From blight alerts to field decisions: Are forecast apps finally accurate enough?

By Lukie Pieterse, Potato News Today

Decision-support apps for late blight have matured, but trust still hinges on whether risk scores translate into better spray timing, fewer applications, and cleaner fields. The newest platforms couple hyperlocal weather with calibrated disease models and farm telemetry, then push clear actions rather than vague alerts.

Why this matters now

Late blight remains the most economically damaging potato disease globally. Growers today face shifting Phytophthora infestans populations, evolving fungicide resistance pressures, and tighter sustainability expectations.

In this context, decision-support systems (DSS) promise to convert noisy weather and field signals into a defensible programme: when to start, when to stretch intervals, which actives to rotate, and how to document it all.

The question is not whether models exist—they do—but whether the latest apps are accurate and actionable enough to outperform calendar programs in the real world.

What the modern blight DSS actually does

Contemporary platforms ingest three classes of inputs:

• Weather and canopy wetness: on-farm stations, dense regional networks, and short-term forecasts drive hourly temperature, humidity, rainfall, wind, and leaf-wetness signals.
• Crop context: variety susceptibility, growth stage, canopy closure, irrigation events, and field history.
• Programme and telemetry: last spray date, product and dose, nozzle type and water volume, travel speed, and sometimes equipment data.

A DSS combines these with one or more late blight models to estimate infection risk and protection decay, then generates a specific recommendation: spray now or stretch, suggested interval, a short list of suitable modes of action, and a reminder on application set-up. The better systems also log decisions automatically to create an auditable trail.

Model guts – how blight models, leaf-wetness, and regional calibration fit together

At the core are epidemiological models that link weather to infection events and epidemic growth:

• Trigger rules and indices. The UK’s Hutton Criteria replaced the historical Smith Period as a more sensitive national trigger: two consecutive days with minimum temperature ≥ 10 °C and at least six hours each day at ≥ 90% relative humidity. Systems use this to flag high-risk windows and to start or tighten programmes.
• Integrated blight models. NEGFRY combines a “negative prognosis” (start criteria) with Fry-type severity values to time subsequent sprays; SIMPHYT and ProPhy perform similar roles in other regions. BLIGHTSIM simulates responses to diurnal humidity–temperature cycles and changing climates. These models are parameterised and validated locally before deployment.
• Regional calibration. EuroBlight and national groups curate country-specific model settings and fungicide attributes, reflecting local pathogen populations and agronomic practices. French MILEOS, for example, was built and maintained for national conditions and regulations.

Leaf wetness is the most contentious input. Many farms do not run leaf-wetness sensors, so platforms estimate wetness duration from humidity, temperature, radiation, and rainfall, or use physical/empirical emulators when sensors fail.

The method matters: over-estimating wetness inflates risk scores and compresses intervals, while under-estimating invites escapes. Where budgets allow, well-sited, maintained wetness sensors plus a quality-controlled estimation backstop is the pragmatic standard.

Data hygiene – the quiet factor that makes or breaks accuracy

A clean model can be undermined by messy data. Three hygiene points dominate:

• Station siting and maintenance. Temperature and humidity sensors too close to buildings, gravel yards, or tree lines can bias dew formation and RH maxima; rainfall gauges need level installation and regular checks. Adhering to WMO siting classes and field-scale best practices reduces systematic error.
• Quality control and gap-filling. Data dropouts are inevitable. Simple back-fills can distort risk curves; robust frameworks use multi-station cross-checks, dynamic time-warping, and machine-learning imputation with flags so agronomists see when values were reconstructed. The rule: it’s safer to show an explicit “data uncertain” banner than silently fill and over-trigger.
• Forecast blending. Near-term forecasts (0-48 h) now drive most “spray or stretch” calls. Platforms blend observed data with mesoscale forecasts—ideally bias-corrected using yesterday’s station truth—to avoid systematic warm/cool or wet/dry drifts.

Good apps make data provenance transparent: which station, distance to field, last calibration, missing intervals, and whether a forecast or observation drove a decision. Transparency builds trust—and helps explain odd alerts.

Actionability – turning a risk index into the next pass

Accuracy alone does not move the sprayer. Growers and agronomists want clear, bounded choices:

• Intervals tied to quantified risk. Rather than a single “red” flag, better DSS output shows a next-7-day risk curve, last spray protection decay, and a default interval with a +/- range. This enables stretching in low-risk lulls and tightening ahead of stacked wet nights.
• Mode-of-action rotation prompts. Integrations with EuroBlight/FRAC attributes help avoid over-use of single modes and steer toward products with proven foliar and tuber protection when conditions turn. Apps should surface why a particular active is proposed (e.g., strong anti-sporulant rating, or robust rainfastness) and flag label-specific limits.
• Application set-up cues. Droplet spectrum, water volume, travel speed, and nozzle choice influence canopy coverage and anti-sporulant performance. Field evidence suggests drift-reduction nozzles can perform well when calibrated, though extremely coarse sprays risk poor lower-canopy coverage; medium to coarse with adequate water volume remains a common compromise. Apps increasingly embed set-up checklists tailored to canopy stage and product.

In practice, the most useful interfaces package all this into a single card per field: “High risk from Wednesday evening; your last spray provides partial cover until Thursday mid-day. Recommended: treat by Thursday with a strong anti-sporulant mix; 250–400 L/ha, boom height 50 cm, medium-coarse droplets; rotate away from last week’s CAA.”

Verification – what side-by-side field logs say

Do DSS-guided programmes actually save sprays without sacrificing control? Controlled trials and meta-analyses over the past two decades consistently show potential for fewer applications with equivalent disease outcomes:

• Trials in Ireland comparing NEGFRY, SIMPHYT, ProPhy and PLANT-Plus to routine 7- to 10-day programs reported substantial reductions in application counts—often 40–60 percent—without significant loss of control under the trial conditions.
• A broad synthesis of 80 experiments across pathosystems concluded that DSS strategies can halve sprays compared with calendar schedules while keeping disease within acceptable bounds—an effect echoed in late blight literature when models are well calibrated and seasons are not extremely epidemic.
• National tools such as MILEOS report meaningful reductions in treatment frequency index (TFI) under French conditions, particularly when combined with varietal resistance.
• Newer regional studies continue to validate NEGFRY-type scheduling as a sustainable alternative in parts of Spain and elsewhere, though efficacy hinges on local calibration and high-quality weather inputs.

Two practical caveats recur in field logs. First, sprayability matters: in wet weeks, access constraints can push a “treat Thursday” recommendation to Saturday. Good platforms anticipate this with earlier “advance” nudges when a high-risk window is forecast and travel days look poor.

Second, inoculum pressure and strain mix vary within and across seasons; pairing DSS with in-season genotyping and regional outbreak monitoring improves confidence in the chosen actives and intervals.

Enterprise scale – rolling out across many farms and buyers’ audit needs

Once a DSS works for a single farm, the next hurdle is scaling across supply groups:

• User roles and permissions. Advisors need read-write across client blocks; growers need edit rights on their fields; processors and buyers often want read-only views and programme conformance checks.
• Device and network diversity. Mixed fleets of on-farm stations (e.g., METOS/FieldClimate, Sencrop, Arable) plus national networks must be harmonised, with APIs and station-level metadata to prevent double-counting and ensure bias correction.
• Audit trails for due diligence. Retailers and processors increasingly require evidence of IPM and resistance stewardship. Platforms that log who decided what and when—complete with weather backing, model state, and product FRAC codes—reduce paperwork and speed audits. Several commercial tools already expose these logs and export PDF/CSV summaries to meet buyer templates.

Commercial landscape – examples of platforms growers actually use

Availability varies by region, but today’s toolbox includes:

• National/Institute-backed DSS: MILEOS (France) with ARVALIS; country implementations of Hutton-based alerts and DSS integrations in Great Britain; university and government dashboards in the USA and Nordics that publish real-time disease risk by station.
• Weather-first platforms with potato models: METOS FieldClimate (Pessl) offers multiple late blight models and APIs; Sencrop provides farm-level Hutton indicators and risk dashboards.
• Crop protection–integrated solutions: BASF’s xarvio Field Manager provides growth stage prediction and disease risk alerts for potatoes in several markets, linking to product planning and documentation; other offers exist from major manufacturers via regional portals.
• Legacy and regional systems: NEGFRY deployments in parts of Europe; SIMPHYT/ProPhy in German-speaking regions; PLANT-Plus/Dacom advice modules in the Netherlands; and regional decision tools linked to processor programmes.

The key is not brand but fit-for-purpose: validated for the local climate and strains, fed by trustworthy weather, and embedded in a workflow growers will actually follow.

Leaf-wetness and microclimate: when “hyperlocal” earns its keep

Potato canopies create their own humidity envelopes after late-day irrigation or light night winds. A station at headland height can miss this, while an in-canopy sensor may over-report. Two pragmatic patterns are emerging:

1. Dual-source wetness: run one well-sited farm station plus one in-canopy wetness sensor in a representative field. Use the farm station as the default and the in-canopy sensor to correct evening/morning wetness duration.
2. Model-plus-sensor fusion: use a physics-based or machine-learned wetness emulator as the baseline, then nudge with a cheap contact sensor—flagging disagreements to the user.

Either way, the app should reveal which source drove the decision. Hidden switches erode trust.

Fungicide choice, rotation, and nozzle set-up: what good apps surface

Translating model outputs into a tank mix is where risk meets stewardship:

• Rotation and resistance stewardship. EuroBlight tables and FRAC guidance provide live guardrails on foliar and tuber protection, rainfastness, and anti-sporulant ratings. Embedding those attributes in an app avoids accidental over-reliance on a single mode—especially in long, wet cycles.
• Droplet size and water volume. For dense canopies or anti-sporulant aims, medium to coarse droplets at adequate water volumes (often 250–400 L/ha in Europe) improve penetration while limiting drift; extremely fine sprays risk drift and inconsistent lower-canopy coverage. Apps that include a pre-spray checklist—boom height, pressure, travel speed, nozzle choice—help standardise quality.

Limitations to respect

Even the best DSS will not compensate for certain realities:

• Access and logistics. Muddy tracks and wind limits can delay the “optimal” window. Programmes must be robust to 24–48 h slippage.
• Strain dynamics. A local emergence of more aggressive or insensitive clones can shorten safe intervals despite low modelled risk; linking to regional monitoring improves resilience.
• Garbage-in, garbage-out. Poorly sited or unmaintained sensors will degrade outputs no matter how elegant the model.
• Edge climates. In maritime or mountainous zones with frequent dew and rapid radiation swings, leaf-wetness estimation remains hard. Sensors help, but uncertainty bands should be visible.

Are apps finally accurate enough?

For many regions and seasons, yes—with conditions. Where models are locally calibrated, weather data are high-quality, and growers follow clear, auditable recommendations, DSS has repeatedly matched or beaten calendar programs on control and reduced spray counts.

The performance ceiling rises further when platforms fuse near-term forecast skill with practical application prompts and resistance-aware product choices. The bigger gap now is less about raw accuracy and more about workflow: clean inputs, timely alerts, and documentation that meets buyer audits.

How to trial a platform on your farm or supply group

• Pick one or two fields per variety class (e.g., one susceptible, one more tolerant) and run the DSS in parallel with your current schedule for a season.
• Instrument for trust. At least one well-sited farm station, plus a leaf-wetness sensor or validated emulator and a QC routine.
• Define decision rules up front. For example: “If high risk is forecast for ≥ 24 h before current protection expires, treat; otherwise stretch by 24–48 h.”
• Document everything. Last spray, product, dose, nozzle, water volume, travel speed, and notes on sprayability constraints.
• Post-season audit. Compare total sprays, intervals, disease outcomes (including tuber infection at grading), and cost. Decide field-by-field whether to scale.

If the platform can produce a clean PDF/CSV trail, you’ll also be sharper for buyer verification and sustainability reporting.

What to ask vendors before you sign

• Which late blight models are available and how are they calibrated for my region and varieties?
• What weather sources power my fields, and how are forecast biases corrected?
• How is leaf-wetness handled—sensor, estimation, or fusion—and how is uncertainty shown?
• Can I see data provenance, QA flags, and an audit trail of who changed what?
• How are EuroBlight/FRAC attributes integrated into product suggestions?
• What happens when data drop out mid-season—how does the app alert me, and what’s the safe fallback?
• Can advisors, growers, and buyers have appropriate role-based access without duplicating records?

Bottom line: Apps won’t spray the field for you, but the best of them now connect credible risk modelling with practical programme decisions and buyer-ready records. Used with discipline—and a healthy respect for data quality—they’re accurate enough to earn a permanent place in the blight playbook.

---

Sources and further reading

Hutton Criteria: new national warning system for potato late blight in Great Britain (AHDB) – https://potatoes.ahdb.org.uk/development-and-implementation-of-a-new-national-warning-system-for-potato-late-blight-in-great-britain-hutton-criteria potatoes.ahdb.org.uk

Bayer Late Blight Knowledge Hub: summary of Hutton Criteria – https://cropscience.bayer.co.uk/late-blight/ cropscience.bayer.co.uk

IPM Decisions factsheet: Hutton Criteria Late Blight Model – https://www.ipmdecisions.net/documents/factsheet-hutton-criteria-late-blight-model/ and PDF – https://www.ipmdecisions.net/media/4jkcvxnf/ipm_factsheet-hutton-criteria-late-blight-model_v0001_print.pdf

EuroBlight: DSS overview and MILEOS – https://agro.au.dk/forskning/internationale-platforme/euroblight/control-strategies/dss-overview and France – Mileos – https://agro.au.dk/forskning/internationale-platforme/euroblight/research-projects/ipmblight20/decision-support-systems-overview/france-mileos

NEGFRY and model background (UC ANR summary) – https://ipm.ucanr.edu/disease/database/potatolateblight.htmldavis

BLIGHTSIM: late blight model addressing diurnal fluctuations – https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7459445/

Field validation of DSS vs calendar programmes (Ireland) – https://t-stor.teagasc.ie/entities/publication/706861f8-5dec-4b2b-b957-546be669350a and summary – https://www.researchgate.net/publication/248116061_Field_validation_of_four_decision_support_systems_for_the_control_of_late_blight_of_potatoes_in_Ireland

Meta-analysis: DSS can halve sprays with acceptable control (Communications Earth & Environment) – https://www.nature.com/articles/s43247-021-00291-8 (open PDF summary) – https://www.ipmdecisions.net/media/gyhpevvs/lizenistrup-dss-halve-fungicide-s43247-021-00291-8.pdf

MILEOS impact and French IPM context – https://www.platform.smartprotect-h2020.eu/en/view/ipm/92 and McDonald’s Flagship Farmer case note – https://www.flagshipfarmers.com/media/1036/potatoes-leprince-france.pdf

Recent regional validation of NEGFRY (Spain) – https://www.mdpi.com/2077-0472/14/5/652 (open PDF: https://soildiveragro.eu/wp-content/uploads/2024/04/agriculture-14-00652.pdf)

WMO station siting and observation standards – https://community.wmo.int/en/activity-areas/imop/wmo-no_8 and siting classification details – https://community.wmo.int/en/activity-areas/imop/siting-classification; Personal weather station siting guide (NWS) – https://www.weather.gov/media/epz/mesonet/CWOP-Siting.pdf

Leaf-wetness: concepts, measurement and estimation (APS Feature) – https://apsjournals.apsnet.org/doi/pdf/10.1094/PDIS-05-14-0529-FE; estimation/emulator approaches – https://www.mdpi.com/2073-4395/11/2/216 and model optimisation examples – https://www.sciencedirect.com/science/article/am/pii/S0168192320301891

Data quality and gap-filling frameworks (agrometeorology) – https://www.mdpi.com/2624-7402/7/6/174 and classic weather-data imputation for agriculture – https://journals.ametsoc.org/view/journals/apme/39/7/1520-0450_2000_039_1176_emwdfa_2.0.co_2.xml

Pessl Instruments (METOS) potato disease models and FieldClimate API – https://metos.global/en/disease-models-potato/ and API docs – https://docsdev.fieldclimate.com/ (v2) / https://api.fieldclimate.com/v1/docs/ (v1)

Sencrop Hutton indicator for potatoes – https://uk.blog.sencrop.com/assessing-the-risk-of-late-blight-on-potatoes-with-sencrop/ and model explainer – https://sencrop.com/eu/hutton-criteria/

xarvio Field Manager (BASF) potato risk alerts and regional expansion – Product page: https://www.xarvio.com/us/en/products/field-manager.html; Japan potato functions news – https://ag.xarvio.com/global/news/GLOBAL_2024-04_BASF-Japan-adds-xarvio-FIELD-MANAGER-growth-stage-prediction-and-disease-alert-functions-for-potato-and-sugar-beet-to-support-key-crop-cultivation-in-Hokkaido; 2025 expansion note – https://www.basf.com/global/en/media/news-releases/2025/02/p-25-016

EuroBlight fungicide tables and methodology – Table: https://agro.au.dk/forskning/internationale-platforme/euroblight/control-strategies/late-blight-fungicide-table; methods preprint – https://www.researchsquare.com/article/rs-6305294/latest (ResearchGate mirror: https://www.researchgate.net/publication/391699813_Methodology_to_determine_fungicide_efficacy_ratings_for_the_EuroBlight_tables_-_potato_late_or_early_blight)

Application set-up and droplet spectrum evidence – Scottish Agronomy note – https://scottishagronomy.co.uk/new-choices-for-optimising-spray-droplet-deposition-in-potatoes/; review on nozzle and drift classes – https://edepot.wur.nl/630802; NWS/extension droplet selection guide – https://bae.k-state.edu/faculty/wolf/PDF/442-031_DropletChart-SelectionGuide.pdf; mixed-nozzle efficacy studies – https://bibliotekanauki.pl/articles/65449.pdf

University and institute risk dashboards – University of Wisconsin Vegetable Pathology (example) – https://vegpath.plantpath.wisc.edu/weather-models/

Author: Lukie Pieterse, Potato News Today
Image: