sylva-rs 2.0.0

Terrestrial and airborne laser scanning processing for forest ecology and remote sensing research.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
// Sylva: LiDAR processing for forest ecology and remote sensing research.
// Copyright (C) 2026 Tim Devereux, The University of Queensland.
// Free software under the GNU General Public License v3.0 or later;
// see the LICENSE file. There is no warranty, to the extent permitted by law.
//! Change between two airborne lidar surveys of one area.
//!
//! The principle is that of [`crate::change`]: every change carries its
//! uncertainty or level of detection, and whatever the data cannot support
//! is labelled rather than reported. Everything runs on catalogues through
//! the chunk engine of [`crate::als`]: each chunk reads the same buffered
//! box from both surveys, and the results do not depend on the chunks or
//! the number of workers.
//!
//! * [`align`]: vertical and horizontal offsets between the surveys on
//!   stable surfaces (ground, and optionally roads or roofs), per block or
//!   as a smooth field, with their uncertainty;
//! * [`surface`]: CHM, DSM and DTM differences with a level of detection per
//!   cell, a comparison of the two sensors, and harmonisation by thinning;
//! * [`gaps`]: canopy gaps of each survey, their formation and closure,
//!   polygons and size distributions;
//! * [`trees`]: airborne trees matched between surveys, height growth,
//!   mortality, damage and recruitment;
//! * [`metrics`]: area-based metrics and plant area index compared.

pub mod align;
pub mod gaps;
pub mod metrics;
pub mod surface;
pub mod trees;

use std::collections::HashMap;

use crate::als::{est_points, in_core, read_chunk, Catalog, Chunk};
use crate::error::{Error, Result};
use crate::{Point, PointCloud};

pub use align::Alignment;

/// ASPRS noise classes, left out of every comparison.
const NOISE: [u8; 2] = [7, 18];

/// Normal quantile of a two-sided confidence level.
pub fn z_of(confidence: f64) -> Result<f64> {
    if !(confidence > 0.0 && confidence < 1.0) {
        return Err(Error::invalid(format!("confidence must be between 0 and 1, got {confidence}")));
    }
    Ok(crate::change::trees::normal_quantile(0.5 + confidence / 2.0))
}

/// Tiles of `cat` whose extent meets `outer`, in catalogue order.
fn files_in(cat: &Catalog, outer: &[f64; 4]) -> Vec<usize> {
    (0..cat.tiles.len())
        .filter(|&i| {
            let t = cat.tiles[i].xy();
            cat.tiles[i].n_points > 0 && t[0] <= outer[2] && outer[0] <= t[2] && t[1] <= outer[3] && outer[1] <= t[3]
        })
        .collect()
}

fn grown(b: &[f64; 4], d: f64) -> [f64; 4] {
    [b[0] - d, b[1] - d, b[2] + d, b[3] + d]
}

/// Points of the second survey expected in a chunk's box grown by `grow`.
pub(crate) fn est_other(cat_b: &Catalog, chunk: &Chunk, grow: f64) -> u64 {
    let outer = grown(&chunk.outer, grow);
    est_points(cat_b, &outer, &files_in(cat_b, &outer))
}

/// Read the second survey over a chunk of the first: every point in the
/// chunk's buffered box grown by `grow`, moved into the first survey's frame
/// by `alignment` (a point `p` becomes `p - offset(p)`), then kept if it lies
/// in the buffered box. `buffer[i]` is true outside the core.
pub(crate) fn read_other(cat_b: &Catalog, chunk: &Chunk, grow: f64, alignment: Option<&Alignment>) -> Result<(PointCloud, Vec<bool>)> {
    let outer = grown(&chunk.outer, grow);
    let files = files_in(cat_b, &outer);
    if files.is_empty() {
        return Ok((PointCloud::default(), Vec::new()));
    }
    let ch = Chunk { index: chunk.index, core: chunk.core, outer, own: None, est_points: est_points(cat_b, &outer, &files), files, name: chunk.name.clone() };
    let mut cloud = read_chunk(cat_b, &ch)?.cloud;
    if let Some(al) = alignment {
        for p in cloud.xyz.iter_mut() {
            let o = al.offset_at(p[0], p[1]);
            p[0] -= o[0];
            p[1] -= o[1];
            p[2] -= o[2];
        }
    }
    let o = chunk.outer;
    let keep: Vec<usize> = (0..cloud.len()).filter(|&i| {
        let p = cloud.xyz[i];
        p[0] >= o[0] && p[0] <= o[2] && p[1] >= o[1] && p[1] <= o[3]
    }).collect();
    let cloud = if keep.len() == cloud.len() { cloud } else { cloud.take(&keep) };
    let buffer = cloud.xyz.iter().map(|p| !in_core(&chunk.core, p[0], p[1])).collect();
    Ok((cloud, buffer))
}

/// Class of every point (0 when the cloud has none).
pub(crate) fn classes(cloud: &PointCloud) -> Vec<u8> {
    match cloud.attr("classification") {
        Some(c) => (0..cloud.len()).map(|i| c.get_f64(i) as u8).collect(),
        None => vec![0; cloud.len()],
    }
}

/// The cloud without noise (classes 7 and 18).
pub(crate) fn without_noise(cloud: PointCloud, buffer: Vec<bool>) -> (PointCloud, Vec<bool>) {
    let cls = classes(&cloud);
    if !cls.iter().any(|c| NOISE.contains(c)) {
        return (cloud, buffer);
    }
    let keep: Vec<usize> = (0..cloud.len()).filter(|&i| !NOISE.contains(&cls[i])).collect();
    let b = keep.iter().map(|&i| buffer[i]).collect();
    (cloud.take(&keep), b)
}

/// Whether each point is a first return (all are without return numbers).
pub(crate) fn first_returns(cloud: &PointCloud) -> Vec<bool> {
    match cloud.attr("return_number") {
        Some(r) => (0..cloud.len()).map(|i| r.get_f64(i) <= 1.0).collect(),
        None => vec![true; cloud.len()],
    }
}

// ------------------------------------------------------------------ thinning

/// splitmix64 (Steele, Lea and Flood 2014): a well-mixed 64-bit hash.
pub(crate) fn mix(mut z: u64) -> u64 {
    z = z.wrapping_add(0x9E37_79B9_7F4A_7C15);
    z = (z ^ (z >> 30)).wrapping_mul(0xBF58_476D_1CE4_E5B9);
    z = (z ^ (z >> 27)).wrapping_mul(0x94D0_49BB_1331_11EB);
    z ^ (z >> 31)
}

/// A uniform number in `[0, 1)` from a hash.
pub(crate) fn unit(h: u64) -> f64 {
    (h >> 11) as f64 / (1u64 << 53) as f64
}

/// Fewest first returns for the check that GPS times tell pulses apart.
const TIME_CHECK_MIN: usize = 16;
/// GPS times do not tell pulses apart when the first returns outnumber the
/// distinct `(gps_time, point_source_id)` values among them by more than
/// this factor.
const TIME_CHECK_RATIO: usize = 8;

/// Whether `gps_time` (with the flight line) tells the pulses of `cloud`
/// apart. A file written without times carries one value (usually 0) on
/// every point, and one truncated to a coarse clock repeats each value over
/// many pulses; either way thousands of first returns share a handful of
/// times. True without a `gps_time` attribute is never returned: there is
/// then nothing to judge.
pub(crate) fn times_identify_pulses(cloud: &PointCloud) -> bool {
    let Some(t) = cloud.attr("gps_time") else { return false };
    let psid = cloud.attr("point_source_id");
    let first = first_returns(cloud);
    let mut seen: std::collections::HashSet<(u64, u64)> = std::collections::HashSet::new();
    let mut n_first = 0usize;
    for i in 0..cloud.len() {
        if first[i] {
            n_first += 1;
            seen.insert((t.get_f64(i).to_bits(), psid.map_or(0, |p| p.get_f64(i) as u64)));
        }
    }
    n_first < TIME_CHECK_MIN || seen.len() * TIME_CHECK_RATIO >= n_first
}

/// A key identifying each point's pulse: its GPS time and flight line, or
/// the point's own coordinates (to the millimetre) when the cloud has no
/// GPS time or its GPS times do not tell pulses apart
/// ([`times_identify_pulses`]: all zero, constant, or truncated). With the
/// coordinates as key every return is its own pulse, so the returns of one
/// pulse are no longer kept together; that is the documented fallback, and
/// far better than treating a whole flight line as one pulse.
pub(crate) fn pulse_keys(cloud: &PointCloud) -> Vec<u64> {
    let psid = cloud.attr("point_source_id");
    match cloud.attr("gps_time").filter(|_| times_identify_pulses(cloud)) {
        Some(t) => (0..cloud.len()).map(|i| mix(t.get_f64(i).to_bits() ^ psid.map_or(0, |p| (p.get_f64(i) as u64) << 48))).collect(),
        None => cloud.xyz.iter().map(|p| {
            let q = |v: f64| (v * 1000.0).round() as i64 as u64;
            mix(q(p[0]) ^ mix(q(p[1]) ^ mix(q(p[2]))))
        }).collect(),
    }
}

/// Settings of harmonisation by thinning.
#[derive(Debug, Clone, Copy, PartialEq)]
pub struct Harmonise {
    /// Side of the square cells (m, on a grid through the origin) in which
    /// pulse densities are compared.
    pub cell: f64,
    /// Seed of the pulse selection.
    pub seed: u64,
}

impl Harmonise {
    /// Checks that `cell` is a positive number of metres.
    pub fn check(&self) -> Result<()> {
        if !(self.cell.is_finite() && self.cell > 0.0) {
            return Err(Error::invalid(format!("density_cell must be a positive number of metres, got {}", self.cell)));
        }
        Ok(())
    }
}

fn density_key(x: f64, y: f64, cell: f64) -> (i64, i64) {
    ((x / cell).floor() as i64, (y / cell).floor() as i64)
}

/// First returns (pulses) of each density cell.
fn pulse_counts(cloud: &PointCloud, cell: f64) -> HashMap<(i64, i64), f64> {
    let first = first_returns(cloud);
    let mut m: HashMap<(i64, i64), f64> = HashMap::new();
    for (i, p) in cloud.xyz.iter().enumerate() {
        if first[i] {
            *m.entry(density_key(p[0], p[1], cell)).or_default() += 1.0;
        }
    }
    m
}

/// Which points of `own` survive thinning towards the pulse density of
/// `other`: in each density cell where `own` has more pulses, each pulse is
/// kept with probability `n_other / n_own`, decided by a hash of the pulse
/// and the seed, so that every return of a pulse shares its fate and the
/// choice does not depend on how the area was divided. A pulse belongs to
/// the cell of its first return (of its own position when that is not in
/// the cloud).
pub(crate) fn thin_mask(own: &PointCloud, other: &PointCloud, h: &Harmonise) -> Vec<bool> {
    let (n_own, n_other) = (pulse_counts(own, h.cell), pulse_counts(other, h.cell));
    let prob = |x: f64, y: f64| -> f64 {
        let k = density_key(x, y, h.cell);
        let a = n_own.get(&k).copied().unwrap_or(0.0);
        let b = n_other.get(&k).copied().unwrap_or(0.0);
        if a > b && a > 0.0 { b / a } else { 1.0 }
    };
    let keys = pulse_keys(own);
    let first = first_returns(own);
    let mut pulse_p: HashMap<u64, f64> = HashMap::new();
    for (i, p) in own.xyz.iter().enumerate() {
        if first[i] {
            pulse_p.insert(keys[i], prob(p[0], p[1]));
        }
    }
    let seed = mix(h.seed ^ 0x5EED_5EED_5EED_5EED);
    (0..own.len())
        .map(|i| {
            let p = pulse_p.get(&keys[i]).copied().unwrap_or_else(|| prob(own.xyz[i][0], own.xyz[i][1]));
            p >= 1.0 || unit(mix(keys[i] ^ seed)) < p
        })
        .collect()
}

/// Thin both clouds of a chunk towards each other's pulse density.
pub(crate) fn harmonise_pair(a: (PointCloud, Vec<bool>), b: (PointCloud, Vec<bool>), h: &Harmonise) -> ((PointCloud, Vec<bool>), (PointCloud, Vec<bool>)) {
    let ka = thin_mask(&a.0, &b.0, h);
    let kb = thin_mask(&b.0, &a.0, h);
    let apply = |(c, buf): (PointCloud, Vec<bool>), keep: &[bool]| -> (PointCloud, Vec<bool>) {
        let idx: Vec<usize> = (0..c.len()).filter(|&i| keep[i]).collect();
        let b2 = idx.iter().map(|&i| buf[i]).collect();
        (c.take(&idx), b2)
    };
    (apply(a, &ka), apply(b, &kb))
}

// ------------------------------------------------------------------ neighbours and planes

/// A 2-D bucket grid over points, for radius searches whose results come
/// back in index order (so sums over them do not depend on a tree's shape).
pub(crate) struct Grid2 {
    cell: f64,
    map: HashMap<(i64, i64), Vec<usize>>,
}

impl Grid2 {
    pub(crate) fn new(points: &[Point], cell: f64) -> Grid2 {
        let mut map: HashMap<(i64, i64), Vec<usize>> = HashMap::new();
        for (i, p) in points.iter().enumerate() {
            map.entry(((p[0] / cell).floor() as i64, (p[1] / cell).floor() as i64)).or_default().push(i);
        }
        Grid2 { cell, map }
    }

    /// Indices of the points within `r` (in x, y) of `(x, y)`, ascending.
    pub(crate) fn within(&self, points: &[Point], x: f64, y: f64, r: f64) -> Vec<usize> {
        let k = (r / self.cell).ceil() as i64;
        let (cx, cy) = ((x / self.cell).floor() as i64, (y / self.cell).floor() as i64);
        let mut out = Vec::new();
        for dx in -k..=k {
            for dy in -k..=k {
                if let Some(v) = self.map.get(&(cx + dx, cy + dy)) {
                    out.extend(v.iter().copied().filter(|&i| (points[i][0] - x).hypot(points[i][1] - y) <= r));
                }
            }
        }
        out.sort_unstable();
        out
    }

    /// Distance (in x, y) to the nearest point, searching out to `max`;
    /// infinity if none is that near.
    pub(crate) fn nearest(&self, points: &[Point], x: f64, y: f64, max: f64) -> f64 {
        let (cx, cy) = ((x / self.cell).floor() as i64, (y / self.cell).floor() as i64);
        let kmax = (max / self.cell).ceil() as i64 + 1;
        let mut best = f64::INFINITY;
        for k in 0..=kmax {
            // Ring k of cells around the centre cell.
            for dx in -k..=k {
                for dy in -k..=k {
                    if dx.abs() != k && dy.abs() != k {
                        continue;
                    }
                    if let Some(v) = self.map.get(&(cx + dx, cy + dy)) {
                        for &i in v {
                            best = best.min((points[i][0] - x).hypot(points[i][1] - y));
                        }
                    }
                }
            }
            // Every point in a farther ring is at least k cells away.
            if best <= k as f64 * self.cell {
                break;
            }
        }
        if best <= max { best } else { f64::INFINITY }
    }
}

/// Sort points canonically (x, then y, then z), so that index order is a
/// property of the points and not of how they were read.
pub(crate) fn canonical(mut pts: Vec<Point>) -> Vec<Point> {
    pts.sort_unstable_by(|a, b| a[0].total_cmp(&b[0]).then(a[1].total_cmp(&b[1])).then(a[2].total_cmp(&b[2])));
    pts
}

/// A plane `z = h + gx (x - x0) + gy (y - y0)` fitted by least squares, with
/// the RMS of its residuals and the covariance of its gradient
/// `[[var gx, cov], [cov, var gy]]` from that RMS.
#[derive(Debug, Clone, Copy, PartialEq)]
pub(crate) struct Plane {
    pub h: f64,
    pub gx: f64,
    pub gy: f64,
    pub rms: f64,
    pub n: usize,
    pub cov_g: [[f64; 2]; 2],
    /// Variance of `h` per unit residual variance.
    pub var_h: f64,
}

/// Inverse of a 3 x 3 matrix; None if (numerically) singular.
pub(crate) fn inv3(m: &[[f64; 3]; 3]) -> Option<[[f64; 3]; 3]> {
    let c = |i: usize, j: usize| {
        let (r0, r1) = ((i + 1) % 3, (i + 2) % 3);
        let (c0, c1) = ((j + 1) % 3, (j + 2) % 3);
        m[r0][c0] * m[r1][c1] - m[r0][c1] * m[r1][c0]
    };
    let det = m[0][0] * c(0, 0) + m[0][1] * c(0, 1) + m[0][2] * c(0, 2);
    let diag = (m[0][0] * m[1][1] * m[2][2]).abs();
    if !(det.is_finite() && diag > 0.0 && det.abs() > 1e-12 * diag) {
        return None;
    }
    let mut out = [[0.0; 3]; 3];
    for (i, row) in out.iter_mut().enumerate() {
        for (j, v) in row.iter_mut().enumerate() {
            *v = c(j, i) / det;
        }
    }
    Some(out)
}

/// Fit a plane through `idx` of `pts`, centred on `(x0, y0)`.
pub(crate) fn fit_plane(pts: &[Point], idx: &[usize], x0: f64, y0: f64) -> Option<Plane> {
    let n = idx.len();
    if n < 3 {
        return None;
    }
    let mut m = [[0.0; 3]; 3];
    let mut b = [0.0; 3];
    for &i in idx {
        let p = pts[i];
        let row = [1.0, p[0] - x0, p[1] - y0];
        for r in 0..3 {
            for c in 0..3 {
                m[r][c] += row[r] * row[c];
            }
            b[r] += row[r] * p[2];
        }
    }
    let inv = inv3(&m)?;
    let s = [0, 1, 2].map(|i| inv[i][0] * b[0] + inv[i][1] * b[1] + inv[i][2] * b[2]);
    let ss: f64 = idx.iter().map(|&i| {
        let p = pts[i];
        let e = p[2] - (s[0] + s[1] * (p[0] - x0) + s[2] * (p[1] - y0));
        e * e
    }).sum();
    let rms = if n > 3 { (ss / (n - 3) as f64).sqrt() } else { 0.0 };
    let v = rms * rms;
    Some(Plane { h: s[0], gx: s[1], gy: s[2], rms, n, cov_g: [[v * inv[1][1], v * inv[1][2]], [v * inv[2][1], v * inv[2][2]]], var_h: inv[0][0] })
}

/// Median of a slice (NaN when empty); sorts it.
pub(crate) fn median(v: &mut [f64]) -> f64 {
    if v.is_empty() {
        return f64::NAN;
    }
    v.sort_unstable_by(|a, b| a.total_cmp(b));
    let n = v.len();
    if n % 2 == 1 { v[n / 2] } else { 0.5 * (v[n / 2 - 1] + v[n / 2]) }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::pointcloud::Attr;

    #[test]
    #[allow(clippy::needless_range_loop)]
    fn planes_and_inverses() {
        let pts: Vec<Point> = (0..25).map(|k| {
            let (x, y) = ((k % 5) as f64, (k / 5) as f64);
            [x, y, 1.0 + 0.2 * x - 0.1 * y]
        }).collect();
        let idx: Vec<usize> = (0..25).collect();
        let p = fit_plane(&pts, &idx, 2.0, 2.0).unwrap();
        assert!((p.h - (1.0 + 0.4 - 0.2)).abs() < 1e-12 && (p.gx - 0.2).abs() < 1e-12 && (p.gy + 0.1).abs() < 1e-12 && p.rms < 1e-12);
        let m = [[4.0, 1.0, 0.5], [1.0, 3.0, 0.2], [0.5, 0.2, 2.0]];
        let inv = inv3(&m).unwrap();
        for i in 0..3 {
            for j in 0..3 {
                let v: f64 = (0..3).map(|k| m[i][k] * inv[k][j]).sum();
                assert!((v - if i == j { 1.0 } else { 0.0 }).abs() < 1e-12);
            }
        }
        assert!(inv3(&[[1.0, 2.0, 3.0], [2.0, 4.0, 6.0], [0.0, 0.0, 1.0]]).is_none());
    }

    #[test]
    fn grid_searches_match_brute_force() {
        let pts: Vec<Point> = (0..400).map(|k| [((k * 37) % 101) as f64 * 0.13, ((k * 53) % 97) as f64 * 0.11, 0.0]).collect();
        let g = Grid2::new(&pts, 0.7);
        for &(x, y) in &[(3.0, 4.0), (0.0, 0.0), (12.9, 10.4)] {
            let got = g.within(&pts, x, y, 1.1);
            let want: Vec<usize> = (0..pts.len()).filter(|&i| (pts[i][0] - x).hypot(pts[i][1] - y) <= 1.1).collect();
            assert_eq!(got, want);
            let near = pts.iter().map(|p| (p[0] - x).hypot(p[1] - y)).fold(f64::INFINITY, f64::min);
            assert_eq!(g.nearest(&pts, x, y, 50.0), near);
        }
        assert_eq!(g.nearest(&pts, 100.0, 100.0, 1.0), f64::INFINITY);
    }

    #[test]
    fn constant_gps_times_do_not_make_one_pulse_of_a_flight_line() {
        // 400 single returns, all with gps_time 0 (a file written without
        // times): each must stay its own pulse.
        let xyz: Vec<Point> = (0..400).map(|k| [(k % 20) as f64 * 0.5, (k / 20) as f64 * 0.5, 10.0 + (k % 7) as f64]).collect();
        let mut c = PointCloud::new(xyz.clone());
        c.attrs.insert("return_number".into(), Attr::U8(vec![1; 400]));
        c.attrs.insert("gps_time".into(), Attr::F64(vec![0.0; 400]));
        assert!(!times_identify_pulses(&c));
        let keys = pulse_keys(&c);
        let distinct: std::collections::HashSet<u64> = keys.iter().copied().collect();
        assert_eq!(distinct.len(), 400);
        // The same keys as a cloud without the attribute.
        assert_eq!(keys, pulse_keys(&PointCloud::new(xyz)));
        // Real times are used: both returns of a pulse share a key.
        c.attrs.insert("gps_time".into(), Attr::F64((0..400).map(|k| 1e5 + (k / 2) as f64 * 1e-5).collect()));
        c.attrs.insert("return_number".into(), Attr::U8((0..400).map(|k| 1 + (k % 2) as u8).collect()));
        assert!(times_identify_pulses(&c));
        let keys = pulse_keys(&c);
        assert!(keys[0] == keys[1] && keys[1] != keys[2]);
    }

    #[test]
    fn thinning_matches_densities_and_keeps_pulses_whole() {
        // Epoch a: one pulse per 0.1 m along x in [0, 20); epoch b: one per 0.25 m.
        let mk = |step: f64, t0: f64| {
            let n = (20.0 / step) as usize;
            let mut xyz = Vec::new();
            let (mut rn, mut nr, mut t) = (Vec::new(), Vec::new(), Vec::new());
            for k in 0..n {
                let x = k as f64 * step;
                for r in 1..=2u8 {
                    xyz.push([x, 0.5, 10.0 / r as f64]);
                    rn.push(r);
                    nr.push(2u8);
                    t.push(t0 + k as f64);
                }
            }
            let mut c = PointCloud::new(xyz);
            c.attrs.insert("return_number".into(), Attr::U8(rn));
            c.attrs.insert("number_of_returns".into(), Attr::U8(nr));
            c.attrs.insert("gps_time".into(), Attr::F64(t));
            c
        };
        let (a, b) = (mk(0.1, 0.0), mk(0.25, 1e6));
        let h = Harmonise { cell: 10.0, seed: 1 };
        let ka = thin_mask(&a, &b, &h);
        let kb = thin_mask(&b, &a, &h);
        assert!(kb.iter().all(|&k| k), "the sparser epoch is not thinned");
        let kept = ka.iter().filter(|&&k| k).count() as f64 / 2.0;
        assert!((kept - 80.0).abs() < 20.0, "{kept} pulses kept of 200, expected about 80");
        for i in (0..a.len()).step_by(2) {
            assert_eq!(ka[i], ka[i + 1], "both returns of a pulse share its fate");
        }
        assert_eq!(ka, thin_mask(&a, &b, &h));
    }
}