There is a workflow that almost every agricultural researcher knows. Data gets recorded in a field notebook, transferred to Excel, analysed in SPSS or R, figures get made in GraphPad or Excel again, tables get formatted manually in Word, and somewhere in that chain of copy-paste operations, a value gets transcribed wrong, a column gets misaligned, and a result that took three months of fieldwork to generate becomes unreliable for reasons that have nothing to do with the science.
I know this workflow intimately. My MSc research at the Dhaka University Nanotechnology Centre involves comparing nano-urea — nitrogen fertilizer processed into nanoparticles for foliar application — against conventional granular urea on wheat, across eight treatment combinations and 30 measured parameters. When I sat down to begin analysis I realised the fragmented pipeline would cost me more time than the lab work itself. So I built something different.
SPADE (Statistical Platform for Agronomic Data Evaluation) is the tool I wish had existed when I started.
The problem it addresses
Nitrogen use efficiency studies have a specific and frustrating inconsistency problem. There are five established NUE indices (PFP-N, AE-N, RE-N, PE-N, and the Nitrogen Harvest Index), each measuring a different dimension of how applied nitrogen is converted to grain. Most published papers report only one or two, usually the easiest ones. Recovery efficiency (RE-N) and physiological efficiency (PE-N) require tissue nitrogen data from both grain and straw, so they get skipped. This means the published record on nano-urea is systematically incomplete. You can find dozens of papers showing that nano-urea increases agronomic efficiency, but you cannot easily find papers that tell you why, because the indices that answer the mechanism question are missing.
SPADE computes all five indices automatically the moment you have the data. The barrier to complete reporting drops to zero.
Related postNUE Indices: AE, RE, PE, and NHI Explained — the five indices SPADE computes, with formulas and a worked example.
The second problem is statistical. The compact letter display — the system of letters (a, b, ab) used in agricultural publications to summarise which treatments are significantly different from each other — is routinely implemented incorrectly. The naive algorithm assigns letters by scanning treatments in order of their means, which produces wrong groupings whenever the significance pattern is non-monotonic, a very common situation in fertilizer dose-response experiments. SPADE implements the Piepho (2004) maximal clique method, which is mathematically guaranteed to produce correct letter assignments for any pattern. This is not a minor technical detail. Incorrect CLD letters in a published paper mean the conclusions about treatment comparisons are wrong.
Third: with only three replicates per treatment, which is standard in pot experiments, most statistical tools will happily run tests that are statistically inappropriate. Shapiro-Wilk for normality at n=3 has almost no power — it almost never rejects regardless of the actual distribution, making the test meaningless. Dixon's Q test is built for samples of this size and is what SPADE uses instead. The outlier results include a plain-language explanation: not just "Q=0.987 > Q_crit 0.970" but "R2 value (2.1) is the lowest replicate and deviates 92.7% below the other values (mean of others = 8.25). Q=0.9870 exceeds the critical value 0.970." You can read that sentence and understand what it is telling you.
Related postsOne-Way ANOVA in R for Agricultural Research and Tukey HSD: Post-Hoc Testing After ANOVA — the analytical steps SPADE automates, explained in full.
What SPADE actually contains
The dashboard has eight tabs covering the full analytical pipeline.
Manages 30 parameters across growth, harvest, and soil categories in editable tables directly in the browser. It saves to Excel with named sheets. Treatment descriptions are editable from the UI; changes propagate immediately to every table, figure, and report without touching code. There is also a data quality flags system: mark individual observations as Verified, Suspect, or Excluded with a free-text note, creating an audit trail that lives alongside the data.
Runs the standard agricultural analysis (F-test, Tukey pairwise comparisons, CLD letters) but adds two things most tools omit: effect sizes (η² and ω²) alongside every result, and a residual Q-Q plot for normality assessment on the ANOVA residuals rather than on the raw replicates. If the Q-Q plot shows a departure from normality, a Kruskal-Wallis non-parametric test with Dunn's post-hoc is one click away.
Decomposes the treatment effect into its component factors (N rate and foliar application type) with an interaction plot. This matters because analysing eight treatments as a flat one-way ANOVA, which is what most nano-fertilizer papers do, cannot tell you whether it was the N rate or the application method driving the difference, or whether the two factors interact.
Shows all five NUE indices simultaneously, grain protein content, and the root-to-shoot dry weight ratio with its own ANOVA. The R:S ratio uses dry weights by default (scientifically correct) and falls back to fresh weights with an explicit note if dry weights have not yet been entered.
The tab that does the work most pot experiment papers omit. Multi-nutrient uptake for N, P, K, S, Ca, and B is computed from tissue concentrations. Biological yield (grain + straw + root dry weight) and the correct harvest index are computed. A nitrogen balance table shows what happened to applied nitrogen: how much ended up in grain, how much in straw, how much changed in the soil relative to the pre-experiment baseline, and how much is unaccounted for. The pre/post soil comparison table shows the percentage change in every soil parameter between baseline and post-harvest for each treatment.
Runs Dixon's Q and Grubbs on every treatment group for any selected parameter, with the plain-language explanations described above. A full scan button tests all 30 parameters simultaneously. The results note explicitly that at n=3, both tests have very low power — a flag means something needs looking at, not that the value must be removed.
Four chart types: bar charts with Okabe-Ito colourblind-safe colours, CLD letters, and optional significance brackets; strip plots that show actual replicate values rather than error bars (more honest for small n); scatter plots with Pearson correlation; and a radar chart that normalises multiple parameters to the same scale for multi-treatment comparison. Everything exports as PNG or JPG at twice screen resolution.
Generates a formatted Word document containing ANOVA tables, treatment means (Mean ± SD, not SE; SD describes actual variability while SE at n=3 compresses the spread misleadingly), NUE indices, N balance, and field event documentation. It also generates a significance summary sentence per parameter:
"Grain yield was significantly affected by treatment (F(7,16) = 12.34, p < 0.001 **, ω² = 0.71 [large effect], LSD₀.₀₅ = 0.84). The highest mean was recorded in T3 (13.72 ± 0.21 SD), which was significantly superior to T1 (p = 0.001) and T6 (p = 0.023)."
Why it runs locally
SPADE requires no internet after installation. No subscription, no login, no data leaves your machine. This is a deliberate design choice for researchers at institutions in South Asia and similar settings where cloud services may be unreliable, restricted, or unaffordable. The entire analysis pipeline, from raw replicate entry to formatted Word report, runs on a laptop with Python installed.
Data is saved to a standard Excel file that you can open, inspect, and back up independently of SPADE. If SPADE were to disappear tomorrow, your data would still be there.
The broader context
This dashboard emerged from a specific experiment: nano-urea versus conventional urea on BARI Gom 33 wheat, eight treatments, Completely Randomised Design, 24 pots at the DUNTC facility. The research question is whether foliar nano-urea at 50 to 75% of the recommended nitrogen dose can maintain grain protein quality and nitrogen recovery equivalent to full-dose conventional urea. If it can, the implications for input costs, groundwater nitrate contamination, and greenhouse gas emissions from smallholder wheat production in Bangladesh are real and worth pursuing.
The experiment was interrupted by a hailstorm during grain-filling, which destroyed most of the grain in some pots. A lesser outcome would have been to discard the study. Instead, the analytical focus shifted to parameters the hail could not touch — tissue nitrogen partitioning, nitrogen harvest index, root:shoot ratio, and post-harvest soil nutrient dynamics. SPADE documents this explicitly: the field event is recorded, pots with total grain loss are excluded from yield analysis with a note, and all tissue and soil data are retained. This is how uncontrolled field events should be handled: transparently documented, not omitted.
Related postWhat Happens to Urea After It's Applied: A Visual Explainer of Nitrogen Cycling — the nitrogen loss pathways SPADE's NUE indices are designed to measure and report completely.
Get it
SPADE is open-source under the MIT licence. The full source code, a sample dataset, installation instructions, and a detailed user guide are all available in the repository below.
Source code · sample dataset · installation guide · user documentation. A methods paper is in preparation targeting Computers and Electronics in Agriculture. If you are working on nitrogen management, fertilizer trials, or controlled crop experiments and want to discuss the tool, reach out through sajjadur-rahman.com.