Maps – Peter Horton

Is Nevada Really The Most Mountainous State?

Which State Has the Most Mountains over 10k Feet?

I was recently asked which state has the most mountains over 10,000 feet. I guessed Colorado, but was surprised to hear that it is allegedly Nevada.

If you Google this, you can find some evidence for it, in particular this Opinion piece from the Great Falls Tribune, which states:

If measured by the state with the number of named mountain ranges, then the distinction goes to Nevada, according to a member of Blurtit.com, an online question and answer community.

“Nevada has more than 300 named mountain ranges, all running north-south as part of the Great Basin complex. Elevations range from 2,000 to 3,000 feet. The state has the most number of peaks above 10,000 feet. “
Great Falls Tribune edit board

While I don’t doubt that Nevada has a large number of mountain ranges, I found it hard to believe that Nevada had the most peaks above 10,000 feet, particularly because California contains the bulk of the Sierra Nevada range, and Colorado is known for its many 14,000 foot peaks and has significant areas of land above 10,000 feet in elevation.

Curious to find the answer, I looked at topographic data and mountain summit data to dig into it further. From that, here are my findings: Colorado, not Nevada, has the most mountains over 10,000 feet tall.

Running the Numbers

The first source I looked at was summit data from the Environmental Systems Research Institute. It provided data on mountain peaks, including height in meters and the state and county in which each peak is located. Filtering down the dataset to mountains over 3048 meters (10k feet), here’s how many peaks there are in each state:

State	Number of 10,ooo foot Mountains
Colorado	1,593
California	506
Wyoming	474
Utah	265
Montana	230
New Mexico	169
Idaho	119
Nevada	117
Alaska	108
Hawaii	38
Arizona	25
Washington	13
Oregon	5

Source: Environmental Systems Research Institute (Redlands, Calif.)

The number for Nevada from this data closely matches that of this list on Peakbagger.com, which lists 134 peaks above 10,000 feet.

The Colorado number appears harder to verify, but if anything appears vastly undercounted relative to the numbers listed in this article. That link mentions 48 14k footers, 804 13k footers, 1,062 12k footers, 716 11k footers, and 527 10k footers (with at least 300′ clear prominence).

Perhaps the quote above listing Nevada on top is incorporating some sort of prominence metric, but it seems hard to justify putting Nevada on top when there are 7 other states with more 10k footers.

Visualizing

This question gave me a good opportunity to play around with RASTER data, where I used elevation data from the USGS and to map that data alongside state boundary polygons and the summit point data mentioned above. As you can see from the map, Colorado clearly has the most mountains over 10k feet, which are marked in blue.

Below is a map without the mountains.

Mapping and Analyzing 16 Years of Data on Top HS Basketball Recruits

For this post, I analyzed ESPN100 men’s HS basketball prospect ratings from 2007 to 2022. For a given year, the dataset looks something like this:

For this analysis, I cleaned the dataset and used API Ninja to Geocode hometown information. I ended up analyzing:
– Basic player information like name, height, and weight
– Top high schools for producing ESPN100 players and the colleges players most frequently attend
– Maps of where ESPN100 players come from each year and over time
– Maps of where particular colleges’ ESPN100 players come from
– An analysis of ESPN100 players forgoing college basketball

Notes and some additional information on processing are including at the bottom of the post. In general, I did not edit the dataset and primarily filled in gaps where they existed. The original dataset was accessed manually through the ESPN100 website.

Basic Prospect Data (Names, Height, Weight)

Names

The most common name for ESPN100 recruits from 2007 to 2022 is either Jordan or Jalen, with 21 recruits each having those names. The other names are:

Jordan – 21 recruits
Jalen – 21 recruits
Brandon – 19 recruits
Isaiah – 16 recruits
Josh – 15 recruits
Chris – 15 recruits
Anthony – 14 recruits
Justin – 14 recruits
Tyler – 14 recruits
James – 13 recruits

Height

The tallest player in the dataset is Mamadou Ndiaye, who is listed at 7’5″. Ndiaye was the #74 ranked player in 2013 and went on to play basketball at UC Irvine.

The shortest player in the dataset is Erving Walker, who is listed at 5’6″. Walker was the #75 ranked player in 2008 and went on to play basketball at Florida.

View the full distribution below:

Weight

The heaviest player in the dataset is Sim Bhullar, who is listed at 7’4″ and 330 lbs. Bhullar was the #82 ranked player in 2011 and went on to play basketball at New Mexico State before becoming the first player of Indian descent to play in the NBA.

There are 8 different players tied for the lightest player in the dataset at 150 pounds.

View the full distribution below:

Top High Schools and Colleges

The graphic below shows the high schools or prep schools that had the most ESPN100 prospects.

High schools that produced the most ESPN100 prospects

In terms of colleges signing the highest number of ESPN100 recruits, Kentucky and Duke are a clear tier above the rest.

Colleges that recruited the most ESPN100 prospects

Recruits by Hometown Every Year and in Total

The two graphics depict where ESPN100 recruits came from in a given year. Some possible trends include a lot fewer players from the Bay Area over time and more players from the Twin Cities and Seattle.

Mapping where top basketball players came from on ESPN100 in total across 2007 to 2022

Where Colleges Get Their Recruits

The following section shows maps for the 76 schools that signed more than 5 ESPN100 recruits from 2007 to 2022. The slideshow is loaded in descending order, with schools with more numbers of recruits at the beginning. For each school, the map plots the hometown of every recruit they signed, with a catchall “Overseas” category.

In many cases, one could come pretty close to guessing the school just by the location of its top recruits.

Players Forgoing College Basketball

During this time period, players were not allowed to enter the NBA directly out of high school. Nonetheless, 26 players on the list did not end up signing with a college to play basketball. The number of such players peaked at 8 in 2020, which coincided with the inaugural season of the NBA G League Ignite.

Notable prospects to forgo college basketball (2007 – 2002):

Jaden Hardy: #2 in ’21
Jalen Green: #1 in ’20
Jonathan Kuminga: #4 in ’20
LaMelo Ball: #21 in ’19
Anfernee Simons: #9 in ’18
Mitchell Robinson: #11 in ’17
Emmanuel Mudiay: #5 in ’14
Brandon Jennings: #1 in ’08
Terrelle Pryor: #39 in ’08

Note: Terrelle Pryor appears to be the only player from 07-22 to forgo basketball entirely. He went on to play football at Ohio State

Data Processing and Cleaning

For each class year, the ESPN100 data provides the following fields
– Rank
– Player Name
– Position
– Hometown (including High School name and City / State)²
– Height
– Weight
– Stars
– Grade (0 to 100 recruit grade determined by ESPN)
– College (School where they signed or committed)

A few notes about the dataset:
– For the “Hometown” field, the City / State information is not always the City / State of the High School, and appears to be the actual City / State where the player is from. At least for this analysis, I am more interested in a player’s hometown and not the location of the school where they played their basketball season. See (appendix) for more details.
– The dataset includes international recruits.
– In the original ESPN100 dataset, not every player has complete “College” data. In most cases the name of the college where they signed a letter of intent is listed, but in some cases, the field just lists where a player committed to play or simply provides a list of school
– In certain cases, a single college or destination is not provided, in those instances I looked into the player’s career and generally marked where they played next played basketball.

Notes

¹ For whatever reason, ESPN100 does not always list exactly 100 recruits, and in some years, fewer recruits are listed.

² Many elite HS basketball recruits attend high schools or prep schools to play basketball. In general, it seems like the ESPN100 list provides the name of the high school or prep school but the city and state of the player’s hometown, and not the location of the school. For example, Brandon Jennings is listed as “Los Angeles, CA Oak Hill Academy” in the dataset, even though Oak Hill Academy is in Mouth of Wilson, VA. I deferred to ESPN and did not significantly edit these values.

Zip Codes, ZCTAs and ZCTAs that Cross State Boundaries

Zip codes are used by the United States Postal Service (USPS) to deliver mail and are not always polygon areas. The Census Bureau releases Zip Code Tabulation Areas which approximate the boundaries of zip codes. Zip Code Tabulation Areas do not cover the entire country, as represented by the grey areas in the maps below.

One problem that comes up when working with Zip Codes is the fact that some Zip Codes contain areas in multiple states. While this does not use official USPS zip code data, the lower map shows Zip Code Tabulation areas that cross state lines, of which there are 137.

The below table lists out the ZCTAs highlighted in red above.

ZCTA5	States	Number of States
86514	{UT, NM, AZ}	3
82082	{WY, NE, CO}	3
57717	{MT, SD, WY}	3
71749	{AR, LA}	2
69168	{NE, CO}	2
69201	{SD, NE}	2
69212	{SD, NE}	2
69216	{SD, NE}	2
69337	{SD, NE}	2
71953	{OK, AR}	2
71937	{OK, AR}	2
72338	{TN, AR}	2
72644	{MO, AR}	2
73949	{OK, TX}	2
75556	{AR, TX}	2
79835	{NM, TX}	2
69026	{KS, NE}	2
02861	{MA, RI}	2
79922	{NM, TX}	2
66541	{KS, NE}	2
59275	{ND, MT}	2
59847	{ID, MT}	2
63673	{MO, IL}	2
65729	{MO, AR}	2
65733	{MO, AR}	2
65761	{MO, AR}	2
66955	{KS, NE}	2
68978	{KS, NE}	2
68325	{KS, NE}	2
68327	{KS, NE}	2
68420	{KS, NE}	2
68719	{SD, NE}	2
68755	{SD, NE}	2
68943	{KS, NE}	2
79837	{NM, TX}	2
81120	{NM, CO}	2
80737	{NE, CO}	2
86515	{NM, AZ}	2
99128	{ID, WA}	2
99033	{ID, WA}	2
97913	{ID, OR}	2
97910	{ID, OR}	2
97635	{CA, OR}	2
89832	{ID, NV}	2
89439	{CA, NV}	2
89421	{OR, NV}	2
89061	{CA, NV}	2
89060	{CA, NV}	2
89019	{CA, NV}	2
89010	{CA, NV}	2
88430	{NM, TX}	2
87328	{NM, AZ}	2
86504	{NM, AZ}	2
59221	{ND, MT}	2
86044	{UT, AZ}	2
84536	{UT, AZ}	2
84531	{UT, AZ}	2
84034	{UT, NV}	2
83856	{ID, WA}	2
83342	{UT, ID}	2
83127	{ID, WY}	2
83120	{ID, WY}	2
82930	{UT, WY}	2
82801	{MT, WY}	2
82701	{SD, WY}	2
82063	{WY, CO}	2
81324	{UT, CO}	2
81137	{NM, CO}	2
59270	{ND, MT}	2
58653	{ND, SD}	2
03579	{NH, ME}	2
38549	{TN, KY}	2
54540	{MI, WI}	2
52626	{MO, IA}	2
52573	{MO, IA}	2
52542	{MO, IA}	2
51640	{MO, IA}	2
51557	{IA, NE}	2
51360	{MN, IA}	2
51023	{SD, IA}	2
51001	{SD, IA}	2
42602	{TN, KY}	2
42223	{TN, KY}	2
40965	{TN, KY}	2
38852	{AL, MS}	2
38769	{AR, MS}	2
38326	{TN, MS}	2
56136	{MN, SD}	2
38079	{TN, KY}	2
37752	{VA, TN}	2
37642	{VA, TN}	2
36855	{AL, GA}	2
30741	{TN, GA}	2
30165	{AL, GA}	2
28675	{NC, VA}	2
27048	{NC, VA}	2
24622	{VA, WV}	2
24604	{VA, WV}	2
21912	{DE, MD}	2
21874	{DE, MD}	2
20135	{VA, WV}	2
19973	{DE, MD}	2
56027	{MN, IA}	2
56144	{MN, SD}	2
58649	{ND, SD}	2
57660	{ND, SD}	2
58639	{ND, SD}	2
58623	{ND, SD}	2
58621	{ND, MT}	2
58568	{ND, SD}	2
58439	{ND, SD}	2
58436	{ND, SD}	2
58413	{ND, SD}	2
58225	{ND, MN}	2
58053	{ND, SD}	2
58043	{ND, SD}	2
58041	{ND, SD}	2
58030	{ND, MN}	2
57735	{SD, NE}	2
57724	{MT, SD}	2
57648	{ND, SD}	2
56164	{MN, SD}	2
57645	{ND, SD}	2
57642	{ND, SD}	2
57641	{ND, SD}	2
57638	{ND, SD}	2
57430	{ND, SD}	2
57255	{ND, SD}	2
57068	{MN, SD}	2
57034	{SD, IA}	2
57030	{MN, SD}	2
57026	{MN, SD}	2
56744	{ND, MN}	2
56257	{MN, SD}	2
56220	{MN, SD}	2
56219	{MN, SD}	2
99362	{WA, OR}	2

Moving Alaska and Hawaii for Mapping in National Maps

Problems Mapping Alaska and Hawaii

As anyone who has tried to make a map of the United States would tell you, the true locations of Alaska and Hawaii make creating good national maps difficult. Including these two states in their true locations leaves a lot of blank space on the map. As such, maps will oftentimes either cut out Alaska and Hawaii and only show the continental United States or resize and move them so as they appear close to the other states. The latter option is definitely preferable if you want to include Alaska and Hawaii in your map.

Scaling and Translating Alaska and Hawaii

Scaling changes the size of a geometry and translating it moves left, right, up or down. Given Alaska’s size, particularly in coordinate reference systems like 3857, Alaska and Hawaii need to be scaled and translated to fit in nicely with the lower 48 states. One quirk with scaling geographic data is that if you are attempting to plot sub-state geographies of Alaska and Hawaii, the scaling step may pull these geographies apart, as in the example below:

In the above map, the counties in Alaska and Hawaii being plotted are not scaled around a fixed point, but instead to the center of their own respective geometries. This is mostly fine for Hawaii as its counties largely consist of islands, but destroys the county adjacency relationship in Alaska. If you scale the county geometries around a fixed point however, the adjacency relationships are maintained and Alaska looks just like you’d expect it to, as in the map below:

Code to Scale and Translate Alaska and Hawaii

I’ve had to perform this operation many times and have found myself digging back into old code to find the exact numbers used in the scaling and translation. The code below includes the necessary scaling and translation to move data for Alaska and Hawaii in a Python GeoDataFrame to these locations.

The fourth parameter used when performing the scale operation is the fixed point. Please note that the code currently modifies the GeoDataFrame to be in crs 3857. For other coordinate reference systems, different scaling and translating values may be required.

The code takes in a GeoDataFrame containing data for Alaska and Hawaii, a string to refer to the column name where Alaska and Hawaii can be filtered and then inputs for the values for Alaska and Hawaii in that column.

def move_alaska_hawaii(gdf, filter_col, ak_id, hi_id):
    
    gdf = gdf.to_crs(3857)
        
    alaska = gdf[gdf[filter_col] == ak_id]
    hawaii = gdf[gdf[filter_col] == hi_id]
    
    remaining = gdf[~gdf[filter_col].isin([ak_id, hi_id])]
    
    alaska = alaska.set_geometry(alaska.scale(.2,.2,.2,(-13452629.057,3227683.786)).translate(.215e7, -1.36e6))
    hawaii = hawaii.set_geometry(hawaii.scale(1.5,1.5,1.5,(-14384434.819, 2342560.248)).translate(.57e7, 1e6))
    
    gdf = gp.GeoDataFrame(pd.concat([alaska, hawaii, remaining]), crs = 3857)
        
    return gdf

Baltimore Police District Redistricting

It was announced last week that the Baltimore Police Department is planning on redrawing its districts for the first time since 1959.

Although exact population stats are hard to find, when the current 9 districts were drawn in 1959, they all contained roughly equal population (or to be more accurate they were called “more nearly even” relative to the prior districts – see the Baltimore Sun article “Station Sites Talked Over” from March 1, 1956).

Since 1959, changes in population within the city have led to huge population disparities among the districts, as you can see below, where 2020 census data is aggregated to each district.

As the total 2020 census population for Baltimore was 585,708, an equal population split of the city would result in 9 districts with 65,078 people each, give or take a person. On the current plan, the Northeastern District covers 122,022 people and the Eastern District only 33,546.

As the current districts were drawn on slightly different geographies than those of 2020 census blocks, the below totals are calculated by intersecting each block with a shapefile of police districts. If parts of a block are within two or more different districts, the population within the block is assigned to the district containing the largest portion of its area.

Also of interest may be the racial demographics of the various police districts, which can be seen below.

Python Shapefile Adjacency Code – Queen vs. Rook

When calculating adjacency, there is sometimes a distinction made between queen adjacency and rook adjacency.

Queen vs. Rook Adjacency

For a given shape X, the Queen adjacent shapes are all the shapes that touch X. For a given shape X, the Rook adjacent shapes are all the shapes that touch X at more than just one particular point.

The two types of adjacencies are named as such as a nod to the movement of the chess pieces. Rooks can only move up or down or left to right, whereas queens can move up, down, left, right or any direction diagonally. From a given square X on a chess board, the rook adjacent squares would be those that border square X and could be traveled to by a rook, and likewise for the queen adjacency squares.

Source: “Investigating Commuting Time in a Metropolitan Statistical Area Using Spatial Autocorrelation Analysis” by S. Hessam Miri

Calculating the Two

The shapefile adjacency code I shared here, which uses a buffer, will only return the queen adjacency.

The new code, shared below, which does not user a buffer, allows a user to specify whether they want point adjacencies to be return or not. This occurs by intersecting the shapefile with itself, without a buffer and then, depending on the “include_point_adjacency” parameter, filtering out the intersection that are just of Point geometry type. These are the point intersections

import pandas as pd
import geopandas as gp
from collections import defaultdict

def calculate_adjacency(gdf, unique_col):
    '''
    Code that takes a geodataframe and returns two dataframes: the first with rook adjacencies and the second with queen adjacencies.
    
    Both dataframes have two columns the first column with the unique column values, and the second column with a list of adjacent geometries, listed by their unique column value.
    ''' 
    
    # Confirm that unique_col is actually unique
    if not(max(gdf[unique_col].value_counts(dropna = False) == 1)):
        raise ValueError("Non-unique column provided")
    
    # Intersected the GeoDataFrame with the buffer with the original GeoDataFrame
    all_intersections = gp.overlay(gdf, gdf, how = "intersection", keep_geom_type = False)
   
    # Filter out self-intersections
    filtered_intersections = all_intersections[all_intersections[unique_col+"_1"]!=all_intersections[unique_col+"_2"]]
    
    # Separate out point intersections
    point_intersections = filtered_intersections[filtered_intersections.geom_type == "Point"]
    non_point_intersections = filtered_intersections[filtered_intersections.geom_type != "Point"]
    
    # Define a tuple of zips of the unique_col pairs present in the non-point intersections
    non_point_intersections_tuples = tuple(zip(non_point_intersections[unique_col+"_1"], non_point_intersections[unique_col+"_2"]))
    
    # Define a dictionary that will map from a unique_col value to a list of other unique_cols it is adjacent to
    rook_dict = defaultdict(list)
    
    # Iterate over the tuples
    for val in non_point_intersections_tuples:        
        rook_dict[val[0]].append(val[1])

    # Some shapes will only intersect with themselves and not be added to the above
    not_added = list(set(gdf[unique_col]).difference(set(rook_dict.keys())))
    for val in not_added:
        
        # For each of these, add a blank list to the dictionary
        rook_dict[val] = []
     
    # Create DataFrame of rook intersections
    df_rook = pd.DataFrame()
    df_rook['GEOID20'] = rook_dict.keys()
    df_rook["ADJ_GEOMS"] = rook_dict.values()
        
    # Make a copy of the dictionary so we can add the point intersections
    queen_dict = {key: value[:] for key, value in rook_dict.items()}
    
    # Define a tuple of zips of the unique_col pairs present in the point intersections
    point_intersection_tuples = tuple(zip(point_intersections[unique_col+"_1"], point_intersections[unique_col+"_2"]))
    for val in point_intersection_tuples:        
        queen_dict[val[0]].append(val[1])
         
    # Create DataFrame of queen intersections
    df_queen = pd.DataFrame()
    df_queen['GEOID20'] = queen_dict.keys()
    df_queen["ADJ_GEOMS"] = queen_dict.values()
    
    return df_rook, df_queen

Comparing the Code

As you can see in comparing this image with the earlier image, AZ + CO and UT + NM are point adjacent to one another

Geopandas Shapefile Adjacency

The below code takes in a geodataframe and a unique column and returns a dictionary mapping from each unique column value to a list of the column values it is adjacent too.

As written, the code uses a buffer of 1 in the 3857 crs and, as you can see below, accounts for point (Queen’s) adjacency.

The next version of the code will attempt to do the same thing without using a buffer and return an adjacency matrix rather than a dictionary.

def calculate_adjacency(gdf, unique_col):
    '''
    Code that takes a geodataframe and returns a dictionary of adjacencies
    '''
    
    # Convert to a crs to make sure the buffer area works
    gdf = gdf.to_crs(3857)
    
    # Make a copy of the GeoDataFrame
    gdf_buffer = gdf.copy(deep = True)
    
    # Add a buffer of 1 to the geometry of the copied GeoDataFrame
    gdf_buffer["geometry"] = gdf.buffer(1)
    
    # Intersected the GeoDataFrame with the buffer with the original GeoDataFrame
    test_intersection = gp.overlay(gdf_buffer, gdf, how = "intersection")
    
    # Define a tuple of zips of the unique_col pairs present in the intersection
    test_intersection_tuples = tuple(zip(test_intersection[unique_col+"_1"], test_intersection[unique_col+"_2"]))
    
    # Define a dictionary that will map from a unique_col value to a list of other unique_cols it is adjacent to
    final_dict = {}
    
    # Iterate over the tuples
    for val in test_intersection_tuples:
        
        # The shapes will intersect with themselves, we don't want to add these to the dictionary
        if val[0] != val[1]:
            
            # If the shape is already in the dictionary
            if val[0] in list(final_dict.keys()):
                
                # Append the adjacent shape to the list
                holder = final_dict[val[0]]
                holder.append(val[1])
                final_dict[val[0]] = holder
            else:
                
                # Otherwise, create a key in the dictionary mapping to a list with the adjacenct shape
                final_dict[val[0]] = [val[1]]
                
    # Some shapes will only intersect with themselves and not be added to the above
    for val in [i for i in gdf[unique_col] if i not in list(final_dict.keys())]:
        
        # For each of these, add a blank list to the dictionary
        final_dict[val] = []
        
    # Return the adjacency dictionary    
    return final_dict

Example output from running the code on a shape file of the US States from the census.

Mapping Baltimore’s “White L” and “Black Butterfly”

Baltimore is known for its “White L” and “Black Butterfly” patterns of settlement.

Demographics

Crime

Screenshot from Baltimore Sun’ Homicide tracker in November 2022. Last 500 homicides pictured.

Interests

Sträva heat map of most popular running routes in Baltimore from November 2022.

Housing Prices

Maps below from Zillow in November, 2022.

Beyond the Black Butterfly & White L

Compared with other large Northeast Corridor cities, Baltimore is also unique in its relatively small non-black and non-white population.

Concentrated Census Tract Population Density

I was interested in looking at state’s where the densest 2020 census tracts are all primarily in one city or area of the state.

In playing around with it, MA, LA, NY, PA, IL, and AK all stood out as states with this sort of concentrated density.

Check out the maps below to see more!

While the number of tracts for each state in the images may seem arbitrary, if you added an additional tract, the maps would no longer be contained in their regions.