about EART60061
0 · Exam Mapcourse + Canvas
“The test will be open book and will draw on course materials and the exercises/tasks you are asked to complete throughout the unit.” (EART60061 unit page, Canvas, 2026/27 1st semester)
Format in the course data: multiple-choice + short-answer, covering every intended learning outcome, in two 1.5-hour sessions. The closest preparation is to rerun the exercises, change a parameter and explain the output.
| If asked to… | First move |
|---|---|
| quality of values | s → SE → t·SE |
| combine values | choose error rule |
| test a line | r² → t → p |
| predict process | state model + check |
| step a model | old values only |
1 · Research ArgumentCh 1
quality testResearch = importance × novelty; rigour decides whether it counts.
Each objective must be achievable, recognisable and visibly linked to the aim. Order closed → open. Judge every section for coherence, completeness and concision. A Gantt chart should show parallel work and leave time for review.
| Part | Marker-facing test |
|---|---|
| General introduction | broad-to-narrow case ending at the requirement |
| Aim | single statement, close to title |
| Specific introduction | why the aim is difficult; terms defined |
| Objectives | 3–6 recognisable tasks, closed-to-open |
| Method | organised by objectives; replication and error explicit |
| Plan | parallel activities, feasible timing, review allowed |
Scope honestly: an MSc project need not have global importance or novelty; state whether each is local, national or wider. Rigour appears in suitable methods, replication and quantified uncertainty.
2 · Spread: Choose the StatisticCh 2
| Measure | Use / warning |
|---|---|
| Range | quick; only min/max, hostage to one extreme |
| Mean absolute deviation | sign-free; no distribution assumption |
| Sample s | uses every value; links spread to probability |
| Population σ | complete population only; divisor n |
three spread formulasx̄ = Σxᵢ/n
MAD = Σ|xᵢ−x̄|/n
s = √[Σ(xᵢ−x̄)²/(n−1)]
Why n−1? The mean was estimated from the same sample, costing one degree of freedom. Replicate field/lab data: STDEV or STDEV.S; STDEVP divides by n. Signed deviations sum to zero; squaring removes sign and weights extremes.
3 · Normal DistributionCh 3
standardised = (x−x̄)/s · 95% range = x̄ ± 1.96s
within ±d: 1.00 → 68.3%; 1.96 → 95%; 2.58 → 99%
For a threshold, sketch first: one tail or both? Tail estimates assume random variation and that sample x̄,s represent the population. A home-built area table needs frequency × class width; coarse classes create resolution error.
| d | Central share | Both tails |
|---|---|---|
| 1.00 | 68.3% | 31.7% |
| 1.96 | 95.0% | 5.0% |
| 2.58 | 99.0% | 1.0% |
Spreadsheet mechanics that get tested
- Set calculation to automatic before trusting changing outputs.
- For a frequency distribution, select all output cells, enter the data range and class limits, then use the array entry method shown in the exercise.
- Fix the mean cell with an absolute reference before copying deviation formulas.
4 · Confidence LimitsCh 4
four-line methodSE = s/√n
df = n−1
95% CL = t₀.₀₅,df × SE
report x̄ ± CL, units; interval [x̄−CL, x̄+CL]
Excel: TINV(0.05,n−1). Examples from the guide: df=2 → t=4.303; df=9 → 2.262; t approaches 1.96 only as n becomes very large. For two means, non-overlap supports a difference at 95%; overlap means do not claim a difference.
| Error bar belongs to | Quantity |
|---|---|
| single readings | s or d·s |
| mean of n | SE=s/√n |
| small-n 95% mean | t·SE |
Quote mean and limit to the same decimal places. Figure legend must state 95% confidence. In sampling design, control tide, location, species/size and handling; replication measures variation that remains.
4b · Decision TrapCh 4 FAQ
SD describes measurements; SE describes their mean. Using 1.96·SD for a mean is too wide; using 1.96·SE with three replicates is too narrow because t=4.303.
| n | df | 95% t | Use |
|---|---|---|---|
| 3 | 2 | 4.303 | triplicate mean |
| 4 | 3 | 3.182 | four replicates |
| 6 | 5 | 2.571 | six replicates |
| 10 | 9 | 2.262 | ten replicates |
| ∞ | ∞ | 1.960 | normal limit |
Biomonitoring design → defensible bars
Seaweed integrates exposure and concentrates metals, avoiding seawater's tide/current variation and salt matrix effects. Hold constant tide stage, sampling position, species and specimen size; replicate within each site. If a substitute species is unavoidable, sample both at a shared site and disclose the correction-factor ratio.
5 · Error PropagationCh 5
select by operationsum/difference: E = √(Eₐ²+Eᵦ²+…)
product/quotient: E_Q/Q = √[(Eₐ/A)²+(Eᵦ/B)²+…]
power xⁿ: E/xⁿ = |n|·Eₓ/x
Assumptions: errors are random and uncorrelated. Direct addition overstates error at the stated confidence. An exact multiplier has zero error: it scales absolute error, not fractional error. The largest fractional input error usually dominates—identify it to say which measurement should improve.
| Unit chain | Result |
|---|---|
| m³/s × mg/L | g/s (1000 L/m³; 1000 mg/g) |
| g/s × 86,400 s/day | g/day |
| g/day ÷ 1000 | kg/day |
5b · Flux & SamplingCh 5 case
Concentration is mass/volume; flux is mass/time and predicts transport. Estuary flux also varies through dilution, pH, adsorption, flocculation, settling, storms and stored metal. A negative tidal-cycle flux can mean net landward movement, not an arithmetic error. Compare a consent with the whole uncertainty interval.
| Process | Design response |
|---|---|
| Dilution | measure discharge |
| pH / adsorption | analyse dissolved + particulate |
| Settling / storage | allow for later resuspension |
| Tide | complete spring + neap cycles |
| Storm | acknowledge rare unsampled export |
Consent: report the result as a confidence interval; if it crosses the limit, the centre alone cannot settle compliance. Negative net flux during one spring tide can indicate temporary landward storage, not the long-run state.
6 · Modelling ProcessCh 6
| Stage | Question |
|---|---|
| Abstraction / assumption | What is omitted or simplified? |
| Formulation | How is the concept written mathematically? |
| Solution | Exact analytical or stepped numerical? |
| Verification | Does smaller Δt/Δx materially change it? |
| Calibration | Which plausible parameter values fit? |
| Validation | Does output match independent reality? |
Verification ≠ validation. Verification checks numerical approximation; validation tests assumptions against observations. Analytical solutions need no numerical verification. Add complexity only when failed validation identifies a missing process.
7 · Regression & TransformationsCh 7
significance from r²r = √r² · df = n−2
t = r√(n−2)/√(1−r²)
p = TDIST(t, n−2, 2); p<0.05 → significant at 95%
Report r² with p. Fit strength and significance answer different questions. Power law y=axᵇ → double-log straight line; exponential → semi-log. Splitting rising/falling hydrograph limbs captures hysteresis but reduces n, so recompute p.
| Result | Read it as |
|---|---|
| large r², p≥.05 | fit without enough evidence |
| small r², p<.05 | real but weak relation |
8 · Growth ModelsCh 8
exponentialdN/dt = rN · Nₜ = N₀eʳᵗ · lnNₜ = lnN₀ + rt
logisticdN/dt = rN(K−N)/K
Nₜ = KN₀ / [N₀+(K−N₀)e⁻ʳᵗ]
Read N₀ and plateau K from data. While N≪K, r=(lnN₂−lnN₁)/(t₂−t₁). Natural logs match e; multiply a base-10 gradient by 2.303. If calibration fails, revisit assumptions.
9 · CompartmentsCh 9
X₁′ = X₁ + Δt[F₀₁−(K₁₀+K₁₂)X₁]
X₂′ = X₂ + Δt[F₀₂+K₁₂X₁−K₂₀X₂]
Use old X₁ and old X₂ on both right-hand sides. Equilibrium: set changes to zero; X₁*=F₀₁/(K₁₀+K₁₂), X₂*=(F₀₂+K₁₂X₁*)/K₂₀. Residence time = 1/ΣKout. Check K·Δt<1; halve Δt and compare.
| Output clue | Action |
|---|---|
| logistic still misses | revisit model structure |
| store below X* and rising | system is a sink |
| ΣKout·Δt>1 | refine the step |
| Δt/2 changes result | not verified |
10 · Fick Difference StepCh 10
Fᵢ→ᵢ₊₁ = Kd(Cᵢ−Cᵢ₊₁)
Cᵢ(t+1)=Cᵢ(t)+Kd[Cᵢ₋₁−2Cᵢ+Cᵢ₊₁]
Check total mass each step. The guide's grid has a limiting Kd about 0.33; above it the neighbour overtakes the centre, and beyond 0.5 the centre becomes negative. For a conservative spike, variance = 2Kd·t.