Temperature Anomaly - Climatology Baseline Methodology

The question a raw temperature number cannot answer

July 28 in Ames, Iowa: maximum temperature 34°C. Is that alarming? Normal? Unusually cool?

The raw number is uninformative without context. 34°C is above average for late July in central Iowa - but how far above? How often does late July reach 34°C historically? Is this part of a sustained heat pattern or an isolated day? These questions can only be answered by comparing today’s observation against a historical baseline - a climatology.

MSCIP’s temperature anomaly computation provides that context by expressing each day’s temperature as a deviation from what has historically been normal for that location and calendar date.

Data source: NASA POWER

All temperature data used in MSCIP anomaly calculations comes from NASA POWER (Prediction of Worldwide Energy Resources), a gridded meteorological dataset that provides daily maximum temperature (T2M_MAX), minimum temperature (T2M_MIN), and mean temperature (T2M) at 0.5° × 0.5° spatial resolution globally.

NASA POWER derives its historical data from the MERRA-2 reanalysis (Modern-Era Retrospective analysis for Research and Applications, Version 2), which blends satellite observations, surface station data, and atmospheric model output to produce spatially consistent daily gridded fields back to 1981. For MSCIP’s crop monitoring regions - the US Corn Belt, US Wheat Belt, Mato Grosso, and Paraná - MERRA-2 coverage is dense and well-validated against ground stations.

Daily POWER data is available via API with typically 2–5 day latency, sufficient for MSCIP’s weekly crop intelligence refresh cadence.

The 30-year climatological baseline: 1991–2020

MSCIP’s baseline period is 1991–2020, the current standard climatological normal period established by the World Meteorological Organization (WMO). The WMO updates its reference period every decade; 1991–2020 replaced the previous 1981–2010 normal in 2021.

For each monitored location and each calendar day (January 1 through December 31, treating leap-year days consistently), MSCIP computes:

• Daily mean normal (μ): the average of observed T2M values for that calendar day across all years 1991–2020
• Daily standard deviation (σ): the standard deviation of that same 30-year sample

The resulting daily anomaly is:

$$\text{Anomaly}\text{day} = T\text{obs} - \mu_\text{day}$$

Positive values indicate warmer than normal; negative values indicate cooler than normal. Units are °C.

Why daily granularity outperforms monthly

Monthly anomaly summaries - “July was 1.2°C above the 30-year average” - mask the timing information that matters most for crop outcomes. A July that was 3°C above normal for the first ten days but normal for the remaining 21 days will produce a monthly anomaly of roughly +1°C. But those 10 anomalous days coincide with corn pollination in much of the US Corn Belt, and that timing makes the month agronomically very different from a month where the same average heat was distributed uniformly.

Daily granularity preserves the growth-stage alignment that drives crop outcomes. MSCIP computes:

• Rolling 7-day anomaly: week-over-week temperature departures for current crop monitoring
• Rolling 30-day anomaly: monthly context for sustained warm or cool patterns
• Season-to-date cumulative anomaly: aggregate thermal departure from planting through current date, which provides a GDD-correlated indicator separate from (but consistent with) the direct GDD accumulation count

The 30-day rolling anomaly is the primary metric shown in MSCIP crop condition summaries. It smooths out weather noise while preserving timing resolution at the weekly scale.

Crop region geographies

MSCIP monitors temperature anomaly across fixed crop region bounding boxes:

Within each bounding box, MSCIP area-weights the gridded 0.5° cells by their overlap with the agricultural area (cropland fraction) from MODIS land-use classification, rather than equally averaging all cells including non-agricultural land. This better represents the temperature experienced by the actual crop.

Why the 1991–2020 baseline matters

Climatological normals are not fixed absolute references - they are rolling 30-year averages that shift as the climate changes. Using the current 1991–2020 normal rather than an older baseline (1961–1990 or 1971–2000) matters for several reasons:

Anomaly interpretation accuracy. Global average surface temperatures have risen approximately 0.4–0.5°C from the 1971–2000 baseline to the 1991–2020 baseline. Using an older baseline would systematically inflate “positive anomalies” in a way that doesn’t reflect what farmers and crop models consider unusual. A day that looks 1.5°C above normal against a 1971–2000 baseline might be 1.0°C above normal against the 1991–2020 baseline - a meaningful difference when these anomalies are used as model inputs.

Consistency with agronomy research. Yield-anomaly studies published in the last decade increasingly calibrate against the 1991–2020 or recent WMO normal, making MSCIP anomalies directly comparable to reported research findings.

Application in ensemble crop yield models

Temperature anomaly enters MSCIP’s Phase 4 crop yield ensemble as one of five primary features alongside GDD accumulation, EDD accumulation, water stress index, and NDVI (from MODIS MOD13A2 16-day composites).

Research on the relative contribution of temperature anomaly to yield variance is consistent across studies: temperature (in one form or another) is the single most common predictor class across published crop yield prediction surveys, appearing more frequently than water stress, precipitation, or spectral indices. However, temperature anomaly works primarily as an interaction term: a hot anomaly during vegetative growth has a different effect on final yield than the same anomaly during reproductive stages, which is why anomaly alone (without growth-stage context from GDD accumulation) is an incomplete signal.

In MSCIP content, temperature anomaly is reported as context and as one input to the ensemble - never as a standalone yield forecast.

Data sources

World Meteorological Organization (2021). The 1991–2020 Standard Climatological Normal Period. WMO-No. 1251. Geneva: WMO.

MERRA-2 reanalysis dataset: NASA Global Modeling and Assimilation Office (GMAO), Goddard Space Flight Center.

See also

• Growing Degree Days - how thermal time accumulation differs from temperature anomaly, and why both are needed
• Water Stress Index - temperature is one side of the evapotranspiration equation; water stress is the other
• MSCIP Forecast Maturity Framework - how temperature anomaly enters ensemble models and what validation is required before it drives forecasts