By Austin Little
AI disclosure: Drafted with AI assistance for Apiary. The listening and dashboard comparisons are teaching examples, not reports of any named place’s migration on a particular night.
Turn off the news for ten minutes and step somewhere safe outside on a migration night. You may hear the neighbor’s air conditioner, a truck across town, one note that might belong to a bird, and then nothing definite. Meanwhile, a map of your region indicates birds in flight. The temptation is to settle which source was wrong. Before doing that, look at what each machine, including your ears, was trying to detect.
The map may come from weather radar data processed to estimate the movement of birds across a broad area. The note you heard came from a particular position on the ground. A bird can pass far above you, use a route across another part of the region, call outside the interval you listened, or say nothing you could recognize. That is why a radar map can show a strong migration night that is acoustically quiet from one porch. It is also why a loud little burst of calls won’t tell you how many birds crossed the county. BirdCast: frequently asked questions · Horton and colleagues: radar, thermal imaging and acoustic comparison
Ask what evidence would reach your own site
A prediction about a broad region and a quiet place in it are not two measurements on the same instrument. One describes a phenomenon that may be spatially spread and largely aloft. The other describes what one observer could detect with a particular view or listening range. They can differ even when both are working as intended. This is the distinction to settle before arguing about which instrument is right.
Imagine being told that many vehicles crossed a county during the day, and then looking at one side street for five minutes. You could see none even if the county observation is accurate. The analogy has limits. Birds do not follow road geometry, the map may represent traffic rather than a census, and an acoustic event is not a vehicle count. But the scale difference is useful: a high regional flow does not specify what any short sample at one address will contain.
Name the local outcome before you check the map. “Did I hear at least one sound consistent with a flight call in ten minutes?” is a manageable question. “How many birds passed directly overhead?” is not answerable from unaided hearing alone. “Was this a night with a strong regional radar signal?” is a defensible reading of the relevant observation when coverage and units are clear. One person can record all three questions while recognizing only two may have observable answers.
Start with where you can observe safely and with permission. A quieter site isn't automatically a better one if reaching it means leaving an approved route or crossing a road in the dark. A town that closes its parks is not offering an overnight sample site. The article's examples are about passive observation from an already permitted place. Don't use lights, playback or approaches intended to change where wildlife moves so you'll have a more dramatic result.
Is the map a forecast or a view of the night?
BirdCast offers forecasts and observation-based migration tools. A forecast estimates where and when movement is likely from a prediction model. Its useful counterpart is a weather forecast, not a photograph of the future. The live maps use weather surveillance radar observations during the seasons and overnight periods documented by BirdCast. The local dashboard offers summaries based on those observations for a selected county or state. A forecast and an observation-based estimate may both carry the same date. Confirm which one is under the screenshot before describing what happened. BirdCast: forecast maps · BirdCast: live migration maps
Night adds a calendar problem. At 1 a.m., a map may still be summarizing an event that began the preceding evening. A chart can label a time in the selected area’s time zone; a broad graphic may use another display convention. BirdCast describes local dashboard times and explains its separate Eastern time convention for graphics on the live-map page. Save the date, clock and location the product actually used instead of labeling a screenshot “last night” and hoping a future reader remembers when you took it. BirdCast: migration dashboard and local times · BirdCast: live-map time convention
There is also a difference between a map that stopped getting data and a map full of zeros. Radar stations may be unavailable. In some terrain, radar coverage can be limited even when birds are moving where the signal doesn’t sample them well. Rain and other signals can complicate the estimate. A dashboard may be out of its live season altogether. When unable to find a meaningful estimate, report the missing measure rather than telling everyone the sky was empty. A blank map is still a useful clue about the measurement system if you can establish why it is blank. BirdCast: live map coverage and radar availability
Use a map screen as a dated record, not an interchangeable picture
Record the product name and the date or time label it actually showed. A screenshot of familiar colors may not identify whether you opened a forecast or an observation. If the product updates as new data arrives, include when you read it, not just when you took a walk. The issued time of a prediction, the interval being predicted, the radar observation time, and your listening window belong to separate entries.
Suppose a synthetic evening starts November 4 in a region on local standard time. A listener begins at 23:57 and stops at 00:07. Write a start date and time and an end date and time, both with the chosen time zone. A short notation such as “night of Nov 4, 23:57 to Nov 5, 00:07 local” carries information that “Nov 5 at midnight” doesn't. If the displayed plot follows a different convention, use the product labels and make chronology explicit rather than mixing the timezone and night conventions.
Next note whether the window fits one plotted radar interval, crosses several, or doesn't align with the product resolution. There is no need to interpolate a fine local value from a broader graph just to line up the timestamps. A better comparison may say a ten-minute listening bout fell within a one-hour interval of elevated estimated regional traffic. This is a statement about overlap, not an assertion that identical conditions persisted through all ten minutes.
A common trap is a casually saved snapshot of the whole-night total before the night ends. It might be partial. A later complete total can differ without retroactively invalidating the early observation; the question is what had happened and what was known at the time. If timing or missing coverage isn't clear, mark it unresolved. A product showing limited data isn't the same as a report of little movement. Different histograms can look alike even when one is missing the relevant window.
Keep the method lightweight. A table can record source, viewed time, region, units, timestamp basis, coverage note and qualitative reading. Include a saved written interpretation rather than relying on a screenshot alone. A later reader may not remember what magenta meant on a particular color scale. If the website redraws its design, the plain text will travel better than the colors.
Read the number’s unit before giving the number a headline
The dashboard describes more than one kind of number. A count of birds estimated to have overflown a selected region during the night is a cumulative estimate. A number of birds in flight at a particular moment is an estimate of an airborne population in that area then. One is like the number of people whose route has passed through a station during a shift. The other is like the number occupying the platforms at a moment. Do not compare them as if one were a revision of the other. BirdCast: meanings of the dashboard measures
The live map can show migration traffic rate. BirdCast describes this in birds per kilometer per hour, as traffic across a line perpendicular to the direction of flight. That is not birds per square kilometer standing on the landscape. It’s a rate for movement expressed against a line instead of a count for a whole place. If you find three figures from the same night and one uses a rate, one uses a cumulative crossing estimate and one uses an instant population estimate, you have three different measurements. An honest caption can use only the one it needs and name the rest if they add context. BirdCast: reading the colors and migration traffic rate
Consider a made-up dashboard for a county. At 10 p.m., it estimates 8,000 birds in flight. By morning it estimates 35,000 crossed the county overnight. The first figure does not say every bird present at 10 stayed all night. The second does not say the county has 35,000 new residents. People posting observations from a park in the morning may see migrants using the landscape, but the radar numbers do not tell you where individual birds landed, or whether they landed in the selected region at all.
A further wrinkle: the dashboard’s expected species list is based on a separate source of observations, eBird, and its procedure is not a species breakdown of the radar return. You can read it as context for plausible nocturnal migrants in the place and period, not as a proof that a listed species was present in the airborne estimate on the selected night. If the expected list is missing, check BirdCast’s explanation about data sufficiency. The availability of a useful species prediction and the availability of radar data can differ. BirdCast: expected nocturnal migrants
Work the units without converting them into birds above the porch
A traffic rate can describe how many birds are estimated to cross a kilometer of line oriented across migration in an hour. To reason from rate to a crossing count, you would need a specified transect, time interval, direction and assumptions about whether the rate applies uniformly. A porch doesn't provide a calibrated transect amid a varying corridor. Dividing the rate by sixty and multiplying by listening minutes can look scientific while assuming the local detection geometry away.
For an explicitly fictional arithmetic example, suppose an analysis refers to a uniform rate of 120 birds per kilometer per hour across a hypothetical two-kilometer transect for half an hour. Multiplying the rate by the width and duration gives 120 estimated crossings under those idealized assumptions. This is a unit exercise, not an animal count for any real block. If wind, route direction, place or time changes the flow across the line, the simplified example doesn't apply. A listener's detection area also has no straight boundary corresponding to that transect.
A dashboard's estimated birds over a selected county during a whole night is another quantity. Swapping the preceding calculation into a county total because the numbers share the word birds would be a category error. A snapshot birds-in-flight metric offers a momentary estimate, which cannot automatically be summed across repeated graph samples to get unique individuals. The same moving birds can contribute to more than one momentary display. Keep map units in your notebook exactly as labeled.
When comparing neighborhoods, read denominator and geography before ranking. A large county's night total and a smaller locality's total answer different geographic questions. A high regional value on one night in your selected view doesn't prove the local observer had proportionately better odds of hearing anything. It might help you choose candidate nights for a listening exercise. Use the tool as a context selector rather than a promise that local sensory evidence will scale one for one.
A call is an event, not a counted bird
A nocturnal flight call can be brief enough that a person on the ground only notices it after it is gone. Some calls can be recognized, which makes acoustic study valuable for questions about identity that weather radar may not answer. But simply hearing something doesn’t mean you can assign a species. Some birds don’t make a detectable call while in range. Some may call several times; local traffic or wind can mask notes; a listener may miss a faint sound while writing down the previous one. It’s possible to learn something real by counting sounds without treating the count as individuals. Béasse and colleagues: evaluating acoustic estimates of nocturnal migration
A good first note for a beginner is “one short, high call overhead; source unconfirmed.” It may be tempting to search the expected migrants list for a name and put it there after hearing a rough resemblance. If your evidence isn’t strong enough, known possibilities can stay in a second column. The original observation should still be legible even if that tentative name later changes. Distinguish sounds you heard yourself from sounds in a recording that a reviewer examined later. That difference helps if an account is trying to establish which evidence is available to check.
A recorder and a person also differ. A file can be replayed, and it can preserve a sound the listener missed. It can also have background noise, limited sensitivity and its own placement problems. An empty result from software that classifies sounds means the software did not register a result under its settings. It doesn’t, without reviewing quality and scope, prove the file contains no calls. On the other side, an automatic high-confidence label may include nonspeech or nonbird noise in some circumstances or confuse similar calls. Treat the file as evidence you can review rather than a press release the tool wrote for you.
Keep the ecological stakes small for a first listening visit. There is no need to play bird calls, set up lights, disturb animals or enter a closed site. You can enjoy a quiet listen near home or check historical maps from indoors. If making a recording, respect other people’s privacy and don’t publish incidental conversations. A written field note about what you heard gets you most of the learning without having to capture everything around you.
Work through three different kinds of silent window
Consider three invented sessions. Session A is ten minutes under good listening conditions; the recorder was checked; no candidate call was noted. Session B begins, then a passing vehicle and wind obscure most of the interval. Session C is canceled because the observer cannot use the usual spot. A spreadsheet that writes zero in all three rows erases most of what happened.
Record A as a completed no-detection observation within the stated method. This doesn't establish zero passing birds, but is a meaningful local result. Flag B as interference or partly usable, and include only defensible effective listening time if you can determine it rather than simply naming the planned duration. Mark C as not observed. There is no normal source for an invented detection count in an observation you never made.
Suppose the map shows high estimated regional traffic for all three nights. The first is a clear local non-detection in a strong regional context. The second may teach which noise conditions make the site unproductive. The third simply widens your missing record. None requires throwing out the radar data, and none can be treated as equally strong evidence that birds were quiet. A group can summarize these results without manufacturing exact counts: one complete window with no detections, one interrupted run, one missed outing.
Now reverse the comparison. A person hears two candidate chirps during a night with a relatively weak broad signal. A rare local encounter can coexist with a low regional aggregate. A resident bird, insect, or noise might also be mistaken for a flight call. Record the sounds as observed and the interpretation separately. Don't upgrade a probable sound to a species-level identification solely because an expected species list makes it seem reasonable.
For long-term observation, the interruptions can be as useful as detections. If most weak-signal nights were pleasant and most strong-signal nights were noisy, your notes may reflect access and weather as much as migration. Report what fraction of planned sessions were completed and under which stated conditions. A half-finished field log can support better planning without supporting a detection probability.
Scientists find a mismatch worth explaining, not a reason to throw away a sensor
In Delaware, Kyle Horton and colleagues compared weather radar, thermal imagery and an acoustic recorder over spring and fall migration periods. The strongest movement signals in radar and thermal imagery tended to occur around or shortly after evening twilight, while the acoustic peak tended toward the approach of morning twilight. Agreement varied substantially from night to night. In their measurements, flight height mattered to how each instrument detected movement. These are findings from the study site and design, not a rule telling a listener what hour all birds will be audible on every coast. Horton and colleagues: Delaware comparison study
A study comparing several acoustic metrics with dedicated vertical radar in southwest France addressed some of the same limits with a separate design. Calling patterns, weather and flight height can influence what an acoustic monitor records; choices about which acoustic measure to calculate matter. This is why it helps to identify whether a research report counted calls, attempted to distinguish separate passages or compared diversity. The same microphones can support several metrics. They don’t become interchangeable just because the input files are the same. Béasse and colleagues: acoustic variables and radar
For a local report, these comparisons suggest a useful approach. When sensors disagree, look for a difference in view and time before picking the number that sounds more impressive. Radar estimates can be powerful for describing large patterns, while acoustic evidence can add identity information when calls can be confidently assigned. A source with more detail about one bird species may capture fewer birds overall. A source covering broad movement may say little about individual species. Those are complementary strengths.
The point is not to make the instruments compete for your trust. First decide why you were looking. If you want to understand a regional movement pattern, begin with the regional evidence. If you are documenting a particular sound near where you live, preserve that local record independently. The join between them can be a good story without making either piece carry what it cannot measure.
Make the passive listen comparable without overengineering it
Use a stable start cue and duration. One person can note that ten minutes begin after arriving and setting up. Another can use a fixed offset after local twilight if it is relevant and easy to document. What matters is that the protocol doesn't quietly move to the moment a promising sound first appears. If a spontaneous sound prompts an extra listen, log it as an opportunistic observation separate from the planned count.
Control ordinary disturbances within your own behavior. Put away noisy gear, don't aim lamps into habitat, and stand only where you are already allowed. If conversation, appliances, or movement interfere, mark it instead of deleting the entire session after seeing an inconvenient result. A useful observer log can record audible wind, steady road noise or individual interruptions qualitatively when no calibrated sound measurement is available. Do not call those descriptions decibels.
If using a permitted recorder, preserve settings, placement and times. An automatic gain mode might respond differently after a loud nearby event, making a later faint signal less comparable. A phone screen showing sound activity isn't proof of uniform sensitivity in all directions. Do not set out to collect private conversation or ordinary street activity as sample material. If local rules or privacy conditions make recording inappropriate, ordinary written listening notes can still support a bounded observation.
Review uncertain sounds on their own evidence. A small excerpt may preserve timing and shape, but an app label is a suggestion rather than a final observation. Where multiple people compare a recording, ask them to mark candidate windows, noise and confidence before showing a name suggested by the map's species context. Otherwise an attractive label can turn several independent listeners into repeaters of the same first guess. When no appropriate identification is available, “unidentified candidate sound” is a worthwhile field entry.
A ten-minute comparison you can actually describe later
Choose a safe ordinary place and a listening window. Record the time and time zone, duration, and facts that could affect hearing: wind, intermittent traffic or equipment running nearby. State what events you tried to count. A handwritten tally of probable calls with identity unknown can be more honest than a precise species total you worked backward from later. If you hear nothing recognizable, leave the zero in the right category: no confirmed call during this effort.
Then open the dashboard, using a historical night if the live tool is off season. Record the selected region and date, and identify one number or graph relevant to the window. Do not count a missing time series as quiet. Do not silently move the morning date to the next night if the regional display still groups it with the evening before. If your ten-minute interval does not overlap the chart you can confidently read, say so. That is a more useful result than comparing the wrong hours with great confidence.
A possible finished note is: “Ten-minute listen at one residential location, high road noise in the first five minutes; one probable overhead call, identity unknown. The county dashboard estimated substantial overnight movement. The sound record is much smaller in coverage and does not identify the county’s migrants.” The note can be that simple. It keeps the listening window and the broad estimate alongside one another without presenting one as confirmation of the other.
Suppose you run the same short watch two nights in a row. The first night your note shows three probable calls. The next night you hear six and stop. Doubling the local tally may reflect more birds, more calls per bird, a noisier first night, a lower flight path or random variation across two short windows. While you can record twice as many probable calls, you should not say the migration increased twofold unless the evidence actually supports a migration comparison. Repeating windows, describing conditions and checking broader data would strengthen the next question; they need not erase the simple result from this one.
Use neighborhood comparisons to find confounders first
If two neighbors want to ask whether their observations track the same broad signal, standardize the things they can control: passive recording or listening method, start and end times, no-detection notation, and what qualifies as a candidate. Differences will remain. One porch fronts a treeline; the other is beside a commercial entrance and wakes to a fan. Those are observation conditions, not reasons to discard a site because it produced the wrong answer.
Each neighborhood can keep a simple pair record: site label, session window, concurrent map interval, conditions, usable duration, candidate sound count, and uncertainty. Group reports can omit home addresses. Pick a neutral site ID and say whether participants compare notes before labeling sounds. Differences in experience are part of the method. A novice with strong road noise has a different detection context from a trained observer listening in quiet, even if both follow a clock precisely.
Suppose a fictional four-night comparison shows the quieter site reporting more candidate calls on three nights. Don't announce that three times as many birds migrated over that street. The observations alone don't distinguish better detection from real differences overhead. A useful next step is to repeat under matched timing and, if both places are safe and authorized, have consenting observers swap sites in a later comparison. A change in who listens can help reveal one confounder. It doesn't turn a volunteer series into a calibrated spatial survey.
Consider unseen disagreements too. One site may successfully collect twenty minutes and the other ten due to interruption. A raw tally is not comparable as if effort were equal. Dividing by observed time gives a descriptive candidate-event rate, not an individual-bird rate. If the method encouraged noting only the first event of each cluster, rates from a different protocol won't necessarily line up. State what was counted and what its denominator can do.
When a neighborhood group wants to compare sites
If a nature group has one listener near a busy landing and another on a quiet residential street, give both the same scheduled window before you compare their tallies. Keep the site notes separate: people disagree about what counts as "quiet," but they can agree that loud engine noise obscured the first four minutes of one watch. If one listener can’t make it that night, record a missing session rather than an acoustic zero. That distinction will matter if the group later plots each site’s results. One absent listener should not make a noisy graph seem to show a decline in birds.
A map can supply shared context for the sites. Resist tagging each household with its own estimated share of the county’s flight, because the listening points are not a partition of the county and the radar number is not divided among individual porches. If the intent is to tell a local story, compare what duration and conditions each listener experienced, and use the county chart separately to show why the group picked the night. Site differences may prompt a useful repeat observation. They do not require more certainty than you have about what flew directly over any person.
Keep morning observations out of the overnight denominator
The morning after a strong regional signal can be interesting even if the porch was quiet. A few birds noticed in shrubs after sunrise are observations of a ground site at that time. They might be local residents, newly noticed individuals, or birds using that habitat after migratory movement. A nightly radar visualization generally does not identify the specific visitors to a particular shrub. The morning trip has its own place, timing, duration, and list rules.
Imagine a park opens later than sunrise and the most accessible path passes only along an edge. If observers see few birds, this by itself says little about what was available before opening or deeper in the park. Don't ask anyone to bypass barriers or pursue animals to validate a prediction. A short authorized observation can reveal a local presence or non-detection under that protocol; it doesn't have to explain what happened aloft.
Record a morning note as a separate type. Include target habitat if known, start and stop time, distance if useful, observed animals or sounds within agreed abilities, and visibility or interruption. If contributing to an established citizen-science project, follow its own data and access rules. Do not copy a radar-derived total into a checklist or count likely migrants that were never seen or heard.
The view from a busy morning park is a third piece of the story
Migration researchers may also check which birds people see on the ground. For a visitor, walking a regular park path in the morning after checking an overnight estimate can be a satisfying routine. The tool does not promise a show. Birds that passed overhead may have continued, dropped into habitat elsewhere or been difficult to detect where you walked. Your first few identifiable birds may include residents. Label the area you surveyed and the activity you observed instead of assuming that every earful belongs to the overnight migrants.
Similarly, a busy morning can follow a quiet short listening window. If you see a flock or hear calls from deep in a canopy, you’ve learned something local about that moment. To connect particular birds to a recent flight would require evidence beyond the timing of the sighting. Put the changes next to each other as observations rather than inventing a route for them.
An editor who wants a simple graphic can provide context without overpromising. A useful pairing is one regional chart clearly labeled with date and measure, plus a separate panel with duration and limits of a local watch. If the story includes a species name, say whether it came from an identified call, a ground observation or a separate list of expected migrants. A caption that preserves those sources gives readers a way to check the story and makes the quiet result worth showing.
That is the little trick to enjoying this: you can let your street stay quiet without treating the darkness as empty, and you can let the radar offer a large-scale view without expecting it to tell you which bird went over your roof. What each source misses can become the question you bring to the next night. There is no need to fill every part of the sky with an answer on the first try.
Show what you can conclude from a complete mini-log
Here is a minimal invented report: on Night 1, regional observation showed elevated traffic during the chart interval containing a completed ten-minute listen; no candidate calls heard in a quiet window. On Night 2, the listening window had two unidentified candidate sounds; the relevant radar interval was unavailable. On Night 3, a forecast predicted stronger traffic; no local session was attempted. None of the three rows supports a comparison of local counts between high and low measured traffic.
It does give a good plan for the next outing. Attempt another preselected quiet window with usable regional coverage and write the chosen threshold or qualitative categories in advance. Compare completed cases where the products and windows overlap. Keep forecasts in a planning column; do not retrospectively promote them into measurements because you couldn't obtain the corresponding observation.
Keep the summary modest when observations grow. “Among six completed listening sessions with usable concurrent regional products, two included candidate events” is an honest complete-case count. List interrupted or missed outings elsewhere and don't call the two detections a species total. If the collection was opportunistic, write that too. The practical point is not to collect enough reports to disprove silence. It is to discover which evidence you gathered, what it can say, and what question is still worth asking next time.
Check whether your sample design can hear disagreement
Try a short teaching exercise before anyone goes out. Give half the group a synthetic night chart containing one broad hump and the other half a set of fictional porch logs with both completed and interrupted windows. Ask each reader to write only the claims their sheet supports. Then reveal both sheets and ask for a joint statement. If the statement adds a count of individual birds that nobody measured, the exercise has exposed where the two kinds of evidence got mixed.
A stronger practice pair includes a conflict that cannot be solved in the room. Imagine a chart with a clear regional signal and a quiet local recording with no identified sounds. Resist asking the group to rank two working instruments as one winner and one loser. The right answer can state what each observed and suggest a follow-on check for timing or audibility without assuming which explanation will survive.
Document what evidence would make you revise your account. Finding the chart and porch notes were dated one night apart would change the comparison. Recovering a clear call in a previously unreviewed segment would change the local observation, subject to uncertainty about identification. Neither discovery necessarily changes the regional estimate. A good field note can tell a future reader not only what the group concluded but which claim the new evidence should change.
Sources and field references
- BirdCast: migration dashboard. Definitions, date conventions, coverage and eBird-based expected species.
- BirdCast: live maps. Migration traffic rate, radar sites and interpretation.
- BirdCast: forecast maps. Model predictions versus observations, and movement aloft rather than ground presence.
- BirdCast: frequently asked questions. Limits of radar species identification and everyday interpretation.
- Horton and colleagues: radar, thermal imaging and acoustic recording. Primary comparison of nocturnal detection biases in Delaware.
- Béasse and colleagues: quantifying nocturnal migration using acoustics. Primary study using acoustic measures and dedicated vertical radar.