Update by update
The auto estimate and the official intensity
How one update was made
Each row is one input to this update: the previous estimate carried forward on its trend, and every reading that arrived in the 3-hour window, calibrated on aircraft truth. Bars are 90% error ranges; the bigger the dot, the more the filter trusted it. The blue line is the result; the black dashed line is the official intensity.
Where the number comes from
Each voice's share of the estimate, through the storm's life
Every update
The ledger
Reading this page
Official. The grey dashed line is the working best track as it stands now: the agency revises it as new data arrive, so it can differ from what was issued at the time.
Auto-intensity. A damped-trend Kalman filter run forward only on objective voices — the UW-CIMSS ADT, CIMSS SATCON, the CIMSS machine-learning estimates when JTWC quotes them, the climatlas satellite estimator on NOAA GMGSI imagery, and aircraft (the larger of 0.9 × the 700-hPa flight-level wind and the SFMR surface wind). Each voice is inverted through its measured response to aircraft truth and anchored per basin, so what enters is its calibrated reading with an error bar that widens where the sensor saturates. The value at a synoptic time uses readings through the following three hours — the window in which that advisory is written and issued. Method and backtest: climatlas Auto-Intensity.
The shares. At each 3-hour step the filter blends its previous estimate, carried forward on its trend, with the readings that arrived, each weighted by its precision. Because the filter is linear, that recursion can be unrolled: the current number is exactly the sum of what every voice's readings contributed, back to the storm's first hour. The share chart is that sum.
climatlas Best Track. The green dashed line is the released climatlas Best Track (CBT), a reanalysis: it smooths the same satellite voices in both directions and takes the official book as one more voice, so it knows how the storm turned out. It is drawn for comparison and is not an input. It updates with each CBT release, not every six hours, so it can stop short of the latest updates.
Context, not inputs. Subjective Dvorak fixes (TAFB, SAB, CPHC, JTWC and the agencies JTWC quotes) and other agencies' 10-minute winds (converted toward 1-minute) can be switched on above. None of them enters the estimate.