Research Question 3: How can an MRV framework be designed to reliably quantify soil carbon sequestration and N2O emission reductions through drip irrigation under different contracting and management scenarios in Türkiye — and how should the measurement effort itself be allocated to maximize net carbon revenue?
Part 1 · The Rules
What Verra VM0042 Requires
The six requirements below are the parts of the VM0042 / VMD0053 methodology that actually shape the measurement design. The last two — the uncertainty deduction and its two error sources — are the bridge into the optimization problem in Part 2.
1
Measure-and-Model (Approach 1)
VM0042 offers three quantification approaches. This project uses Approach 1: measure-and-model — a process-based biogeochemical model predicts the emission reduction for every field, and direct field measurements are used to calibrate and validate that model rather than to credit each field one by one.
VM0042 · Approach 1
2
Calibrate, then validate independently
The model is calibrated on one body of measured data, then validated against a separate, independent dataset it never saw during calibration. Only validation performance decides whether the model is allowed to generate credits.
VMD0053 · Model validation
3
Validation is coverage, not a fixed count
There is no “collect N samples” rule. The validation dataset must span the project’s conditions: every climate zone present, the three predominant soil textures, and a clay-content range of at least 15 percentage points — with enough independent observations (≈10+) to run the statistical tests.
VMD0053 · §5.2
4
Two validation gates must pass
To be usable, the model must clear both: (1) Bias gate — mean bias ≤ the model’s own prediction uncertainty (PMU); (2) 90% coverage gate — the model’s 90% prediction interval must contain at least 90% of the independent validation observations.
VMD0053 · §5.2.4–5.2.5
5
The uncertainty deduction
Credits are discounted for uncertainty. The combined error of the net reduction becomes a deduction factor UNC, and:
\( \widehat{ER} = ER \,(1 - UNC) \)
More uncertainty → fewer credits. This single rule is what converts measurement quality directly into money.
VM0042 · eq. 63 & 74
6
Two sources of error
The deduction has two ingredients: model prediction error (how well the model matches reality — an irreducible floor) and sampling error (field-to-field variability — reducible by measuring more points). The whole design problem is: how much to spend shrinking the second, given the first.
Law of total variance
Part 2 · The Math
Problem Formulation
The measurement program is chosen before any data are seen, to maximize the implementer’s expected net present value under uncertainty. This is a stochastic optimization problem; specifically, a pre-posterior optimal sampling design.
Decision variables — what we choose
| Symbol | Meaning | Role |
|---|
| \(n_d\) | Soil / model sampling points in domain \(d\) | Reduces sampling error (\(\sim 1/n_d\)) |
| \(k_d\) | Project validation fields measured (N2O) | Optional; sharpens \(\sigma_{\text{model}}\) |
| \(a_d\) | MRV strategy: Model-only / Sample / Defer | Which error channel to attack |
| \(y_{d,t}\) | Whether domain \(d\) is enrolled by year \(t\) | Cohort roll-out schedule |
Uncertain parameters — what we simulate
| Symbol | Meaning | Source |
|---|
| \(ER_d\) | Emission reduction (tCO2e/ha) | Adana: Scenario B ≈ 3.59 (floor) to A ≈ 6.18; else literature |
| \(\sigma_{\text{model},d}\) | Model prediction error (irreducible floor) | Published model-validation range (RRMSE) |
| \(\sigma_{\text{obs},d}\) | Between-field spatial variability | Adana field campaign + literature transfer |
| \(P_t\) | Carbon price in year \(t\) | Market scenarios |
| \(\theta_d\) | Bass adoption \((p_d,q_d,\bar m_d)\) | Rogers’ diffusion, math form; Sultan 1990 + interviews |
Derived — the VM0042 deduction chain
The floor that drives everything. As \(n_d\to\infty\) the sampling term vanishes, but \(s^2_{\text{mod},d}\) remains. The deduction can never fall below an
irreducible model-error floor, so the value of extra sampling
saturates — and that saturation point is exactly what decides, per domain, whether to
measure or just
model (\(a_d\)). The full objective, constraints (capacity, budget, enrollment logic) and an interactive version live on the
MRV Feasibility & VoI Tool →
Part 3 · The Plan
Methodology — Step by Step
The logic is a single sentence: simulate the uncertainty → evaluate each candidate design across the scenarios → optimize the design (not the scenarios). Five steps:
Define the domains / strata
Partition the program into D1–D9, each a unique crop × crop-functional-group × region (climate zone × soil texture). Each domain is credited and analyzed separately, so uncertainty is assessed where it actually varies.
Quantify the two error sources / without running a new model
Model error (\(\sigma_{\text{model}}\)) comes from published model-validation studies, taken as a range (RRMSE) rather than a single number. Sampling error (\(\sigma_{\text{obs}}\)) comes from the Adana N2O field campaign (between-field SD) plus literature transfer. Both enter as uncertain inputs — no DNDC run is required for the core result.
Simulate the uncertainty / Monte Carlo
Draw \(ER\), \(\sigma_{\text{model}}\), \(\sigma_{\text{obs}}\), price, and Bass adoption jointly over many scenarios \(\omega\). Three representative adoption paths (conservative / moderate / optimist) are used only to report results, not to enumerate the space.
Optimize the measurement design / pre-posterior
For each candidate design — how many soil points \(n_d\), which domains to measure \(a_d\), the enrollment schedule \(y_{d,t}\) — compute \(\mathbb{E}[\mathrm{NPV}]\) across the scenarios, then pick the design that maximizes it. The decision is made before any data are observed; risk is handled with a Sharpe-ratio / chance-constraint variant.
Operate adaptively / and deliver the tool
As real validation and sampling data arrive, the ranges narrow (prior → posterior) and the design is re-solved for each new cohort — forward-looking only, since already-issued credits are never changed. Packaged as a decision-support tool (two modes below).
Mode 1 · before data
Design mode
Inputs are uncertainty ranges. Output is the starting program: which domains to measure, the first cohort’s sample sizes, and the validation plan.
Answers: “Where do I spend my MRV budget first?”
Mode 2 · while running
Operation mode (adaptive)
Real model error (from validation) and real spatial variance (from collected soil data) replace the ranges. The tool re-optimizes the next cohort’s sampling.
Answers: “Given what I’ve learned, how much do I sample next year?”
Two procedural caveats (being verified against the methodology text): (1) updated uncertainty affects only future crediting — issued VCUs are not retroactively adjusted; (2) sample-size changes must follow the registered monitoring plan, so adaptivity is anchored to monitoring / verification events rather than arbitrary annual changes.
Integration with RQ1 & RQ2: The domains and adoption parameters come from RQ1 (which basins and crops, and the diffusion interviews); the costs, carbon price, and NPV engine come from RQ2 (the techno-economic model). RQ3 adds the measurement-design layer on top — deciding how MRV effort should be allocated across those domains over time.