Cambridge Research Proposal Example: Food systems modeller to agricultural policy (Score 93)

Agricultural policy in low-income countries increasingly depends on quantitative projections of crop yield under climate stress, yet the models generating those projections are rarely designed with the decision horizons of national food security agencies in mind. Most integrated assessment frameworks optimise for long-run equilibrium outcomes rather than the near-term, threshold-based signals that procurement planners and subsidy designers actually require. The central research question is: at what compound climate stress thresholds — defined as co-occurring drought and heat events within a single growing season — do maize and sorghum yields in smallholder-dominated farming systems cross policy-relevant tipping points, and how sensitive are those thresholds to plausible variation in soil carbon and input-access conditions? Two subsidiary questions follow. First, how well do current regional crop models, calibrated primarily on commercial-scale trial data, reproduce yield responses observed in smallholder plot-level datasets, and where do systematic biases appear? Second, can a simplified threshold-indicator derived from the modelling exercise be operationalised within the planning cycles of a national agricultural ministry without requiring continuous high-resolution climate forcing data? The policy relevance is concrete. Governments managing strategic grain reserves or input subsidy programmes need decision triggers, not probability distributions extending to 2100. A model output identifying that a combination of more than fifteen consecutive dry days after anthesis and mean temperatures exceeding thirty-two degrees Celsius for five or more days reduces expected yield by over forty percent is directly actionable in a way that a long-run scenario ensemble is not. Closing the distance between simulation output and administrative decision is the bounded contribution this project aims to make. Two bodies of scholarship are directly relevant, and their limited dialogue defines the gap this project occupies. The first is the crop modelling literature. Process-based models such as DSSAT and APSIM have been applied extensively across sub-Saharan Africa to project yield responses to temperature and precipitation change. This tradition has established that compound stress events — simultaneous heat and water deficit — produce non-linear yield losses that single-stressor analyses underestimate. Lobell and colleagues have demonstrated that heat stress during reproductive stages is particularly damaging for C4 cereals and that interaction with soil moisture status modifies the damage function substantially. However, this literature is predominantly calibrated on experimental station data or commercial farm records, and comparative studies have found that smallholder plot responses diverge from station-calibrated predictions, partly because input use, soil organic matter, and planting date variability are compressed in the training data. The second body concerns food policy and agricultural decision-making under uncertainty. Scholars in the tradition of adaptive governance — including work associated with the CGIAR Research Program on Climate Change, Agriculture and Food Security — have argued that the translation gap between model output and policy uptake is structural rather than communicative: models are not designed to produce indicators that administrative systems can absorb. This literature, however, rarely engages with how model architecture would need to change to produce threshold-based outputs, nor does it empirically test whether simplified indicators derived from existing models retain sufficient predictive accuracy. The gap is therefore methodological and applied simultaneously. No published study has systematically tested whether compound stress thresholds derived from a recalibrated smallholder-parameterised crop model can be compressed into observable indicators that retain predictive validity across the agro-ecological diversity of a single national context. This project proposes to do that for one country case as a proof-of-concept extensible to comparative work. The project proceeds in three phases, each with a defined output. Phase one: model recalibration. I will work with an existing DSSAT configuration for maize and sorghum, repairing the smallholder parameterisation gap by incorporating plot-level yield and management data from the Living Standards Measurement Study Integrated Surveys on Agriculture (LSMS-ISA), which cover Ethiopia, Tanzania, Uganda, and Malawi at plot resolution with GPS coordinates, crop variety records, and input-use data. One country will be selected as the primary calibration case based on data completeness; a second will serve as out-of-sample validation. Recalibration will adjust soil organic carbon distributions, planting date variance, and fertiliser application rates to reflect smallholder conditions. Model performance will be evaluated against observed yield distributions using root mean square error and the coefficient of variation of residuals across agro-ecological zones. Phase two: compound stress threshold identification. Using the recalibrated model, I will run factorial simulations varying heat stress duration and intensity, soil moisture deficit timing relative to phenological stage, and soil carbon levels across the LSMS-ISA observed range. The output will be a response surface mapping yield loss against compound stress combinations. Threshold identification will use change-point detection applied to simulated yield distributions, identifying stress combinations at which yield loss accelerates non-linearly. Uncertainty in threshold location will be characterised via bootstrap resampling across the parameter space. Phase three: indicator simplification and policy translation. Identified thresholds will be expressed as functions of observable meteorological variables available from ERA5 reanalysis and national meteorological service station networks, assessing whether thresholds can be monitored without continuous high-resolution model runs. I will then conduct structured interviews with three to five agricultural ministry planning officers to test whether simplified indicators map onto existing administrative decision points. This qualitative component is deliberately limited: diagnostic rather than ethnographic, its purpose is to identify structural barriers to uptake rather than characterise institutional culture. The LSMS-ISA data are publicly available through the World Bank microdata catalogue under standard academic use terms; no primary data collection is required for quantitative phases. DSSAT is open-source. ERA5 reanalysis data are freely accessible through the Copernicus Climate Change Service. The main access risk is that plot-level GPS coordinates in some LSMS-ISA waves are spatially displaced to protect respondent privacy; I have reviewed the displacement protocols and agro-ecological zone assignments remain valid, though field-level soil data will need to be drawn from gridded products rather than matched point observations. The interview component will require ethics review through the appropriate Cambridge departmental process. Participants will be ministry officials in a professional capacity; no vulnerable populations are involved. Informed consent and data anonymisation will follow standard social science protocols. Timeline across twelve months: months one to three, literature consolidation and model recalibration; months four to six, compound stress simulations and threshold analysis; months seven to nine, indicator simplification and reanalysis validation; months ten to eleven, interviews and policy translation analysis; month twelve, thesis writing and revision. Scope is deliberately bounded to one primary country case with one validation case; comparative extension is flagged as a doctoral direction rather than an MPhil deliverable. The Department of Geography at Cambridge, and specifically research activity associated with the Cambridge Centre for Environment, Energy and Natural Resource Governance, provides the closest institutional home for this project. The department's engagement with food systems governance, land use modelling, and climate adaptation policy aligns directly with the applied modelling orientation of this proposal. The project requires high-performance computing for the DSSAT simulation ensemble; the Cambridge High Performance Computing Service provides this infrastructure to registered graduate students. The LSMS-ISA and ERA5 datasets are accessible without institutional data-sharing agreements beyond standard registration. I am seeking supervision from a researcher with expertise in crop modelling or agricultural systems analysis combined with an interest in policy-relevant climate adaptation; I have identified the department's published work in this area as the basis for this alignment and understand that formal supervisory agreement is subject to departmental matching processes. The MPhil in Environment and Sustainability provides the research training framework — particularly in quantitative environmental methods and governance — within which this project sits. The one-year timeline and bounded country-case scope are calibrated to MPhil feasibility, and the thesis is designed to produce a methodological proof-of-concept with clear doctoral extension pathways rather than a comprehensive regional assessment.