Exemplo n.º 1
0
def convert_blackhurst_data_to_gal_per_year(df, attr):

    import flowsa
    from flowsa.mapping import add_sectors_to_flowbyactivity
    from flowsa.flowbyfunctions import clean_df, fba_fill_na_dict, harmonize_units

    # load the bea make table
    bmt = flowsa.getFlowByActivity(flowclass=['Money'],
                                   datasource='BEA_Make_Table',
                                   years=[2002])
    # clean df
    bmt = clean_df(bmt, flow_by_activity_fields, fba_fill_na_dict)
    bmt = harmonize_units(bmt)
    # drop rows with flowamount = 0
    bmt = bmt[bmt['FlowAmount'] != 0]

    bh_df_revised = pd.merge(
        df,
        bmt[['FlowAmount', 'ActivityProducedBy', 'Location']],
        left_on=['ActivityConsumedBy', 'Location'],
        right_on=['ActivityProducedBy', 'Location'])

    bh_df_revised.loc[:, 'FlowAmount'] = ((bh_df_revised['FlowAmount_x']) *
                                          (bh_df_revised['FlowAmount_y']))
    bh_df_revised.loc[:, 'Unit'] = 'gal'
    # drop columns
    bh_df_revised = bh_df_revised.drop(
        columns=["FlowAmount_x", "FlowAmount_y", 'ActivityProducedBy_y'])
    bh_df_revised = bh_df_revised.rename(
        columns={"ActivityProducedBy_x": "ActivityProducedBy"})

    return bh_df_revised
Exemplo n.º 2
0
def scale_blackhurst_results_to_usgs_values(df_to_scale, attr):
    """
    Scale the initial estimates for Blackhurst-based mining estimates to USGS values. Oil-based sectors are allocated
    a larger percentage of the difference between initial water withdrawal estimates and published USGS values.

    This method is based off the Water Satellite Table created by Yang and Ingwersen, 2017
    :param df_to_scale:
    :param attr:
    :return:
    """

    import flowsa
    from flowsa.flowbyfunctions import harmonize_units

    # determine national level published withdrawal data for usgs mining in FBS method year
    pv_load = flowsa.getFlowByActivity(flowclass=['Water'],
                                       years=[str(attr['helper_source_year'])],
                                       datasource="USGS_NWIS_WU")
    pv_load = harmonize_units(pv_load)
    pv_sub = pv_load[(pv_load['Location'] == str(US_FIPS)) & (
        pv_load['ActivityConsumedBy'] == 'Mining')].reset_index(drop=True)
    pv = pv_sub['FlowAmount'].loc[
        0] * 1000000  # usgs unit is Mgal, blackhurst unit is gal

    # sum quantity of water withdrawals already allocated to sectors
    av = df_to_scale['FlowAmount'].sum()

    # calculate the difference between published value and allocated value
    vd = pv - av

    # subset df to scale into oil and non-oil sectors
    df_to_scale['sector_label'] = np.where(
        df_to_scale['SectorConsumedBy'].apply(lambda x: x[0:5] == '21111'),
        'oil', 'nonoil')
    df_to_scale['ratio'] = np.where(df_to_scale['sector_label'] == 'oil',
                                    2 / 3, 1 / 3)
    df_to_scale['label_sum'] = df_to_scale.groupby(
        ['Location', 'sector_label'])['FlowAmount'].transform('sum')
    df_to_scale.loc[:, 'value_difference'] = vd.astype(float)

    # calculate revised water withdrawal allocation
    df_scaled = df_to_scale.copy()
    df_scaled.loc[:, 'FlowAmount'] = df_scaled['FlowAmount'] + \
                                     (df_scaled['FlowAmount'] / df_scaled['label_sum']) * \
                                     (df_scaled['ratio'] * df_scaled['value_difference'])
    df_scaled = df_scaled.drop(
        columns=['sector_label', 'ratio', 'label_sum', 'value_difference'])

    return df_scaled
Exemplo n.º 3
0
def main(method_name):
    """
    Creates a flowbysector dataset
    :param method_name: Name of method corresponding to flowbysector method yaml name
    :return: flowbysector
    """

    log.info("Initiating flowbysector creation for " + method_name)
    # call on method
    method = load_method(method_name)
    # create dictionary of data and allocation datasets
    fb = method['source_names']
    # Create empty list for storing fbs files
    fbs_list = []
    for k, v in fb.items():
        # pull fba data for allocation
        flows = load_source_dataframe(k, v)

        if v['data_format'] == 'FBA':
            # ensure correct datatypes and that all fields exist
            flows = clean_df(flows,
                             flow_by_activity_fields,
                             fba_fill_na_dict,
                             drop_description=False)

            # clean up fba, if specified in yaml
            if v["clean_fba_df_fxn"] != 'None':
                log.info("Cleaning up " + k + " FlowByActivity")
                flows = getattr(sys.modules[__name__],
                                v["clean_fba_df_fxn"])(flows)

            # if activity_sets are specified in a file, call them here
            if 'activity_set_file' in v:
                aset_names = pd.read_csv(flowbysectoractivitysetspath +
                                         v['activity_set_file'],
                                         dtype=str)

            # create dictionary of allocation datasets for different activities
            activities = v['activity_sets']
            # subset activity data and allocate to sector
            for aset, attr in activities.items():
                # subset by named activities
                if 'activity_set_file' in v:
                    names = aset_names[aset_names['activity_set'] ==
                                       aset]['name']
                else:
                    names = attr['names']

                log.info("Preparing to handle subset of flownames " +
                         ', '.join(map(str, names)) + " in " + k)
                # subset fba data by activity
                flows_subset = flows[
                    (flows[fba_activity_fields[0]].isin(names)) |
                    (flows[fba_activity_fields[1]].isin(names))].reset_index(
                        drop=True)

                # extract relevant geoscale data or aggregate existing data
                log.info("Subsetting/aggregating dataframe to " +
                         attr['allocation_from_scale'] + " geoscale")
                flows_subset_geo = subset_df_by_geoscale(
                    flows_subset, v['geoscale_to_use'],
                    attr['allocation_from_scale'])

                # Add sectors to df activity, depending on level of specified sector aggregation
                log.info("Adding sectors to " + k)
                flow_subset_wsec = add_sectors_to_flowbyactivity(
                    flows_subset_geo,
                    sectorsourcename=method['target_sector_source'],
                    allocationmethod=attr['allocation_method'])
                # clean up fba with sectors, if specified in yaml
                if v["clean_fba_w_sec_df_fxn"] != 'None':
                    log.info("Cleaning up " + k +
                             " FlowByActivity with sectors")
                    flow_subset_wsec = getattr(sys.modules[__name__],
                                               v["clean_fba_w_sec_df_fxn"])(
                                                   flow_subset_wsec, attr=attr)

                # map df to elementary flows
                log.info("Mapping flows in " + k +
                         ' to federal elementary flow list')
                if 'fedefl_mapping' in v:
                    mapping_files = v['fedefl_mapping']
                else:
                    mapping_files = k

                flow_subset_mapped = map_elementary_flows(
                    flow_subset_wsec, mapping_files)

                # clean up mapped fba with sectors, if specified in yaml
                if "clean_mapped_fba_w_sec_df_fxn" in v:
                    log.info("Cleaning up " + k +
                             " FlowByActivity with sectors")
                    flow_subset_mapped = getattr(
                        sys.modules[__name__],
                        v["clean_mapped_fba_w_sec_df_fxn"])(flow_subset_mapped,
                                                            attr, method)

                # if allocation method is "direct", then no need to create alloc ratios, else need to use allocation
                # dataframe to create sector allocation ratios
                if attr['allocation_method'] == 'direct':
                    log.info('Directly assigning ' +
                             ', '.join(map(str, names)) + ' to sectors')
                    fbs = flow_subset_mapped.copy()
                    # for each activity, if activities are not sector like, check that there is no data loss
                    if load_source_catalog(
                    )[k]['sector-like_activities'] is False:
                        activity_list = []
                        for n in names:
                            log.info('Checking for ' + n + ' at ' +
                                     method['target_sector_level'])
                            fbs_subset = fbs[(
                                (fbs[fba_activity_fields[0]] == n) &
                                (fbs[fba_activity_fields[1]] == n)) |
                                             (fbs[fba_activity_fields[0]] == n)
                                             |
                                             (fbs[fba_activity_fields[1]] == n
                                              )].reset_index(drop=True)
                            fbs_subset = check_if_losing_sector_data(
                                fbs_subset, method['target_sector_level'])
                            activity_list.append(fbs_subset)
                        fbs = pd.concat(activity_list, ignore_index=True)

                # if allocation method for an activity set requires a specific function due to the complicated nature
                # of the allocation, call on function here
                elif attr['allocation_method'] == 'allocation_function':
                    log.info(
                        'Calling on function specified in method yaml to allocate '
                        + ', '.join(map(str, names)) + ' to sectors')
                    fbs = getattr(sys.modules[__name__],
                                  attr['allocation_source'])(
                                      flow_subset_mapped, attr, fbs_list)

                else:
                    # determine appropriate allocation dataset
                    log.info("Loading allocation flowbyactivity " +
                             attr['allocation_source'] + " for year " +
                             str(attr['allocation_source_year']))
                    fba_allocation = flowsa.getFlowByActivity(
                        flowclass=[attr['allocation_source_class']],
                        datasource=attr['allocation_source'],
                        years=[attr['allocation_source_year']
                               ]).reset_index(drop=True)

                    # clean df and harmonize unites
                    fba_allocation = clean_df(fba_allocation,
                                              flow_by_activity_fields,
                                              fba_fill_na_dict)
                    fba_allocation = harmonize_units(fba_allocation)

                    # check if allocation data exists at specified geoscale to use
                    log.info("Checking if allocation data exists at the " +
                             attr['allocation_from_scale'] + " level")
                    check_if_data_exists_at_geoscale(
                        fba_allocation, attr['allocation_from_scale'])

                    # aggregate geographically to the scale of the flowbyactivty source, if necessary
                    fba_allocation = subset_df_by_geoscale(
                        fba_allocation, attr['allocation_from_scale'],
                        v['geoscale_to_use'])

                    # subset based on yaml settings
                    if attr['allocation_flow'] != 'None':
                        fba_allocation = fba_allocation.loc[
                            fba_allocation['FlowName'].isin(
                                attr['allocation_flow'])]
                    if attr['allocation_compartment'] != 'None':
                        fba_allocation = fba_allocation.loc[
                            fba_allocation['Compartment'].isin(
                                attr['allocation_compartment'])]

                    # cleanup the fba allocation df, if necessary
                    if 'clean_allocation_fba' in attr:
                        log.info("Cleaning " + attr['allocation_source'])
                        fba_allocation = getattr(sys.modules[__name__],
                                                 attr["clean_allocation_fba"])(
                                                     fba_allocation, attr=attr)
                    # reset index
                    fba_allocation = fba_allocation.reset_index(drop=True)

                    # assign sector to allocation dataset
                    log.info("Adding sectors to " + attr['allocation_source'])
                    fba_allocation_wsec = add_sectors_to_flowbyactivity(
                        fba_allocation,
                        sectorsourcename=method['target_sector_source'])

                    # call on fxn to further clean up/disaggregate the fba allocation data, if exists
                    if 'clean_allocation_fba_w_sec' in attr:
                        log.info("Further disaggregating sectors in " +
                                 attr['allocation_source'])
                        fba_allocation_wsec = getattr(
                            sys.modules[__name__],
                            attr["clean_allocation_fba_w_sec"])(
                                fba_allocation_wsec, attr=attr, method=method)

                    # subset fba datasets to only keep the sectors associated with activity subset
                    log.info("Subsetting " + attr['allocation_source'] +
                             " for sectors in " + k)
                    fba_allocation_subset = get_fba_allocation_subset(
                        fba_allocation_wsec,
                        k,
                        names,
                        flowSubsetMapped=flow_subset_mapped,
                        allocMethod=attr['allocation_method'])

                    # if there is an allocation helper dataset, modify allocation df
                    if attr['allocation_helper'] == 'yes':
                        log.info(
                            "Using the specified allocation help for subset of "
                            + attr['allocation_source'])
                        fba_allocation_subset = allocation_helper(
                            fba_allocation_subset, attr, method, v)

                    # create flow allocation ratios for each activity
                    # if load_source_catalog()[k]['sector-like_activities']
                    flow_alloc_list = []
                    group_cols = fba_mapped_default_grouping_fields
                    group_cols = [
                        e for e in group_cols
                        if e not in ('ActivityProducedBy',
                                     'ActivityConsumedBy')
                    ]
                    for n in names:
                        log.info("Creating allocation ratios for " + n)
                        fba_allocation_subset_2 = get_fba_allocation_subset(
                            fba_allocation_subset,
                            k, [n],
                            flowSubsetMapped=flow_subset_mapped,
                            allocMethod=attr['allocation_method'])
                        if len(fba_allocation_subset_2) == 0:
                            log.info("No data found to allocate " + n)
                        else:
                            flow_alloc = allocate_by_sector(
                                fba_allocation_subset_2,
                                k,
                                attr['allocation_source'],
                                attr['allocation_method'],
                                group_cols,
                                flowSubsetMapped=flow_subset_mapped)
                            flow_alloc = flow_alloc.assign(FBA_Activity=n)
                            flow_alloc_list.append(flow_alloc)
                    flow_allocation = pd.concat(flow_alloc_list,
                                                ignore_index=True)

                    # generalize activity field names to enable link to main fba source
                    log.info("Generalizing activity columns in subset of " +
                             attr['allocation_source'])
                    flow_allocation = collapse_activity_fields(flow_allocation)

                    # check for issues with allocation ratios
                    check_allocation_ratios(flow_allocation, aset, k,
                                            method_name)

                    # create list of sectors in the flow allocation df, drop any rows of data in the flow df that \
                    # aren't in list
                    sector_list = flow_allocation['Sector'].unique().tolist()

                    # subset fba allocation table to the values in the activity list, based on overlapping sectors
                    flow_subset_mapped = flow_subset_mapped.loc[
                        (flow_subset_mapped[fbs_activity_fields[0]].
                         isin(sector_list)) |
                        (flow_subset_mapped[fbs_activity_fields[1]].
                         isin(sector_list))]

                    # check if fba and allocation dfs have the same LocationSystem
                    log.info(
                        "Checking if flowbyactivity and allocation dataframes use the same location systems"
                    )
                    check_if_location_systems_match(flow_subset_mapped,
                                                    flow_allocation)

                    # merge fba df w/flow allocation dataset
                    log.info("Merge " + k + " and subset of " +
                             attr['allocation_source'])
                    fbs = flow_subset_mapped.merge(
                        flow_allocation[[
                            'Location', 'Sector', 'FlowAmountRatio',
                            'FBA_Activity'
                        ]],
                        left_on=[
                            'Location', 'SectorProducedBy',
                            'ActivityProducedBy'
                        ],
                        right_on=['Location', 'Sector', 'FBA_Activity'],
                        how='left')

                    fbs = fbs.merge(
                        flow_allocation[[
                            'Location', 'Sector', 'FlowAmountRatio',
                            'FBA_Activity'
                        ]],
                        left_on=[
                            'Location', 'SectorConsumedBy',
                            'ActivityConsumedBy'
                        ],
                        right_on=['Location', 'Sector', 'FBA_Activity'],
                        how='left')

                    # merge the flowamount columns
                    fbs.loc[:, 'FlowAmountRatio'] = fbs[
                        'FlowAmountRatio_x'].fillna(fbs['FlowAmountRatio_y'])
                    # fill null rows with 0 because no allocation info
                    fbs['FlowAmountRatio'] = fbs['FlowAmountRatio'].fillna(0)

                    # check if fba and alloc dfs have data for same geoscales - comment back in after address the 'todo'
                    # log.info("Checking if flowbyactivity and allocation dataframes have data at the same locations")
                    # check_if_data_exists_for_same_geoscales(fbs, k, attr['names'])

                    # drop rows where there is no allocation data
                    fbs = fbs.dropna(subset=['Sector_x', 'Sector_y'],
                                     how='all').reset_index()

                    # calculate flow amounts for each sector
                    log.info("Calculating new flow amounts using flow ratios")
                    fbs.loc[:, 'FlowAmount'] = fbs['FlowAmount'] * fbs[
                        'FlowAmountRatio']

                    # drop columns
                    log.info("Cleaning up new flow by sector")
                    fbs = fbs.drop(columns=[
                        'Sector_x', 'FlowAmountRatio_x', 'Sector_y',
                        'FlowAmountRatio_y', 'FlowAmountRatio',
                        'FBA_Activity_x', 'FBA_Activity_y'
                    ])

                # drop rows where flowamount = 0 (although this includes dropping suppressed data)
                fbs = fbs[fbs['FlowAmount'] != 0].reset_index(drop=True)

                # define grouping columns dependent on sectors being activity-like or not
                if load_source_catalog()[k]['sector-like_activities'] is False:
                    groupingcols = fbs_grouping_fields_w_activities
                    groupingdict = flow_by_sector_fields_w_activity
                else:
                    groupingcols = fbs_default_grouping_fields
                    groupingdict = flow_by_sector_fields

                # clean df
                fbs = clean_df(fbs, groupingdict, fbs_fill_na_dict)

                # aggregate df geographically, if necessary
                # todo: replace with fxn return_from_scale
                log.info("Aggregating flowbysector to " +
                         method['target_geoscale'] + " level")
                if fips_number_key[v['geoscale_to_use']] < fips_number_key[
                        attr['allocation_from_scale']]:
                    from_scale = v['geoscale_to_use']
                else:
                    from_scale = attr['allocation_from_scale']

                to_scale = method['target_geoscale']

                fbs_geo_agg = agg_by_geoscale(fbs, from_scale, to_scale,
                                              groupingcols)

                # aggregate data to every sector level
                log.info("Aggregating flowbysector to all sector levels")
                fbs_sec_agg = sector_aggregation(fbs_geo_agg, groupingcols)
                # add missing naics5/6 when only one naics5/6 associated with a naics4
                fbs_agg = sector_disaggregation(fbs_sec_agg, groupingdict)

                # check if any sector information is lost before reaching the target sector length, if so,
                # allocate values equally to disaggregated sectors
                log.info('Checking for data at ' +
                         method['target_sector_level'])
                fbs_agg_2 = check_if_losing_sector_data(
                    fbs_agg, method['target_sector_level'])

                # compare flowbysector with flowbyactivity
                # todo: modify fxn to work if activities are sector like in df being allocated
                if load_source_catalog()[k]['sector-like_activities'] is False:
                    check_for_differences_between_fba_load_and_fbs_output(
                        flow_subset_mapped, fbs_agg_2, aset, k, method_name)

                # return sector level specified in method yaml
                # load the crosswalk linking sector lengths
                sector_list = get_sector_list(method['target_sector_level'])

                # subset df, necessary because not all of the sectors are NAICS and can get duplicate rows
                fbs_1 = fbs_agg_2.loc[
                    (fbs_agg_2[fbs_activity_fields[0]].isin(sector_list))
                    & (fbs_agg_2[fbs_activity_fields[1]].isin(sector_list)
                       )].reset_index(drop=True)
                fbs_2 = fbs_agg_2.loc[
                    (fbs_agg_2[fbs_activity_fields[0]].isin(sector_list)) &
                    (fbs_agg_2[fbs_activity_fields[1]].isnull())].reset_index(
                        drop=True)
                fbs_3 = fbs_agg_2.loc[
                    (fbs_agg_2[fbs_activity_fields[0]].isnull())
                    & (fbs_agg_2[fbs_activity_fields[1]].isin(sector_list)
                       )].reset_index(drop=True)
                fbs_sector_subset = pd.concat([fbs_1, fbs_2, fbs_3])

                # drop activity columns
                fbs_sector_subset = fbs_sector_subset.drop(
                    ['ActivityProducedBy', 'ActivityConsumedBy'],
                    axis=1,
                    errors='ignore')

                # save comparison of FBA total to FBS total for an activity set
                compare_fba_load_and_fbs_output_totals(flows_subset_geo,
                                                       fbs_sector_subset, aset,
                                                       k, method_name, attr,
                                                       method, mapping_files)

                log.info(
                    "Completed flowbysector for activity subset with flows " +
                    ', '.join(map(str, names)))
                fbs_list.append(fbs_sector_subset)
        else:
            # if the loaded flow dt is already in FBS format, append directly to list of FBS
            log.info("Append " + k + " to FBS list")
            # ensure correct field datatypes and add any missing fields
            flows = clean_df(flows, flow_by_sector_fields, fbs_fill_na_dict)
            fbs_list.append(flows)
    # create single df of all activities
    log.info("Concat data for all activities")
    fbss = pd.concat(fbs_list, ignore_index=True, sort=False)
    log.info("Clean final dataframe")
    # aggregate df as activities might have data for the same specified sector length
    fbss = clean_df(fbss, flow_by_sector_fields, fbs_fill_na_dict)
    fbss = aggregator(fbss, fbs_default_grouping_fields)
    # sort df
    log.info("Sort and store dataframe")
    # add missing fields, ensure correct data type, reorder columns
    fbss = fbss.sort_values(
        ['SectorProducedBy', 'SectorConsumedBy', 'Flowable',
         'Context']).reset_index(drop=True)
    # save parquet file
    store_flowbysector(fbss, method_name)
Exemplo n.º 4
0
def disaggregate_cropland(fba_w_sector, attr, method, years_list, sector_column):
    """
    In the event there are 4 (or 5) digit naics for cropland at the county level, use state level harvested cropland to
    create ratios
    :param fba_w_sector:
    :param attr:
    :param years_list:
    :param sector_column: The sector column on which to make df modifications (SectorProducedBy or SectorConsumedBy)
    :param attr:
    :return:
    """

    import flowsa
    from flowsa.flowbyfunctions import sector_aggregation,\
        fbs_default_grouping_fields, clean_df, fba_fill_na_dict, fbs_fill_na_dict, add_missing_flow_by_fields,\
        sector_disaggregation, sector_ratios, replace_strings_with_NoneType, replace_NoneType_with_empty_cells,\
        harmonize_units
    from flowsa.mapping import add_sectors_to_flowbyactivity

    # tmp drop NoneTypes
    fba_w_sector = replace_NoneType_with_empty_cells(fba_w_sector)

    # drop pastureland data
    crop = fba_w_sector.loc[fba_w_sector[sector_column].apply(lambda x: x[0:3]) != '112'].reset_index(drop=True)
    # drop sectors < 4 digits
    crop = crop[crop[sector_column].apply(lambda x: len(x) > 3)].reset_index(drop=True)
    # create tmp location
    crop = crop.assign(Location_tmp=crop['Location'].apply(lambda x: x[0:2]))\

    # load the relevant state level harvested cropland by naics
    naics_load = flowsa.getFlowByActivity(flowclass=['Land'],
                                          years=years_list,
                                          datasource="USDA_CoA_Cropland_NAICS").reset_index(drop=True)
    # clean df
    naics = clean_df(naics_load, flow_by_activity_fields, fba_fill_na_dict)
    naics = harmonize_units(naics)
    # subset the harvested cropland by naics
    naics = naics[naics['FlowName'] == 'AG LAND, CROPLAND, HARVESTED'].reset_index(drop=True)
    # drop the activities that include '&'
    naics = naics[~naics['ActivityConsumedBy'].str.contains('&')].reset_index(drop=True)
    # add sectors
    naics = add_sectors_to_flowbyactivity(naics, sectorsourcename=method['target_sector_source'])
    # add missing fbs fields
    naics = clean_df(naics, flow_by_sector_fields, fbs_fill_na_dict)
    # drop cols and rename
    # naics = naics.drop(columns=["SectorProducedBy"])
    # naics = naics.rename(columns={"SectorConsumedBy": sector_column})

    # aggregate sectors to create any missing naics levels
    group_cols = fbs_default_grouping_fields
    # group_cols = [e for e in group_cols if e not in ('SectorProducedBy', 'SectorConsumedBy')]
    # group_cols.append(sector_column)
    naics2 = sector_aggregation(naics, group_cols)
    # add missing naics5/6 when only one naics5/6 associated with a naics4
    naics3 = sector_disaggregation(naics2, group_cols)
    # drop rows where FlowAmount 0
    # naics3 = naics3[~((naics3['SectorProducedBy'] == '') & (naics3['SectorConsumedBy'] == ''))]
    naics3 = naics3.loc[naics3['FlowAmount'] != 0]
    # create ratios
    naics4 = sector_ratios(naics3, sector_column)
    # create temporary sector column to match the two dfs on
    naics4 = naics4.assign(Location_tmp=naics4['Location'].apply(lambda x: x[0:2]))
    # tmp drop Nonetypes
    naics4 = replace_NoneType_with_empty_cells(naics4)

    # for loop through naics lengths to determine naics 4 and 5 digits to disaggregate
    for i in range(4, 6):
        # subset df to sectors with length = i and length = i + 1
        crop_subset = crop.loc[crop[sector_column].apply(lambda x: i+1 >= len(x) >= i)]
        crop_subset = crop_subset.assign(Sector_tmp=crop_subset[sector_column].apply(lambda x: x[0:i]))
        # if duplicates drop all rows
        df = crop_subset.drop_duplicates(subset=['Location', 'Sector_tmp'], keep=False).reset_index(drop=True)
        # drop sector temp column
        df = df.drop(columns=["Sector_tmp"])
        # subset df to keep the sectors of length i
        df_subset = df.loc[df[sector_column].apply(lambda x: len(x) == i)]
        # subset the naics df where naics length is i + 1
        naics_subset = naics4.loc[naics4[sector_column].apply(lambda x: len(x) == i+1)].reset_index(drop=True)
        naics_subset = naics_subset.assign(Sector_tmp=naics_subset[sector_column].apply(lambda x: x[0:i]))
        # merge the two df based on locations
        df_subset = pd.merge(df_subset, naics_subset[[sector_column, 'FlowAmountRatio', 'Sector_tmp', 'Location_tmp']],
                      how='left', left_on=[sector_column, 'Location_tmp'], right_on=['Sector_tmp', 'Location_tmp'])
        # create flow amounts for the new NAICS based on the flow ratio
        df_subset.loc[:, 'FlowAmount'] = df_subset['FlowAmount'] * df_subset['FlowAmountRatio']
        # drop rows of 0 and na
        df_subset = df_subset[df_subset['FlowAmount'] != 0]
        df_subset = df_subset[~df_subset['FlowAmount'].isna()].reset_index(drop=True)
        # drop columns
        df_subset = df_subset.drop(columns=[sector_column + '_x', 'FlowAmountRatio', 'Sector_tmp'])
        # rename columns
        df_subset = df_subset.rename(columns={sector_column + '_y': sector_column})
        # tmp drop Nonetypes
        df_subset = replace_NoneType_with_empty_cells(df_subset)
        # add new rows of data to crop df
        crop = pd.concat([crop, df_subset], sort=True).reset_index(drop=True)

    # clean up df
    crop = crop.drop(columns=['Location_tmp'])

    # pasture data
    pasture = fba_w_sector.loc[fba_w_sector[sector_column].apply(lambda x: x[0:3]) == '112'].reset_index(drop=True)
    # concat crop and pasture
    fba_w_sector = pd.concat([pasture, crop], sort=True).reset_index(drop=True)

    # fill empty cells with NoneType
    fba_w_sector = replace_strings_with_NoneType(fba_w_sector)

    return fba_w_sector
Exemplo n.º 5
0
def disaggregate_pastureland(fba_w_sector, attr, method, years_list, sector_column):
    """
    The USDA CoA Cropland irrigated pastureland data only links to the 3 digit NAICS '112'. This function uses state
    level CoA 'Land in Farms' to allocate the county level acreage data to 6 digit NAICS.
    :param fba_w_sector: The CoA Cropland dataframe after linked to sectors
    :param attr:
    :param years_list:
    :param sector_column: The sector column on which to make df modifications (SectorProducedBy or SectorConsumedBy)
    :return: The CoA cropland dataframe with disaggregated pastureland data
    """

    import flowsa
    from flowsa.flowbyfunctions import allocate_by_sector, clean_df, flow_by_activity_fields, \
        fba_fill_na_dict, replace_strings_with_NoneType, replace_NoneType_with_empty_cells, \
        fba_mapped_default_grouping_fields, harmonize_units
    from flowsa.mapping import add_sectors_to_flowbyactivity

    # tmp drop NoneTypes
    fba_w_sector = replace_NoneType_with_empty_cells(fba_w_sector)

    # subset the coa data so only pastureland
    p = fba_w_sector.loc[fba_w_sector[sector_column].apply(lambda x: x[0:3]) == '112'].reset_index(drop=True)
    if len(p) != 0:
        # add temp loc column for state fips
        p = p.assign(Location_tmp=p['Location'].apply(lambda x: x[0:2]))
        df_sourcename = pd.unique(p['SourceName'])[0]

        # load usda coa cropland naics
        df_class = ['Land']
        df_years = years_list
        df_allocation = 'USDA_CoA_Cropland_NAICS'
        df_f = flowsa.getFlowByActivity(flowclass=df_class, years=df_years, datasource=df_allocation)
        df_f = clean_df(df_f, flow_by_activity_fields, fba_fill_na_dict)
        df_f = harmonize_units(df_f)
        # subset to land in farms data
        df_f = df_f[df_f['FlowName'] == 'FARM OPERATIONS']
        # subset to rows related to pastureland
        df_f = df_f.loc[df_f['ActivityConsumedBy'].apply(lambda x: x[0:3]) == '112']
        # drop rows with "&'
        df_f = df_f[~df_f['ActivityConsumedBy'].str.contains('&')]
        # create sector columns
        df_f = add_sectors_to_flowbyactivity(df_f, sectorsourcename=method['target_sector_source'])
        # create proportional ratios
        group_cols = fba_mapped_default_grouping_fields
        group_cols = [e for e in group_cols if
                      e not in ('ActivityProducedBy', 'ActivityConsumedBy')]
        df_f = allocate_by_sector(df_f, df_sourcename, df_allocation, 'proportional', group_cols)
        # tmp drop NoneTypes
        df_f = replace_NoneType_with_empty_cells(df_f)
        # drop naics = '11
        df_f = df_f[df_f[sector_column] != '11']
        # drop 000 in location
        df_f = df_f.assign(Location=df_f['Location'].apply(lambda x: x[0:2]))

        # merge the coa pastureland data with land in farm data
        df = p.merge(df_f[[sector_column, 'Location', 'FlowAmountRatio']], how='left',
                     left_on="Location_tmp", right_on="Location")
        # multiply the flowamount by the flowratio
        df.loc[:, 'FlowAmount'] = df['FlowAmount'] * df['FlowAmountRatio']
        # drop columns and rename
        df = df.drop(columns=['Location_tmp', sector_column + '_x', 'Location_y', 'FlowAmountRatio'])
        df = df.rename(columns={sector_column + '_y': sector_column,
                                "Location_x": 'Location'})

        # drop rows where sector = 112 and then concat with original fba_w_sector
        fba_w_sector = fba_w_sector[fba_w_sector[sector_column].apply(lambda x: x[0:3]) != '112'].reset_index(drop=True)
        fba_w_sector = pd.concat([fba_w_sector, df], sort=True).reset_index(drop=True)

        # fill empty cells with NoneType
        fba_w_sector = replace_strings_with_NoneType(fba_w_sector)

    return fba_w_sector
Exemplo n.º 6
0
def convert_statcan_data_to_US_water_use(df, attr):
    """
    Use Canadian GDP data to convert 3 digit canadian water use to us water
    use:
    - canadian gdp
    - us gdp
    :return:
    """
    import flowsa
    from flowsa.values_from_literature import get_Canadian_to_USD_exchange_rate
    from flowsa.flowbyfunctions import assign_fips_location_system, aggregator, fba_default_grouping_fields, harmonize_units
    from flowsa.common import US_FIPS, load_bea_crosswalk

    # load Canadian GDP data
    gdp = flowsa.getFlowByActivity(flowclass=['Money'],
                                   datasource='StatCan_GDP',
                                   years=[attr['allocation_source_year']])
    gdp = harmonize_units(gdp)
    # drop 31-33
    gdp = gdp[gdp['ActivityProducedBy'] != '31-33']
    gdp = gdp.rename(columns={"FlowAmount": "CanDollar"})

    # merge df
    df_m = pd.merge(df,
                    gdp[['CanDollar', 'ActivityProducedBy']],
                    how='left',
                    left_on='ActivityConsumedBy',
                    right_on='ActivityProducedBy')
    df_m['CanDollar'] = df_m['CanDollar'].fillna(0)
    df_m = df_m.drop(columns=["ActivityProducedBy_y"])
    df_m = df_m.rename(columns={"ActivityProducedBy_x": "ActivityProducedBy"})
    df_m = df_m[df_m['CanDollar'] != 0]

    exchange_rate = get_Canadian_to_USD_exchange_rate(
        str(attr['allocation_source_year']))
    exchange_rate = float(exchange_rate)
    # convert to mgal/USD
    df_m.loc[:, 'FlowAmount'] = df_m['FlowAmount'] / (df_m['CanDollar'] /
                                                      exchange_rate)
    df_m.loc[:, 'Unit'] = 'Mgal/USD'

    df_m = df_m.drop(columns=["CanDollar"])

    # convert Location to US
    df_m.loc[:, 'Location'] = US_FIPS
    df_m = assign_fips_location_system(df_m,
                                       str(attr['allocation_source_year']))

    # load us gdp
    # load Canadian GDP data
    us_gdp_load = flowsa.getFlowByActivity(
        flowclass=['Money'],
        datasource='BEA_GDP_GrossOutput_IO',
        years=[attr['allocation_source_year']])
    us_gdp_load = harmonize_units(us_gdp_load)
    # load bea crosswalk
    cw_load = load_bea_crosswalk()
    cw = cw_load[['BEA_2012_Detail_Code', 'NAICS_2012_Code']].drop_duplicates()
    cw = cw[cw['NAICS_2012_Code'].apply(
        lambda x: len(str(x)) == 3)].drop_duplicates().reset_index(drop=True)

    # merge
    us_gdp = pd.merge(us_gdp_load,
                      cw,
                      how='left',
                      left_on='ActivityProducedBy',
                      right_on='BEA_2012_Detail_Code')
    us_gdp = us_gdp.drop(
        columns=['ActivityProducedBy', 'BEA_2012_Detail_Code'])
    # rename columns
    us_gdp = us_gdp.rename(columns={'NAICS_2012_Code': 'ActivityProducedBy'})
    # agg by naics
    us_gdp = aggregator(us_gdp, fba_default_grouping_fields)
    us_gdp = us_gdp.rename(columns={'FlowAmount': 'us_gdp'})

    # determine annual us water use
    df_m2 = pd.merge(df_m,
                     us_gdp[['ActivityProducedBy', 'us_gdp']],
                     how='left',
                     left_on='ActivityConsumedBy',
                     right_on='ActivityProducedBy')

    df_m2.loc[:, 'FlowAmount'] = df_m2['FlowAmount'] * (df_m2['us_gdp'])
    df_m2.loc[:, 'Unit'] = 'Mgal'
    df_m2 = df_m2.rename(
        columns={'ActivityProducedBy_x': 'ActivityProducedBy'})
    df_m2 = df_m2.drop(columns=['ActivityProducedBy_y', 'us_gdp'])

    return df_m2
Exemplo n.º 7
0
def geoscale_flow_comparison(flowclass,
                             years,
                             datasource,
                             activitynames=['all'],
                             to_scale='national'):
    """ Aggregates county data to state and national, and state data to national level, allowing for comparisons
        in flow totals for a given flowclass and industry. First assigns all flownames to NAICS and standardizes units.

        Assigned to NAICS rather than using FlowNames for aggregation to negate any changes in flownames across
        time/geoscale
    """

    # load parquet file checking aggregation
    flows = flowsa.getFlowByActivity(flowclass=flowclass,
                                     years=years,
                                     datasource=datasource)
    # fill null values
    flows = flows.fillna(value=fba_fill_na_dict)
    # convert units
    flows = harmonize_units(flows)

    # if activityname set to default, then compare aggregation for all activities. If looking at particular activity,
    # filter that activity out
    if activitynames == ['all']:
        flow_subset = flows.copy()
    else:
        flow_subset = flows[
            (flows[fba_activity_fields[0]].isin(activitynames)) |
            (flows[fba_activity_fields[1]].isin(activitynames))]

    # Reset index values after subset
    flow_subset = flow_subset.reset_index()

    # pull naics crosswalk
    mapping = get_activitytosector_mapping(flow_subset['SourceName'].all())

    # assign naics to activities
    # usgs datasource is not easily assigned to naics for checking totals, so instead standardize activity names
    if datasource == 'USGS_NWIS_WU':
        flow_subset = standardize_usgs_nwis_names(flow_subset)
    else:
        flow_subset = pd.merge(flow_subset,
                               mapping[['Activity', 'Sector']],
                               left_on='ActivityProducedBy',
                               right_on='Activity',
                               how='left').rename(
                                   {'Sector': 'SectorProducedBy'}, axis=1)
        flow_subset = pd.merge(flow_subset,
                               mapping[['Activity', 'Sector']],
                               left_on='ActivityConsumedBy',
                               right_on='Activity',
                               how='left').rename(
                                   {'Sector': 'SectorConsumedBy'}, axis=1)
    flow_subset = flow_subset.drop(columns=[
        'ActivityProducedBy', 'ActivityConsumedBy', 'Activity_x', 'Activity_y',
        'Description'
    ],
                                   errors='ignore')
    flow_subset['SectorProducedBy'] = flow_subset['SectorProducedBy'].replace({
        np.nan:
        None
    }).astype(str)
    flow_subset['SectorConsumedBy'] = flow_subset['SectorConsumedBy'].replace({
        np.nan:
        None
    }).astype(str)

    # create list of geoscales for aggregation
    if to_scale == 'national':
        geoscales = ['national', 'state', 'county']
    elif to_scale == 'state':
        geoscales = ['state', 'county']

    # create empty df list
    flow_dfs = []
    for i in geoscales:
        try:
            # filter by geoscale
            fba_from_scale = filter_by_geoscale(flow_subset, i)

            # remove/add column names as a column
            group_cols = fba_default_grouping_fields.copy()
            for j in ['Location', 'ActivityProducedBy', 'ActivityConsumedBy']:
                group_cols.remove(j)
            for j in ['SectorProducedBy', 'SectorConsumedBy']:
                group_cols.append(j)

            # county sums to state and national, state sums to national
            if to_scale == 'state':
                fba_from_scale['Location'] = fba_from_scale['Location'].apply(
                    lambda x: str(x[0:2]))
            elif to_scale == 'national':
                fba_from_scale['Location'] = US_FIPS

            # aggregate
            fba_agg = aggregator(fba_from_scale, group_cols)

            # rename flowamount column, based on geoscale
            fba_agg = fba_agg.rename(columns={"FlowAmount": "FlowAmount_" + i})

            # drop fields irrelevant to aggregated flow comparision
            drop_fields = flows[[
                'MeasureofSpread', 'Spread', 'DistributionType',
                'DataReliability', 'DataCollection'
            ]]
            fba_agg = fba_agg.drop(columns=drop_fields)

            # reset index
            fba_agg = fba_agg.reset_index(drop=True)

            flow_dfs.append(fba_agg)
        except:
            pass

    # merge list of dfs by column
    flow_comparison = reduce(
        lambda left, right: pd.merge(
            left,
            right,
            on=[
                'Class', 'SourceName', 'FlowName', 'Unit', 'SectorProducedBy',
                'SectorConsumedBy', 'Compartment', 'Location',
                'LocationSystem', 'Year'
            ],
            how='outer'), flow_dfs)

    # sort df
    flow_comparison = flow_comparison.sort_values([
        'Year', 'Location', 'SectorProducedBy', 'SectorConsumedBy', 'FlowName',
        'Compartment'
    ])

    return flow_comparison
Exemplo n.º 8
0
def convert_blackhurst_data_to_gal_per_employee(df_wsec, attr, method):
    """

    :param df_wsec:
    :param attr:
    :param method:
    :return:
    """

    import flowsa
    from flowsa.mapping import add_sectors_to_flowbyactivity
    from flowsa.flowbyfunctions import clean_df, fba_fill_na_dict,  proportional_allocation_by_location_and_activity, \
        filter_by_geoscale, harmonize_units
    from flowsa.BLS_QCEW import clean_bls_qcew_fba

    bls = flowsa.getFlowByActivity(flowclass=['Employment'],
                                   datasource='BLS_QCEW',
                                   years=[2002])

    bls = filter_by_geoscale(bls, 'national')

    # clean df
    bls = clean_df(bls, flow_by_activity_fields, fba_fill_na_dict)
    bls = harmonize_units(bls)
    bls = clean_bls_qcew_fba(bls, attr=attr)

    # assign naics to allocation dataset
    bls_wsec = add_sectors_to_flowbyactivity(
        bls, sectorsourcename=method['target_sector_source'])
    # drop rows where sector = None ( does not occur with mining)
    bls_wsec = bls_wsec[~bls_wsec['SectorProducedBy'].isnull()]
    bls_wsec = bls_wsec.rename(columns={
        'SectorProducedBy': 'Sector',
        'FlowAmount': 'HelperFlow'
    })

    # merge the two dfs
    df = pd.merge(df_wsec,
                  bls_wsec[['Location', 'Sector', 'HelperFlow']],
                  how='left',
                  left_on=['Location', 'SectorConsumedBy'],
                  right_on=['Location', 'Sector'])
    # drop any rows where sector is None
    df = df[~df['Sector'].isnull()]
    # fill helperflow values with 0
    df['HelperFlow'] = df['HelperFlow'].fillna(0)

    # calculate proportional ratios
    df_wratio = proportional_allocation_by_location_and_activity(df, 'Sector')

    df_wratio = df_wratio.rename(columns={
        'FlowAmountRatio': 'EmployeeRatio',
        'HelperFlow': 'Employees'
    })

    # drop rows where helperflow = 0
    df_wratio = df_wratio[df_wratio['Employees'] != 0]

    # calculate gal/employee in 2002
    df_wratio.loc[:, 'FlowAmount'] = (
        df_wratio['FlowAmount'] *
        df_wratio['EmployeeRatio']) / df_wratio['Employees']
    df_wratio.loc[:, 'Unit'] = 'gal/employee'

    # drop cols
    df_wratio = df_wratio.drop(
        columns=['Sector', 'Employees', 'EmployeeRatio'])

    return df_wratio