Forests and lakes as areas, big datasets clipped to your place

Changes since 2.15.1.

#New

• Places as areas. When OrionAI looks up a named forest, lake, wetland, park or reserve, it now adds the place's outline instead of a single point. Questions that need an area work straight away: "Track NDVI month by month in Karura Forest" measures the forest itself, and land cover or water around a lake is summarised inside the lake's shape. Dams, towns, peaks and river gauges still come in as points, ready for watersheds and buffers.
• Large open datasets clipped to your area. Country-wide OpenStreetMap layers, such as every road in Kenya, used to fail with "too large to copy". OrionAI now cuts them to the area you ask about, for example "Map the roads of Nairobi". The first time such a dataset is used, Orion prepares it, which takes a few minutes; OrionAI tells you, and loading it again afterwards takes seconds. When an area holds more features than a map layer takes (Nairobi has about 93,000 road segments), OrionAI keeps the ones that matter most at that scale, main roads or rivers first and then the longest, and says that the smallest are left out. Downloads always give you the full dataset.
• The wards of a county, the sub-counties of a district. "Show the wards of Nairobi county" or "Average NDVI per sub-county in Machakos" now loads only that area's units, each tagged with the county it sits in. Before, OrionAI loaded every ward in the country and had no reliable way to pick the right ones, because the published ward and county borders don't line up exactly.

#Improved

• Fewer wasted steps. When OrionAI gives a setting a slightly different name (for example "start" and "end" for the two periods of a change map, or "monthly" for a time series), the step now runs instead of failing and being retried.
• Clearer land cover from satellite imagery. When OrionAI groups an image into land cover classes without training data, clusters of the same kind are now one class (no more "Dense vegetation 1" to "Dense vegetation 5"), and the colours say what each class is: blue for water, greens for vegetation, tans for bare or built-up ground.
• Bigger layers from the Orion catalogue. A catalogue layer now brings up to 10,000 features onto the map (was 2,000), so "How many schools are there in each sub-county of Nairobi?" counts every school, not the first 2,000.
• Clipping mixed layers. Clipping, intersecting or erasing a layer that mixes lines and areas (OpenStreetMap roads include a few area roads) now keeps every feature. Before, only the areas survived, so "roads of Nairobi" clipped to the county came back with 23 roads.
• Clearer messages, fewer retries. Asking for every road in a whole country now says to zoom to the place first, and a query that names a layer the wrong way is told the right table names.
• More time for big downloads. Loading open data may now take up to two and a half minutes before OrionAI gives up on it, so large layers finish instead of being retried.
• Points of interest from Orion data. Requests for airports, markets, schools, hospitals, roads or rivers go straight to Orion's OpenStreetMap data, instead of searching satellite image catalogues.

#Fixed

• Boundaries when geoBoundaries is down. Country, county and district boundaries now load from geoBoundaries' copy on GitHub when the geoBoundaries website is unavailable, and Orion keeps the boundaries it has loaded across updates. On 31 October the website stopped answering and OrionAI could not load any boundaries.
• Orion tools on a busy server. When the server was busy, Map Studio sometimes started a request before the Orion data tools were ready, and OrionAI replied that they were not available. It now waits for them.