing repairs a validation design that was never there. This is the step that most often separates a result a journal accepts from one it sends back.
Build in accuracy and uncertainty from the start
A result reported without an accuracy figure and a statement of uncertainty is a description, not a measurement — and reviewers in this field now ask for both as a matter of course. A land-cover map needs its confusion matrix and error metrics; a continuous estimate needs its error bars and the limits of where it holds. The trouble is that these cannot be manufactured at write-up time: they depend on reference data and a validation design that had to exist during the work. Planning for the accuracy assessment up front is what makes it possible to report one at all.
Make it reproducible before you have to defend it
Reproducibility is decided by what you record while the work happens, not by what you can reconstruct under reviewer pressure months later. The sensor and platform, the exact acquisition dates, the processing level, the atmospheric correction applied, the index definitions and the model parameters all need to be documented as you go. A method section that cannot be followed by another analyst is a liability in review and a dead end for anyone — including your future self — who wants to extend the work. Good documentation is cheap during the project and nearly impossible to recover after it.
Where this helps
TerraX takes on the remote-sensing component of grant-funded research — the design, the analysis to a publishable standard, and the method documentation that survives peer review. The contribution is strongest when it starts at the proposal stage, where it can shape the sampling and validation design before the budget and the fieldwork are fixed. Where the work genuinely warrants it, co-authorship is welcome; where it is a service, it is delivered as one.
If a study is going to rest on remote sensing, the time to get the method right is before the data exists. That is the core of our research support service — tell us the research question, the data and the funding timeline, and we will help shape the analysis while the decisions are still open.