2023-24
Verified actual- Production
- 1.362M t
- Yield
- 24.67t/ha
District series, area 55,200 ha
Harvest forecast
Where the trend puts Hathras's potato crop in 2024-25 and 2025-26 (estimates), and how sure we can be.
Note: the 2024-25 and 2025-26 harvests have already taken place, but official figures aren't in ColdFlow's data yet (latest record: 2023-24). The figures shown for these seasons are model estimates pending official data.
2024-25
ESTIMATED (Pending Actuals)- Estimated production
- 1.401M t
- Estimated yield
- 24.93t/ha
95% range 1.36M t – 1.44M t · area ~56,253 ha
2025-26
ESTIMATED (Pending Actuals)- Estimated production
- 1.429M t
- Estimated yield
- 25.10t/ha
95% range 1.39M t – 1.47M t · area ~57,052 ha
Actuals and projection
OLS linear trend on season index
- Actual
- Linear-trend estimates
- 95% prediction interval
- Shaded seasons: ESTIMATED (Pending Actuals)
Model statistics
One straight-line fit per metric, n = 6
| Metric | Trend / season | R² |
|---|---|---|
| Production | +28,686 t | 0.971 |
| Yield | +175 kg/ha | 0.872 |
| Area | +799 ha | 0.981 |
Recorded season and estimates
Point estimate and 95% interval
| Season | Production (t) | Yield (kg/ha) |
|---|---|---|
| 2023-24 Verified actual |
1,362,000 recorded |
24,673 recorded |
| 2024-25 ESTIMATED (Pending Actuals) |
1,400,733 1.36M t – 1.44M t |
24,927 24,397 – 25,457 |
| 2025-26 ESTIMATED (Pending Actuals) |
1,429,419 1.39M t – 1.47M t |
25,101 24,510 – 25,693 |
Six data points give a wide interval. Plan to the range, not the point. Weather, prices and seed supply are not modelled. Methodology →
How accurate has the forecast been?
Rolling-origin test: from each season, forecast the next one or two using only earlier data, then compare with what actually happened.
| Model | 1 season ahead | 2 seasons ahead | ||||
|---|---|---|---|---|---|---|
| MAE | WAPE | Bias | Skill vs naive | WAPE | Bias | |
| Linear trend (current model) | 16.4k t | 1.22% | +1.22% | +23% | 1.84% | +1.84% |
| Damped exponential smoothing lowest error | 9.9k t | 0.74% | +0.39% | +54% | 0.56% | +0.53% |
| Drift baseline baseline | 18.6k t | 1.39% | +1.01% | +13% | 1.46% | +1.46% |
| Naive baseline (last season) baseline | 21.3k t | 1.59% | -1.59% | — | 3.80% | -3.80% |
| Historical mean baseline baseline | 68.6k t | 5.11% | -5.11% | -221% | 6.70% | -6.70% |
Read with care. Only 3 one-season-ahead tests are possible with 6 seasons of history, so these errors are indicative, not proof of accuracy. A single unusual season could reorder the models. Treat the forecast range, not the point, as the planning figure.
- The linear trend (used for the headline forecast) beat the naive “same as last season” baseline and beat the drift baseline on one-season-ahead error.
- Its bias is +1.22%: it has over-forecast on average, so headline figures may run slightly high; for capacity planning that errs on the side of more space.
- Damped exponential smoothing had the lowest error here, but with 3 tests the difference is not reliable enough to switch the headline model; both forecasts are shown above.
Every test forecast (25)
| Model | Trained to | Predicted | Horizon | Forecast | Actual | Error |
|---|---|---|---|---|---|---|
| Linear trend (current model) | 2020-21 | 2021-22 | 1 | 1,339,000 | 1,312,000 | +27,000 |
| Linear trend (current model) | 2020-21 | 2022-23 | 2 | 1,379,000 | 1,351,000 | +28,000 |
| Damped exponential smoothing | 2020-21 | 2021-22 | 1 | 1,334,345 | 1,312,000 | +22,345 |
| Damped exponential smoothing | 2020-21 | 2022-23 | 2 | 1,365,678 | 1,351,000 | +14,678 |
| Drift baseline | 2020-21 | 2021-22 | 1 | 1,338,000 | 1,312,000 | +26,000 |
| Drift baseline | 2020-21 | 2022-23 | 2 | 1,378,000 | 1,351,000 | +27,000 |
| Naive baseline (last season) | 2020-21 | 2021-22 | 1 | 1,298,000 | 1,312,000 | -14,000 |
| Naive baseline (last season) | 2020-21 | 2022-23 | 2 | 1,298,000 | 1,351,000 | -53,000 |
| Historical mean baseline | 2020-21 | 2021-22 | 1 | 1,259,000 | 1,312,000 | -53,000 |
| Historical mean baseline | 2020-21 | 2022-23 | 2 | 1,259,000 | 1,351,000 | -92,000 |
| Linear trend (current model) | 2021-22 | 2022-23 | 1 | 1,352,000 | 1,351,000 | +1,000 |
| Linear trend (current model) | 2021-22 | 2023-24 | 2 | 1,383,900 | 1,362,000 | +21,900 |
| Damped exponential smoothing | 2021-22 | 2022-23 | 1 | 1,344,024 | 1,351,000 | -6,976 |
| Damped exponential smoothing | 2021-22 | 2023-24 | 2 | 1,361,578 | 1,362,000 | -422 |
| Drift baseline | 2021-22 | 2022-23 | 1 | 1,343,333 | 1,351,000 | -7,667 |
| Drift baseline | 2021-22 | 2023-24 | 2 | 1,374,667 | 1,362,000 | +12,667 |
| Naive baseline (last season) | 2021-22 | 2022-23 | 1 | 1,312,000 | 1,351,000 | -39,000 |
| Naive baseline (last season) | 2021-22 | 2023-24 | 2 | 1,312,000 | 1,362,000 | -50,000 |
| Historical mean baseline | 2021-22 | 2022-23 | 1 | 1,272,250 | 1,351,000 | -78,750 |
| Historical mean baseline | 2021-22 | 2023-24 | 2 | 1,272,250 | 1,362,000 | -89,750 |
| Linear trend (current model) | 2022-23 | 2023-24 | 1 | 1,383,100 | 1,362,000 | +21,100 |
| Damped exponential smoothing | 2022-23 | 2023-24 | 1 | 1,362,332 | 1,362,000 | +332 |
| Drift baseline | 2022-23 | 2023-24 | 1 | 1,384,250 | 1,362,000 | +22,250 |
| Naive baseline (last season) | 2022-23 | 2023-24 | 1 | 1,351,000 | 1,362,000 | -11,000 |
| Historical mean baseline | 2022-23 | 2023-24 | 1 | 1,288,000 | 1,362,000 | -74,000 |
MAE = mean absolute error. WAPE = total absolute error ÷ total actual harvest. Bias = total error ÷ total actual (positive = over-forecast). Skill = how much lower the MAE is than the naive baseline. Minimum training window 3 seasons.