One block per source for the selected period. Mean is the area-weighted average over the selection (cos-latitude weighted, so northern cells are not over-counted); Median is the area-weighted 50th percentile, which for rainfall sits well below the mean because the distribution is strongly right-skewed; Maximum is the wettest single cell; and Wet Share is the share of the area above 0.1 mm — the figure that separates “light rain everywhere” from “one heavy storm cell” at the same mean. Volume converts the area integral to million m³ (1 mm over 1 km² is 1,000 m³), and Coverage states how much of the selection the grid actually answered for, which is what keeps a dry 0 mm distinct from a gap. Ground Stations are reported the same way over per-gauge accumulations, and are withheld below 1D because the network records one total per day.
No enabled source reported for Kingdom of Saudi Arabia over .
The precipitation field actually loaded for the selected frame, painted from the same grid values and the same colour scale the map is using — not an illustration. The outline is the real boundary of the selection (or the drawn ring / probed cell), so the figure and the statistics above describe the same footprint. Cells below 0.1 mm are left transparent, which is why a dry day shows the basemap through.
The same four distribution metrics as the summary, one row per enabled product, so the products can be compared against each other rather than read one at a time. Disagreement between rows is expected and informative: the products differ in resolution, retrieval algorithm and whether they are gauge-adjusted. Rows are ordered as the map's layer list orders them.
The same period and the same colour scale in every panel, over the same window, so a difference in colour is a difference in rainfall and nothing else. Two products can share a mean and disagree completely about where the rain fell — one concentrating it on an escarpment, the other spreading it across a province — and the table above cannot show that. Cell size is printed on each panel: a 0.25° product cannot resolve a storm a 0.04° one can, so some of the difference between panels is resolution rather than disagreement.
Reading the Ground Stations panel. The Ground Stations panel is points, not a field. One dot per reporting gauge, filled on the same colour scale as the grids beside it, with dry gauges (0.1 mm or less) drawn hollow so a measured zero stays distinct from ground no gauge covers. Nothing is interpolated between the dots: the space between them is unmeasured, which is what the grids are there to fill and what makes the comparison worth printing. Expect the dots to be higher than the cells around them — a gauge samples an orifice a few hundred millimetres across while a cell reports the average over 19 km² at 0.04° or ~700 km² at 0.25°, so a storm that filled one gauge is spread across a whole cell. Agreement in total with disagreement in place is the common and the important finding.
Which gauges are shown, and why. Only gauges reporting at least 80% of this period’s days are drawn or counted. The archive is not a uniform daily record: over 6,986 station-years, gauges reporting 1–37 days a year average 9.7 mm per reported day with 97% of those days wet, against 0.29 mm and 6% for gauges reporting 330–365. A sparse gauge is logging the rainy days, so its record is a mean over rain days, and reading it beside a complete neighbour compares depths measured over different numbers of days. Records before 2000 are an event log rather than a daily one — 17 reported days a year, every one of them wet — and are excluded from any per-day figure; period and annual totals are not filtered, because a rain-only log still captures most of the depth. A gauge that does not qualify is ABSENT from the map, the mean and the count — never drawn or counted as zero, which is the same rule the grids follow for a missing frame. The panel header states how many gauges survived.
Each reporting gauge inside the selection is paired with the grid cell that CONTAINS it, over the same accumulation window, and the products are scored against that sample — the standard point-to-pixel comparison. The two are not measurements of the same quantity: a gauge samples an orifice a few hundred millimetres across, a cell reports the average over its whole footprint, so part of the scatter below is that mismatch rather than product error, and it is largest for isolated convective storms and for the coarser 0.25° products. Magnitude agreement and event detection are reported separately because a product can carry almost no mean bias while missing the events a gauge recorded and inventing an equal number elsewhere: Bias is the mean of grid − gauge (positive = wetter than the gauges), MAE and RMSE are the typical and the outlier-weighted error, and r is inflated by the many pairs where both are dry. POD (Probability of Detection) is the share of gauge-observed events the product detected, FAR (False Alarm Ratio) the share of its own wet cells the gauges did not confirm, CSI (Critical Success Index) the two combined, and the bias ratio whether it produced the right NUMBER of events. Gauges are treated as the reference, not as truth — they under-catch in wind and heat — and only days and coarser windows are compared, since the network records one total per day. Finally, these scores describe THIS window over THIS area, not the product in general: one day is one sample of weather, and the same product can look dry-biased on a day when a storm sat between cells and wet-biased over the month containing it. A monthly or yearly frame, or a larger area, gives the more stable answer.
TWO charts, one per kind of source, because averaging them is what hides the thing worth seeing: a month where the gauges and the satellites disagree blends into a single bar that reads as an ordinary measurement. The first is the rain-gauge network alone — monthly means across the gauges in the selection, from their own daily totals. The second is the mean of the enabled satellite products alone: each product's complete set of daily area means is summed into calendar-month totals, then the products that could complete the month are averaged equally. Both charts share ONE vertical scale, so a taller bar means more rainfall in either of them and the pair can be compared by height. A partial month is left out of a product's total rather than reported low, and a month no source could complete is drawn at zero for that source only — neither side ever borrows the other's value, which is what keeps a disagreement visible. Missing values are excluded from a mean; a reported 0 mm is a real dry observation and contributes. Gauges record one total per day, so their chart is withheld below 1D.
The only figure in this report that does not follow the map: its time scale and date range are set in the report itself, so a document can carry an hourly storm profile or a twenty-year monthly series without moving the map off the frame every other section describes. The subject is still the map’s — the same area, ring or location as the rest of the document. One bar per frame, valued as the MEAN of the enabled precipitation layers that publish an archive at the chosen scale; a layer that does not resolve it is excluded rather than counted, and the sources that contributed are listed below the chart. A frame no source answered is drawn as a baseline tick, never as a zero bar. The dashed horizontal line is the mean of the frames that reported; the dotted sloped line is a least-squares fit, reported per 10 frames because at an hourly scale a slope per decade would describe a span the axis never shows — it is a fitted line, not a significance test.
One bar per calendar year across the period chosen for this section — the same years the spatial trend below it is fitted over, so the two are comparable. The solid horizontal line is the mean annual total across reporting years: bars above it are wetter than the record mean and bars below it are drier. The dashed line is a least-squares trend fitted against the real year, so a gap in the record neither shifts nor steepens it. The slope is reported per decade, and only where at least three years reported — two points are not a trend. Years with no data stay empty rather than being drawn as zero. The context figures beside it place the selected period among its own peers using an absolute departure from the median and a rank, because percent-of-normal and standardised anomalies are undefined wherever the normal is zero, which is common here.
One least-squares slope per selected spatial unit, in millimetres per decade, fitted over the period chosen for this section — the same years as the chart above, which is what makes “is this trend local or regional?” a question the pair can answer. At least five years are required, because a slope over two or three annual totals is arithmetic rather than a trend. Province and Governorate modes average the qualified rain-gauge network with every enabled precipitation layer that publishes yearly data, year by year over whichever sources answered. Raster mode area-averages every enabled yearly field onto a common 0.25° grid before blending the products; point gauges are excluded because they are not area-average grid cells. A missing year is skipped, never read as zero, and fewer than eight reporting years is marked as a short record. The diverging scale is symmetric about zero — brown drying, teal wetting — and clipped to cover 95% of fitted units so a few extremes cannot flatten the national pattern. This is a description of the archive, not a forecast or significance test.
The first dropdown choice follows the report's current spatial query exactly: the national, province or governorate boundary; the selected raster-cell location; or the user's rectangle or polygon. The other choices group the Kingdom's 13 provinces into six coherent rainfall regimes. For areas, each raster is area-averaged and each day is the equal mean of the qualified ground-gauge network and every enabled precipitation layer that publishes daily data. A location uses the enabled rasters at that cell because a point encloses no gauge network. A source that does not answer is excluded, never read as zero. A rainy day has at least 1 mm, and monthly and annual totals are compared with the same combined 2000–2025 climatology. RX1day and RX5day are the largest one-day and consecutive five-day totals; SDII is precipitation on rainy days divided by their count; R10mm, R20mm and R50mm count threshold exceedances. An event is a run of consecutive rainy days separated by a dry or missing day, and the event bars show how much of each year's total came from the largest one to three events. Ranges report the mean annual index, not one pooled multi-year extreme.