data trove waterfalls worldwide

Overview

Source: /home/coolhand/html/datavis/data_trove/data/geographic/waterfalls/waterfalls_worldwide.json

Saturn profiled 80,678 rows across 9 columns. The stats below are deterministic and machine-readable; the prose is a language-model interpretation of those stats (opt-in, added after the fact, never sees raw rows).

latitude numeric feature

This column contains geographic latitude coordinates, spanning from -77.72° (Antarctic region) to 78.66° (Arctic region), covering nearly the full terrestrial range. With 80,650 unique values out of 80,678 rows and zero nulls, it is essentially a high-cardinality continuous measurement. The distribution is notably left-skewed (skew = -0.94) with a mean of 27.1° and median of 40.3°, indicating a concentration of records in mid-to-high Northern Hemisphere latitudes but with a meaningful tail toward the Southern Hemisphere. The IQR of 37.8° and near-flat kurtosis (-0.28) suggest a broadly spread, roughly uniform distribution rather than a tight cluster.

Treatment: Use as-is for spatial modelling; consider pairing with longitude and binning into geohash or grid cells for aggregation tasks.

longitude numeric feature

This column is geographic longitude, with values spanning nearly the full valid range of −179.99 to 179.41 degrees, indicating globally distributed records. The distribution is notably flat (kurtosis −0.41, IQR of 100.27°) and only mildly right-skewed (skew 0.29), suggesting broad geographic spread rather than concentration in any single region. The median of 7.80° (near Western Europe/West Africa) sits well below the mean of 0.96°, hinting at a slight pull toward Eastern longitudes. Near-perfect uniqueness (80,650 unique values out of 80,678 rows) confirms these are precise coordinate readings, not bucketed regions.

Treatment: Use as-is for spatial modelling; consider pairing with latitude and applying geographic projections or clustering (e.g., H3/geohash) before feeding into non-spatial models.

name text label

This column contains the names of waterfalls or water features, drawn from what appears to be a global geographic dataset (evidenced by multilingual terms: 'Cachoeira'/'Cascada'/'Cascata'/'Fossen'/'Salto'). The dominant signal is that 48,168 of 80,678 rows — nearly 60% — carry the value 'Unnamed Waterfall', driving a duplicate rate of 65.7% and collapsing effective cardinality to just 27,697 unique values out of 80,678 total. The vocab includes Portuguese, Spanish, Norwegian, and English terms, confirming a multilingual mix that an analyst should be aware of when grouping or filtering by name.

Treatment: Treat 'Unnamed Waterfall' as a missing-name sentinel; flag or separate those 48,168 rows before any name-based grouping or NLP embedding.

description categorical label

This column appears to describe a financial or project methodology type, overwhelmingly dominated by 'Waterfall' (72,565 of 80,678 rows, ~89.9%), with the remaining values being 'Waterfall' variants qualified by a time suffix (e.g., '3m', '2m', '5m'). The extreme concentration in a single value — an entropy ratio of only 0.119 — and the long-tail alert indicate that despite 775 unique values, almost all signal is captured by one category. Surprising: with 775 distinct values but ~90% mass in one label, the tail likely contains hundreds of rare or inconsistently formatted variants that may need normalisation.

Treatment: Normalise tail variants (e.g., parse time suffix into a separate numeric feature), then one-hot or ordinal encode; consider collapsing rare variants below a frequency threshold.

category categorical metadata

This column is a dataset category tag, representing the data source or classification for every record — here uniformly 'usgs_waterfalls'. With cardinality of 1, top_rate of 1.0, and zero nulls across all 80,678 rows, it carries no discriminative information whatsoever. This is a constant column, almost certainly a provenance/partition label added when merging multiple source datasets.

Treatment: Drop before modelling — zero-variance constant; retain only if merging with other source datasets where the value varies.

date categorical other

This column is labeled 'date' but contains no actual date values — every single one of its 80,678 rows holds an empty string, giving it a cardinality of 1 and a top_rate of 1.0. The column is entirely blank with zero nulls, meaning missing values were stored as empty strings rather than proper nulls. It carries zero information and will contribute nothing to any analysis or model.

Treatment: Drop this column; it is entirely empty strings with no informational content.

country categorical feature

This column is intended to capture country of origin or residence, with only 6 distinct values across 80,678 rows. The overwhelming surprise is that 99.97% of records (80,650 out of 80,678) contain an empty string rather than a valid country code, making the field effectively unpopulated. The remaining 28 records split across five ISO country codes (VE with 24 occurrences, and DE, LB, HN, BR each with 1), suggesting the field was rarely filled in rather than being systematically captured.

Treatment: Treat empty strings as missing; with 99.97% blank rate this column carries near-zero signal and should be dropped unless the rare non-empty values have specific analytical value.

height categorical feature

This column purports to store height values but is classified as categorical, with 775 unique string values across 80,678 rows. The dominant signal is alarming: 72,565 rows (89.9%) contain an empty string, meaning the field is effectively missing for nearly 9 in 10 records despite a reported null_rate of 0.0. The non-empty values appear to be small integers (e.g., '1', '2', '3', '5', '10', '20'), suggesting height in some discrete unit, but the extreme sparsity and long-tail alert make this column unreliable as a feature without significant imputation or domain clarification.

Treatment: Treat empty strings as missing (true null_rate ≈ 0.90); investigate unit semantics, then impute or drop depending on task requirements before modelling.

source categorical metadata

This column records the data source attribution for all 80,678 rows, and every single record carries the value 'OpenStreetMap' — making it a constant with cardinality of 1, entropy of 0, and a top_rate of 1.0. It provides zero discriminative information and will contribute nothing to any model or analysis. The imbalance alert is technically correct but understates the situation: this is a fully degenerate column, not merely skewed.

Treatment: Drop before modelling; if provenance tracking is needed, note the constant value in dataset documentation instead.