data trove temperature anomalies nasa giss

Overview

Source: /home/coolhand/html/datavis/data_trove/environmental/temperature_anomalies/temperature_anomalies_1880_2015.csv

Saturn profiled 146 rows across 19 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).

Year numeric timestamp

This column contains calendar years spanning from 1880 to 2025, with every one of the 146 rows holding a distinct year value — making it effectively a year-level time index with no gaps or repeats. The distribution is perfectly symmetric (skew = 0.0, median = mean = 1952.5) and platykurtic (kurtosis ≈ −1.2), consistent with near-uniform coverage across the 145-year range. The IQR of 72.5 years and zero outlier rate reinforce that years are evenly spread with no clustering. The inclusion of 2025 suggests the dataset extends to the present or near-present.

Treatment: Use as a time index for temporal modelling or trend analysis; consider encoding as relative offset from a baseline year if used as a numeric feature.

Jan numeric feature

This column, labelled 'Jan', almost certainly represents January returns (or another January-specific numeric metric) for 146 observations, likely financial assets or funds given the monthly naming convention and value range of −0.81 to 1.38. Values cluster around a near-zero median (−0.01) with a mean of 0.078, consistent with monthly percentage returns (e.g., −81% to +138% if fractional, or −0.81% to +1.38% if already in percent). The distribution is only mildly right-skewed (skew = 0.65) and near-mesokurtic (kurtosis ≈ 0.01), with just 4 outliers and 93 unique values out of 146 rows—suggesting moderate repetition but no severe data quality issues.

Treatment: Use as-is or standardise; mild right skew does not warrant log-transform, but verify unit scale (fractional vs. percentage) before modelling.

Feb numeric feature

This column likely represents February returns or growth rates (e.g., monthly percentage changes) for 146 entities, given its name and bounded numeric range of -0.63 to 1.44. The mean (0.085) sits noticeably above the median (-0.035), consistent with the positive skew of 0.78 and a right tail pulled by a max of 1.44. With 98 unique values across 146 rows and only 2 outliers, the distribution is fairly well-behaved, though the wide IQR of 0.63 suggests substantial cross-entity variability. The values appear to be expressed as decimals (e.g., -0.63 = -63% or -0.63 percentage points), which an analyst should confirm before use.

Treatment: Confirm units (decimal fraction vs. percentage points), then use directly or apply mild winsorization given the 2 outliers before modelling.

Mar numeric feature

This column ('Mar') likely represents March returns or price changes for a financial asset or portfolio, expressed as decimal fractions (e.g., 0.10 ≈ 10%). The distribution is moderately right-skewed (skew ≈ 0.91) with a median of only 0.015 despite a mean of 0.104, indicating most observations cluster near zero with a long positive tail reaching 1.39. The range of -0.64 to 1.39 is plausible for monthly returns across diverse securities or years, and only 4 outliers are flagged. The 87 unique values out of 146 rows suggest repeated return levels, consistent with a cross-sectional or panel dataset.

Treatment: Use as-is or apply mild winsorization at the 1st/99th percentile to dampen the 4 outliers before modelling.

Apr numeric feature

This column named 'Apr' almost certainly represents April monthly returns (proportional or percentage, given values ranging from -0.60 to 1.31), likely for a set of financial assets or funds across 146 observations. The mean (0.076) sits well above the median (-0.015), driven by a positive skew of 0.82 and a max of 1.31, suggesting a handful of strongly positive April months pull the distribution rightward. With 89 unique values across 146 rows and only 3 outliers, the data is reasonably well-behaved, though the wide IQR of 0.5475 and std of 0.42 indicate substantial variability in April performance.

Treatment: Use as-is or apply mild winsorization at the 1.31 max before regression/modelling given positive skew.

May numeric feature

This column likely represents a May-specific return, change, or rate (e.g., monthly financial return or growth rate for May), given its name and the fact that values span -0.56 to 1.15 with a mean near zero (0.065). The distribution is moderately right-skewed (skew ≈ 0.797) with a median of -0.035, suggesting most observations cluster slightly below zero while a tail of positive values pulls the mean up. With only 95 unique values across 146 rows, some values repeat, and 2 outliers at the upper end (max 1.15) may warrant inspection.

Treatment: Use as-is or apply mild Winsorization at the 1.37% outlier boundary before modelling.

Jun numeric feature

This column, named 'Jun', almost certainly represents a June monthly return or change figure (likely percentage or decimal), given its range of -0.52 to 1.2 and mean near zero (0.053). With 146 rows but only 94 unique values, there is moderate value repetition that may reflect rounding to two decimal places. The distribution is mildly right-skewed (skew 0.853) with near-zero kurtosis (-0.115), suggesting a roughly bell-shaped spread with a slight positive tail; 3 outliers at the upper end (max 1.2) stand out given the otherwise tight IQR of 0.515.

Treatment: Use as-is or apply mild winsorization at the upper tail before modelling given the 3 high outliers.

Jul numeric feature

This column almost certainly represents a July monthly return or change figure (likely percentage or decimal-fraction), given its name and the range of values spanning −0.52 to 1.20. The mean (≈0.077) sits notably above the median (−0.03), signalling moderate positive skew (skew ≈ 0.99) — a handful of large positive months are pulling the mean up. With only 87 unique values across 146 rows, there is meaningful repetition, and 3 outliers (max = 1.20) are worth investigating as potential data-entry errors or genuinely extreme events.

Treatment: Use as-is or winsorise at the 97th percentile to dampen the 3 outliers before modelling.

Aug numeric feature

This column almost certainly represents August monthly returns or percentage changes for 146 financial instruments or time-series observations, given its name and the presence of signed decimal values ranging from -0.55 to 1.29. The mean (0.077) sits notably above the median (-0.04), and the skew of ~1.0 indicates a right-leaning distribution with a few strong positive outliers driving the mean upward. With only 88 unique values across 146 rows, there is moderate value repetition, and 3 outliers (~2%) at the upper tail (max 1.29) warrant attention. The IQR of 0.52 around a near-zero median is consistent with month-level return data expressed as decimals or percentages.

Treatment: Winsorize or cap the 3 outliers at the 99th percentile before using as a model feature; consider log-transform or standardization if combining with other monthly return columns.

Sep numeric feature

This column, named 'Sep', likely represents September monthly returns or price changes for 146 financial instruments or time periods, expressed as decimal fractions (e.g., -0.58 to 1.48). The mean (0.083) sits well above the median (-0.055), consistent with the positive skew of 1.11, indicating a right-tailed distribution pulled by a few strong positive months. With only 87 unique values across 146 rows, there is notable repetition, and 4 outliers (2.7% of observations) drive the upper tail toward the 1.48 maximum. The near-zero zero_rate (0.007) confirms this is a continuous return series, not a count or categorical encoding.

Treatment: Use as-is or winsorise at the 97th percentile to dampen the 4 outliers before modelling.

Oct numeric feature

This column likely represents October returns or percentage changes (e.g., monthly asset or portfolio returns), given its name 'Oct' and values spanning -0.58 to 1.34 with a mean near zero (0.107). The distribution is right-skewed (skew = 1.05) with a notably low median of 0.01 versus a mean of 0.107, suggesting a few large positive values are pulling the average up — confirmed by 8 detected outliers (5.5% of rows) and a max of 1.34 that sits well above Q3 (0.2675). The IQR of 0.465 and standard deviation of 0.403 indicate meaningful spread for what appear to be decimal-fraction returns.

Treatment: Winsorize or clip outliers at the 1st/99th percentile before modelling; consider log-transform if used as a target.

Nov numeric feature

This column appears to represent a November monthly return or change metric (likely financial, given the decimal scale and negative values). Values range from -0.57 to 1.40 with a median of 0.02, suggesting most observations cluster near zero — consistent with percentage returns expressed as decimals or similar bounded financial data. The distribution is positively skewed (skew = 1.02) with a mean of ~0.10 notably above the median of 0.02, driven by 6 outliers on the upper tail. The IQR of 0.465 spanning negative to positive territory indicates substantial variability across observations.

Treatment: Investigate and cap or Winsorize the 6 upper outliers before modelling; consider standardizing given skew of 1.02.

Dec numeric feature

This column likely represents a decimal-valued return, change, or score measured in December (or abbreviated 'Dec' for December), given its signed numeric range and the naming convention. Values span -0.82 to 1.37 with a near-zero median (-0.04) and mean (0.073), consistent with percentage returns or normalized deltas. The distribution is modestly right-skewed (skew ≈ 0.76) with platykurtic tails (kurtosis ≈ 0.26), and only 2 outliers out of 146 rows — a relatively clean signal. The 91 unique values out of 146 rows suggests some rounding or bucketing in the source data.

Treatment: Use as-is or apply mild standardisation; check for temporal context (is this a monthly slice?) before including in time-aware models.

J-D numeric feature

This column likely represents a January–December (annual) temperature anomaly or similar climate index, where values indicate deviation from a baseline (negative = below average, positive = above average). The range of -0.49 to 1.28 and median of -0.03 are consistent with global mean temperature anomaly records in tenths-of-a-degree or degree units. Notably, the distribution is positively skewed (skew = 0.98) with the mean (0.082) pulled well above the median (-0.03), suggesting a tail of increasingly warm years — a pattern characteristic of a long climate record spanning pre- and post-industrial warming. Only 83 unique values across 146 rows indicates rounding or repeated anomaly values.

Treatment: Use directly as a numeric feature; consider checking temporal ordering and testing for trend/drift given the positive skew.

D-N numeric feature

This column, labelled 'D-N', appears to represent a signed numeric difference or delta score, likely a day-minus-night (or before-minus-after) measurement given its name and bipolar range of -0.5 to 1.29. With a median of -0.04 and mean of 0.082, values are centred near zero but right-skewed (skew ≈ 0.97), pulling the mean above the median. The IQR of 0.52 spanning Q1 = -0.22 to Q3 = 0.30 is relatively tight, but 3 outliers reach up to 1.29, and 86 unique values across 146 rows suggests genuine continuous variation rather than a categorical encoding.

Treatment: Use as-is or apply a mild transform (e.g. Yeo-Johnson) to reduce right skew before regression or classification.

DJF numeric feature

This column likely represents a December–January–February (DJF) seasonal anomaly or index value, such as a climate variable averaged over the boreal winter season. Values range from -0.68 to 1.36 with a median of -0.03 and mean of 0.078, consistent with standardized or near-standardized anomaly data centered near zero. The distribution is mildly right-skewed (skew = 0.80) with near-mesokurtic shape (kurtosis = 0.19), suggesting a mostly normal spread with a slight positive tail; only 2 outliers are flagged. With 94 unique values across 146 rows, there is moderate repetition, plausible for rounded seasonal averages.

Treatment: Use as-is in modelling; mild skew (0.80) does not warrant transformation, but verify units and whether values are already standardized anomalies.

MAM numeric feature

MAM is a numeric column with 146 observations spanning [-0.58, 1.28], likely representing a seasonal or period-averaged anomaly, index, or return (March–April–May being a common meteorological or financial aggregation window). The distribution is near-symmetric around a median of -0.015, with mild positive skew (0.87) and near-zero excess kurtosis (0.04), suggesting an approximately normal spread. Only 2 outliers are flagged (1.37% of rows) and the IQR of 0.568 is consistent with the std of 0.42, indicating no extreme tail behaviour. The mean (0.081) sitting above the median (-0.015) is consistent with the modest right skew.

Treatment: Use as-is in modelling; mild skew does not warrant transformation, but verify units and whether it represents an anomaly or raw aggregate.

JJA numeric feature

This column, named 'JJA' (the standard meteorological abbreviation for June-July-August), most likely represents a seasonal anomaly or index value — possibly temperature or precipitation departure for the boreal summer season. Values range from -0.5 to 1.23 with a mean near zero (0.069) and median of -0.04, consistent with a centered anomaly series. The distribution is moderately right-skewed (skew = 0.968), meaning above-average summers are more extreme than below-average ones, with only 2 outliers at the 1.23 end. With 146 rows and 89 unique values, there is notable repetition that may reflect rounding to two decimal places.

Treatment: Use as-is or standardize; investigate right skew and 2 high outliers before regression modelling.

SON numeric feature

SON appears to be a continuous numeric score or index, likely a standardized anomaly or difference value (the name 'SON' suggests a meteorological or climatological index, such as a Sea-surface temperature ONset or Southern Oscillation Number). The range spans −0.52 to 1.41 with a median near zero (−0.015), consistent with a centered index, but the distribution is right-skewed (skew = 1.11) and the mean (0.097) is pulled above the median by a handful of high values. With only 86 unique values across 146 rows, there is notable value repetition, and 4 outliers (2.7%) at the upper tail warrant attention.

Treatment: Use as-is or apply mild Winsorization at the 97th percentile to dampen the 4 upper outliers before modelling.