Global Agriculture · Climate Change · Planting Windows
A three-decade analysis of agricultural timing disruption using ERA5 reanalysis data, 1990–2020
Abstract
Using 30 years of ERA5 Land reanalysis data (1990–2020), this study quantifies shifts in optimal agricultural planting windows across the globe. A composite planting score — integrating monthly temperature, precipitation, soil moisture, and frost risk — is computed for three decades and compared to identify regions of significant change. Results indicate that 442 grid cells experienced score declines exceeding 5 points, compared to only 121 that improved, representing a 3.6:1 ratio of degradation to improvement. The most severely affected regions are concentrated in tropical Pacific and Southeast Asia, while marginal gains are observed in Arctic Russia — regions with negligible agricultural populations. Furthermore, 1,617 regions saw their viable planting window shorten by more than one month, against 1,256 that lengthened. These findings suggest a net global redistribution of agricultural timing opportunity away from food-insecure equatorial populations toward uninhabited high-latitude regions.
Key findings
1. Background
Planting too early risks frost damage; too late risks drought or shortened growing seasons. As global temperatures rise and precipitation patterns shift, the windows within which planting conditions are optimal are moving — earlier in some regions, later in others, and disappearing entirely in some. This study uses a composite planting score across four variables to quantify exactly where and by how much those windows have shifted over three decades.
2. Results — Score shift
The map below shows the change in annual planting score between the 1990s and 2010s. Orange and red indicate declining conditions; green indicates improvement. The geographic concentration of losses in Southeast Asia, the Pacific Islands, Central Asia, and parts of Eastern Europe is striking — these are regions with high agricultural dependency and limited adaptive capacity.
Figure 1. Change in annual planting score between the 1990s and 2010s. Diverging colorscale: red = worsening conditions, green = improving. Most significant losses concentrated in tropical Pacific (Papua New Guinea, Solomon Islands) and Central Asia.
3. Timing shift
Of 64,800 land grid cells analyzed, 2,400 saw their best planting month move earlier and 2,370 saw it move later — a near-symmetrical global split that masks strong regional patterns. In high-latitude regions warming faster than average, the optimal window has moved measurably earlier. In some tropical zones experiencing intensified dry seasons, the peak window has shifted toward wetter months.
Figure 2. Change in viable planting window length (months scoring above 50) between the 1990s and 2010s. Green = window lengthened; red = window shortened. Net global loss: 361 more regions shortened than lengthened.
4. Regional analysis
The geographic asymmetry of these findings carries significant humanitarian implications. The regions gaining the most — Arctic Russia, northern Siberia — have effectively zero agricultural populations. The regions losing the most — Papua New Guinea, the Solomon Islands, the Philippines, Borneo — are home to tens of millions of subsistence farmers with limited capacity to adapt. Climate change is transferring agricultural opportunity from those who need it most to land that is functionally inaccessible.
Most worsened regions
Most improved regions
Global summary statistics
5. Methodology
Monthly planting scores are computed for each 1° grid cell using a weighted composite of four climate variables derived from ERA5 Land monthly means: 2m temperature (30%), total precipitation (30%), volumetric soil water layer 1 (20%), and a frost risk index derived from temperature (20%). Each variable is scored on a 0–100 scale using domain-specific piecewise functions. The optimal temperature range is 10–25°C; optimal precipitation 50–150mm/month; optimal soil moisture 0.2–0.4 m³/m³. Scores below thresholds are penalized proportionally. Decade climatologies are computed by averaging monthly values across all years within each period: 1990–1999, 2000–2009, and 2010–2020.